# Autobasket, full site text 1. [Home](https://autobasket.store/) 2. Autobasket Convert # How does agentic checkout let AI agents complete the order on your shop? Agentic checkout lets an AI assistant finish the purchase on your own shop once the customer has chosen, and come back for the refill. Autobasket Convert builds it through MCP, UCP, ACP and AP2: the agent checks stock, fills a cart and pays within the limits the customer signed, and each agent order appears in GA4 as its own channel. [Schedule a technical briefing](https://autobasket.store/contact#technical-briefing) [Read how the protocols fit](https://autobasket.store/learn/agent-commerce-protocols-explained) Search Check Cart Pay Reorder Convert ## Which assistants can complete a checkout today, and where? Agentic checkout opened in the United States first. Google's AI Mode and the Gemini app check out through UCP for eligible US retailers, and ChatGPT completes purchases inside apps through ACP. None of these launches named [Switzerland](https://autobasket.store/learn/agentic-commerce-switzerland), so a Swiss shop prepares now and switches each channel on when it opens in its market. | Channel | Protocol | Status, October 2026 | Source | | --- | --- | --- | --- | | Google AI Mode and the Gemini app | Universal Commerce Protocol (UCP) | Checkout for eligible US retailers since 11 January 2026, with the retailer as seller of record | [Google, 11 January 2026](https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/) | | ChatGPT | Agentic Commerce Protocol (ACP) | Instant Checkout moved into apps inside ChatGPT in March 2026, with ACP kept as the link to merchants | [Digital Commerce 360, 6 March 2026](https://www.digitalcommerce360.com/2026/03/06/openai-shifts-checkout-plans-agentic-commerce-strategy/) | | Shopify stores in AI Mode, Gemini and Microsoft Copilot | UCP, through Shopify | Shopify merchants can sell directly in AI Mode and Gemini, and through Copilot Checkout in Microsoft Copilot | [Shopify, 11 January 2026](https://www.shopify.com/news/ai-commerce-at-scale) | | Payments without the person present | Agent Payments Protocol (AP2) | Announced with more than 60 organisations on 16 September 2025, using signed mandates | [Google Cloud, 16 September 2025](https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol) | | Your own catalogue, for any agent | [Model Context Protocol (MCP)](https://autobasket.store/learn/agent-commerce-protocols-explained) | Open today: an MCP server lets any agent that supports MCP search, check stock and open a checkout on your shop | [modelcontextprotocol.io](https://modelcontextprotocol.io) | ## What does an agent read on your product page? Your customer sees the photo, the name and the price. Their agent reads the offer behind it: the barcode number ([GTIN](https://autobasket.store/learn/why-agents-need-a-gtin)), the price with its currency, the stock and the delivery window. Convert completes that second version and keeps it identical to the first, so the agent can order without guessing. Click the card to turn it. ## What can an agent do in your catalogue once Convert is live? Once Convert is live, an agent can search your catalogue within a budget, check stock for a delivery postcode, read an offer with its shipping and return terms, and open a checkout, all without scraping your pages. It does this through an MCP server, a set of typed tools over JSON-RPC 2.0 shaped to your catalogue during the build. | Tool | What the agent sends | What it gets back | | --- | --- | --- | | `search_catalog` | Category, attributes, maximum price, currency | Matching products with GTIN, price and availability | | `check_stock` | SKU and delivery postcode | Quantity and earliest delivery date | | `get_offer` | SKU and quantity | Price with VAT, volume discount, shipping and return terms | | `create_checkout` | Cart lines and a mandate reference | Checkout ID, total and the next step | | `get_order` | Order ID | Status, tracking and invoice link | ## How do you prove what the customer approved when an agent pays? When software pays, you need proof of what the person approved, including for a refill placed later. The Agent Payments Protocol (AP2), donated by Google to the FIDO Alliance in April 2026, records that approval as a signed mandate. Convert checks the mandate against the cart before accepting the order, so every agent order keeps its audit trail. | AP2 mandate | What it records | When it is used | What Convert checks | | --- | --- | --- | --- | | Intent mandate | The person's instruction and limits, such as items, budget and timing | Purchases completed later without the person present | Signature, expiry, and that the cart stays inside the limits | | Cart mandate | Approval of one specific basket and its final amount | Purchases where the person confirms the basket | That the signed basket matches the checkout to the cent | | Payment mandate | Authorisation shared with the payment network, not with the merchant | Every agent payment | That the payment references the accepted cart | ## How does Convert stop agents abandoning the cart? Agents often abandon a cart when replies grow too long. Each reply adds text the model must hold in mind, and noisy replies push the original instruction out of view or bury the field that matters, which is called [context rot](https://autobasket.store/glossary). Convert keeps the cart on your server and sends short, structured replies, so the agent finishes the order. In practice that means three rules for every tool. Responses carry only the fields the next step needs, never the full product page. The cart lives on the server under an ID, so the agent does not have to repeat the basket in every call. And every error names the field that failed and the value that would succeed, so the agent can correct itself instead of guessing. The limit is that no server design can stop a model from misreading a clear answer. What it can do is remove the causes that sit on the merchant's side, and we test them by running the same buying mission through several assistants before launch, as in the [office reorder under a standing mandate](https://autobasket.store/use-cases/office-supplies-reorder-mandate). An agent only reaches the cart once it can read the offer, which is the work of the [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint). ## How do you see agent orders as their own channel in your analytics? An order placed by an agent never loads a page with your analytics tag, so GA4 records it as direct or unassigned traffic, or misses it. Convert's Direct Agent Conversion Tracking (DACT) sends each agent order to GA4 from your server, with the assistant and protocol it came through, the revenue and the average order value. 1. The order arrives through MCP, UCP or ACP and is accepted. 2. Convert sends a purchase event to GA4 through the Measurement Protocol, with the order value, currency and items. 3. The event carries the channel, for example ChatGPT or Google AI Mode, and the protocol used. 4. Your reports show agent orders as their own channel, next to search, social and email. ## What do CTOs ask before starting Convert? These are the questions CTOs and e-commerce architects ask before we build an agent checkout, answered in full. They cover what happens to the existing checkout, which protocol to implement first, how orders stay within approved limits and how agent orders are counted, so the briefing can start from your stack. ### Does Convert replace our checkout? No. Your checkout keeps serving people. Convert adds the routes agents use: MCP tools that let them read your catalogue and stock and fill a cart, and the checkout protocol of the channels open in your market, connected to the order system you already run. ### Which protocol should we implement first? The one used by the channels your customers buy through. Where Google AI Mode completes checkout, that is UCP. A large ChatGPT audience makes a case for ACP, and reorders without the customer present need AP2 mandate checks. ### Can a Swiss online shop accept orders from AI agents today? Through its own MCP server, yes: any agent that supports MCP can search, check stock and open a checkout. The large assistant checkouts, UCP in Google AI Mode and ACP in ChatGPT, launched for US buyers first. A Swiss shop that prepares its offer data, CHF prices with VAT and delivery terms now can switch each channel on as it opens, instead of starting the build then. ### Is ChatGPT Instant Checkout still available? Not in its first form. OpenAI launched Instant Checkout in September 2025 for US buyers and, in March 2026, moved it into apps inside ChatGPT, keeping the Agentic Commerce Protocol as the link between users and merchants (Digital Commerce 360, 6 March 2026). For a merchant, ACP support still matters; the in-chat button is no longer the model to plan around. ### Does agentic checkout work on Shopify, WooCommerce, Salesforce and SAP? Yes, by different routes. Shopify merchants reach AI Mode, Gemini and Copilot through Shopify's own UCP integration, so Convert configures and tests it. On WooCommerce, Salesforce Commerce Cloud, SAP Commerce and custom stacks, Convert builds the MCP server and the protocol endpoints on top of your existing order system, without replatforming. ### Is it safe to let agents place orders? Orders are accepted only within the limits you set and, for payments without the person present, only after a signed AP2 mandate has been verified. Product content is also checked for planted instructions meant to mislead agents, which the Buyer Journey Check tests separately. ### How are agent orders measured? Each agent order carries its channel and protocol into GA4 as its own source, so it no longer lands as direct traffic, and the monthly report separates orders placed by agents from orders placed by people. ## Map the channels that can buy from you today In a briefing we go through your stack, the assistants your customers use and the protocols open in your markets, and you leave with the order in which Convert would build them. [Schedule a technical briefing](https://autobasket.store/contact#technical-briefing) [Start with an audit](https://autobasket.store/services/buyer-journey-check) --- 1. [Home](https://autobasket.store/) 2. Autobasket Discover # How does GEO get your products named in AI answers? Autobasket Discover is generative engine optimization for online shops. We prepare the pages and product data that ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews read, then measure every month how often each one names you against your competitors. ChatGPT ranks sellers on availability, price, quality and who makes the product, so those fields come first. [Request an AI visibility audit](https://autobasket.store/contact#geo-audit) [Compare retainers](#pricing-table) Ask Fetch Quote Name Measure Discover ## How does an assistant pick the pages it names in an answer? Before an assistant names a brand, it searches, fetches pages, cuts them into passages, scores each passage against the question and keeps only the best few to read. A page that drops out at any step is never cited, however good the product is, so Discover checks and fixes each step in order. 1. ### Search The assistant decides the question needs fresh results and runs its own queries. 2. ### Fetch Its crawler requests your page, if `robots.txt` lets it in. 3. ### Split The page is cut into passages of a few hundred words. 4. ### Rank Each passage is turned into a vector, a list of numbers, and scored for how well it matches the question. 5. ### Place The top passages enter the context window, the limited text the model reads. 6. ### Cite The answer names some sources, and their order shapes the click. The research behind this work is narrower than most sales decks suggest. [Aggarwal and colleagues (KDD 2024)](https://arxiv.org/abs/2311.09735) measured visibility gains of 30 to 40% from adding cited sources, quotations and statistics, and up to 115% for a page ranked fifth in search. [Olivier Martinez's survey of 45 studies](https://arxiv.org/abs/2607.14035) (July 2026) found that none of these techniques holds steady across platforms, and that rewrites aimed at citation can hurt retrieval when they dilute the topic. So Discover writes for being found first: one topic per page, the answer in the first 40 to 60 words of each section, and figures with their source and date added only where they keep the topic intact. Our guide on [how to get your products recommended by ChatGPT](https://autobasket.store/learn/how-ai-assistants-choose-products) explains the pipeline, and [GEO vs SEO for online shops](https://autobasket.store/learn/geo-vs-seo-for-online-shops) sets out what changes from classic search. | What was measured | Source, date and scope | | --- | --- | | ChatGPT reads structured metadata such as price and product description, and ranks merchants selling the same item on availability, price, quality and whether they are the maker or primary seller. | [OpenAI Help Center, Shopping with ChatGPT Search](https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search), fetched 16 October 2026. ChatGPT shopping results | | On Shopify, sessions referred by AI chats convert at nearly 50% higher rates than organic search, and AI-referred orders grew nearly 13 times year on year in the first quarter of 2026. | [Shopify](https://www.shopify.com/enterprise/blog/ai-search-insights), 11 May 2026. Shopify merchants, product page sessions | | 76.1% of the pages cited in Google AI Overviews also rank in the top 10 organic results for the query. | [Ahrefs, study of 1.9 million citations from 1 million AI Overviews](https://ahrefs.com/blog/search-rankings-ai-citations/), 21 July 2025, updated 2 July 2026. Google AI Overviews | | A chocolatier near Geneva was named in 58% of Google AI Overviews and 33% of Perplexity answers we tested, yet none of its 26 rendered product pages declared a price an agent could read. | [Autobasket audit](https://autobasket.store/use-cases/chocolatier-findable-not-buyable), July 2026. Switzerland, one shop, 24 prompts per engine | ## How does Discover give assistants a clean copy of your pages? Discover gives assistants a plain-text and data version of your site, built from your existing pages and kept in step with them, so what they read about your products matches what customers see. It has three parts: an `llms.txt` index, a Markdown copy of each page, and JSON-LD `@graph` data linking your company, products, prices and policies. - What an agent reads matches what a person reads, because every file is rebuilt from the rendered page at each deployment, including the full-text `llms-full.txt` and the HTTP headers that point agents to each Markdown copy - Wrong prices are caught before an agent repeats them, because the product data is validated after every catalogue change - The eight search and shopping agents we track are allowed into your pages, because we review what your crawler rules let in - Google is served by the same structured data, since it states that AI Overviews need no AI text files, so the `llms.txt` and Markdown layer is for assistants that fetch pages directly This site runs the same setup. Open [our llms.txt](https://autobasket.store/llms.txt) or [this page as Markdown](https://autobasket.store/products/autobasket-discover.md), and read what each layer does in [llms.txt, Markdown and schema for online shops](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema). ## How often do AI assistants name your brand, and how is it measured? We ask ChatGPT, Perplexity, Claude and Gemini your customers' questions many times over and record who is named, where and from which sources. The result is your share of AI answers, often called [share of model](https://autobasket.store/learn/share-of-model-explained): the share of answers that name your brand, with a margin of error, because one run of an assistant proves very little. Presence Share of answers that name you, with the range it most likely sits in (95% interval) Position Where you appear when you are named Competitors Who is named instead of you, and how often Sources The pages the assistant cited, so you know which pages to improve first Accuracy Whether the price and facts stated about you are correct ## What does each Discover retainer include? Discover runs as a monthly retainer in three tiers, set by the customer questions tracked, the assistants monitored and how often content is updated. Every tier keeps the assistant-readable copy of your site current, since the way assistants pick sources keeps changing. Prices follow a scoping call, as they depend on catalogue size and languages. | What is included | Growth | Enterprise | Global Scale | | --- | --- | --- | --- | | Customer questions tracked | Up to 250 a month | Up to 1,200 a month | Custom prompt sets, no fixed cap | | Assistants monitored | ChatGPT, Perplexity | ChatGPT, Perplexity, Claude, Gemini | All four, plus models you name | | Assistant-readable copy of your site | llms.txt | llms.txt and Markdown mirrors | Custom endpoints served from the edge | | Content and product data updates | Quarterly | Monthly | As the catalogue changes | | Report on how often you are named | Monthly export | Dashboard, updated every two weeks | API and webhook feed | ## What do teams ask before starting Discover? These are the questions marketing and SEO leads ask before a Discover retainer, answered in full. They cover what the work changes, what nobody can promise about AI answers, which assistants we measure and how long a change takes to show, so you can judge the retainer before the first call. ### What is generative engine optimization? Generative Engine Optimization, or GEO, is the work that gets your pages found, quoted and named by AI assistants such as ChatGPT, Perplexity, Claude and Google AI Overviews. It starts with letting their crawlers in and publishing structured data, and continues with writing each section so the passage an assistant picks up already carries the answer. Discover is our GEO service for shops. ### Why does ChatGPT recommend a competitor and not my shop? Usually because the competitor's offer is easier to read and confirm. OpenAI states that ChatGPT uses structured metadata such as price and description, and ranks merchants selling the same item on availability, price, quality and whether they are the maker or primary seller. A shop whose price sits only in the page text, as on 117 of the 211 Romandie shops we audited in July 2026, gives it less to confirm. ### How do I get my products into ChatGPT shopping results? According to OpenAI, Shopify merchants are included through Shopify Catalog without extra work, and other merchants can apply to send a product feed directly. Inclusion is only the start: product results are not ads and are ranked on relevance, availability and price, so Discover makes sure the feed, the structured data and the visible page say the same thing and answer the questions shoppers ask. ### Is GEO different from SEO? They share a foundation. Google's AI Overviews cite pages that are indexed, and Ahrefs found that 76.1% of cited pages also rank in the organic top 10. GEO adds the passage level, because an assistant quotes a few hundred words rather than ranking a whole page, so each section has to carry its own answer, figure and source. Our guide on GEO vs SEO for online shops sets out the differences. ### Can you guarantee a ranking in AI answers? No. Answers vary from one run to the next and no vendor controls them. We remove the reasons a page is dropped and measure the share of answers that name you, with confidence intervals, so you can see whether it moves over several months. ### Which assistants does Discover measure? ChatGPT and Perplexity on the Growth tier, and ChatGPT, Perplexity, Claude and Gemini on Enterprise, as listed in the retainer table above. Global Scale adds the models you name. ### How long before assistants name you more often? It depends on how often assistants revisit your pages and how much other sites write about your category. We measure a starting point in the visibility audit and report against it every month, which is how a real change is told apart from the normal variation between runs. ## Start with an audit of how AI answers treat your category The audit measures how often assistants name you on the questions your customers ask today and names the pages that hold you back, so the retainer starts from a baseline you can check against every month. [Request an AI visibility audit](https://autobasket.store/contact#geo-audit) [See Autobasket Convert](https://autobasket.store/products/autobasket-convert) --- # Be the next best action with Autobasket A purchase used to start with a search and end on your checkout page. Now it often starts with someone describing a problem to ChatGPT, Perplexity, Claude or Gemini, and the assistant takes it from there: it explains the options, names a few brands, compares them and sometimes places the order and the refill. We make sure that at every step it finds you, trusts you and can buy from you. [Start selling smarter](https://autobasket.store/contact#session) [See how agents shop](#stages) Ask Explain Compare Buy Reorder Your Shop ## How do people and AI assistants now find, judge and buy a product? A purchase that used to run through a search engine and your product page now passes more often through an AI assistant first. The assistant searches, reads your offer as data and shortlists two or three products, and the person decides from that shortlist. Every offer is read twice, by the person and by the software working for them. How people bought 1. Search 2. Website 3. Checkout The shopper searched, opened a few sites and bought on the one that convinced them. Your page did the persuading. How people buy now 1. Ask 2. Explain 3. Compare 4. Buy 5. Reorder The shopper describes the problem and the assistant does the rest, reading your data before any person sees your page. | The moment | The person | The agent working for them | | --- | --- | --- | | Finding your product | In a search result, a social post, a shop they already know, or a name an assistant gave them | In the pages and data it retrieves when the person asks for something like a gravel bike under CHF 4,000 | | Judging your offer | From the photos, the reviews and what they already think of the brand | From the declared price, stock, GTIN, delivery date and return terms, and it drops an offer it cannot read | | Buying | Through your checkout, after opening one or two of the pages on the shortlist | Through your checkout, opened by UCP or ACP where the channel allows it, or by handing the person a link | | Coming back | When they remember you, or when your email reaches them | When a standing instruction tells it to reorder, provided it can confirm stock and price on your side | ## Where do shops lose the sale to an AI assistant? An assistant can drop your shop at four points in the same purchase, and each one looks different from your side. Your traffic may hold steady, because people still find you, while the orders the assistant places go to a shop whose data it could read and whose checkout it could use. - ### The assistant never names you It answers the shopper with the brands whose pages it could retrieve and quote. When your guides and product pages sit behind scripts or say little beyond marketing copy, another brand becomes the answer. [Fixed by Autobasket Discover](https://autobasket.store/products/autobasket-discover) - ### It misreads your offer Price, stock, delivery and returns that are not declared as data get guessed from the page or dropped. None of the 26 product pages in our July 2026 chocolatier audit declared a price an agent could read. [Fixed by Autobasket Discover](https://autobasket.store/products/autobasket-discover) - ### It cannot check out Where the shopper lets the assistant pay, it needs a checkout it can use. Without one it hands the shopper a link, and the order waits on a click that may never come. [Fixed by Autobasket Convert](https://autobasket.store/products/autobasket-convert) - ### It reorders somewhere else A refill months later goes to the shop where the assistant can confirm stock and price. When it cannot confirm yours, the standing order moves to a shop where it can. [Fixed by Autobasket Convert](https://autobasket.store/products/autobasket-convert) ## Why does Autobasket exist? Autobasket exists so that the shops people would pick for themselves stay in the answer when an assistant does the picking. We prepare your content, your catalogue and your checkout for software that reads them before any person does, and we measure each step so you can see what changed. - ### Found Assistants name your products when a shopper describes the problem they solve. - ### Trusted The price, stock and delivery date an agent reads match the page your customers see. - ### Bought An agent can place the order on your own checkout and come back for the refill. ## How does an AI agent buy, stage by stage? Every agent purchase passes through five stages: Discover, Understand, Choose, Buy and Retain. A shop can be strong at one stage and invisible at the next, which is how a brand gets recommended and still loses the sale. Autobasket measures each stage separately and fixes the one that loses the most orders first. Choose a route to follow A person searches, lands on your product page and orders on your site. A person asks ChatGPT, Perplexity, Claude or Gemini, which names products and sends them to the shop that answers best. The agent searches, compares and pays on its own, within the limits the person set, where the market and the platform allow it. 1. ### Discover Search result Named in the AI answer Asks your catalogue directly When a shopper describes the need, are you among the answers? 2. ### Understand Product page Product data the assistant reads Reads your product and offer data Can the product, price and seller be read without guessing? 3. ### Choose Comparison with other shops Shortlist inside the answer Compares your offer with other shops Does your offer win the comparison? 4. ### Buy Your checkout Link to your checkout Pays on your checkout within the approved limit Can the decision become an order? 5. ### Retain Account or order email Follow-up question to the assistant Reorders on a standing instruction Does the next order come back to you? ## How do Autobasket Discover and Convert divide the work? Autobasket works on both versions of your offer, in two products. Autobasket Discover gets your products named in more of the answers assistants give in your category, which is where a shopper's consideration now starts. Autobasket Convert lets an agent place the order on your own checkout, through the open protocols assistants use to buy. Discover, Understand, Choose ### [Autobasket Discover](https://autobasket.store/products/autobasket-discover) For marketing and SEO leads who need their products named in AI answers, and named correctly. - Pages written so assistants can quote your answers and name your products - A plain-text guide to your shop that assistants read first (`llms.txt`, `llms-full.txt` and Markdown copies of your pages), kept current - Prices, stock and product details that assistants read correctly, checked again after every catalogue change - A monthly figure for how often four assistants name you in your category, called Share of Model [See what Discover covers](https://autobasket.store/products/autobasket-discover) Buy, Retain ### [Autobasket Convert](https://autobasket.store/products/autobasket-convert) For CTOs and e-commerce architects who need agents to complete orders on the shop’s own terms. - Agents can search your catalogue, check stock and fill a cart directly, through an MCP server - Agents can pay on your checkout in the assistants that can buy in your market, through UCP or ACP - Each agent order is checked against what the shopper approved before you accept it, using AP2 mandates - Agent orders appear as their own channel in GA4 instead of disappearing into direct traffic [See what Convert covers](https://autobasket.store/products/autobasket-convert) ## What does an agentic commerce agency change in your shop? [Agentic commerce](https://autobasket.store/learn/what-is-agentic-commerce) is buying done by an AI agent on a person's behalf, from the first search to the order and the refill. An agentic commerce agency changes what that agent reads and uses in your shop: product data, policies, feeds, content and checkout. Autobasket does this on the platform you already run, layer by layer. | Layer | What changes for the agent | Where we do it | | --- | --- | --- | | Crawler access and HTML | Search and shopping crawlers are allowed in, and product content exists in the HTML without JavaScript | [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) | | Product and offer data | Price in CHF, stock, [GTIN](https://autobasket.store/learn/why-agents-need-a-gtin), [sizes and variants](https://autobasket.store/learn/sizes-and-variants-for-agents), shipping and returns declared as data that matches the visible page | [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) | | Content assistants quote | Answers, specifications and comparisons written as passages an assistant can lift, with their sources | [Autobasket Discover](https://autobasket.store/products/autobasket-discover) | | Product feeds | Google Merchant Center and the OpenAI product feed filled from the same source as the page | [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) | | Checkout | Cart and payment through [UCP or ACP](https://autobasket.store/learn/agent-commerce-protocols-explained) where the channel is open, with AP2 mandates when the person is not present | [Autobasket Convert](https://autobasket.store/products/autobasket-convert) | | Measurement | How often each assistant names you against competitors, and agent orders counted as their own channel in GA4 | [Discover](https://autobasket.store/products/autobasket-discover) and [Convert](https://autobasket.store/products/autobasket-convert) | ## Which shops does Autobasket prepare for AI agents? Agents shop differently in each category. A bike is chosen on frame size and stock per size, a sofa on dimensions and lead time, a serum on ingredients and reviews, a B2B reorder on a signed mandate. Each solution page shows how agents buy in that category and which stage usually breaks first. ### By industry - [**Sports and outdoor**Bikes, ski, running and camping gear sold by size and fit](https://autobasket.store/solutions/sports-outdoor) - [**Fashion and apparel**Clothing, shoes and accessories bought on size, material and returns](https://autobasket.store/solutions/fashion-apparel) - [**Food and drink**Chocolate, coffee, wine and speciality food sold on origin and gifting](https://autobasket.store/solutions/food-drink) - [**Consumer electronics**Devices and accessories compared on specs, compatibility and price](https://autobasket.store/solutions/consumer-electronics) - [**Home and furniture**Furniture, lighting and homeware bought on dimensions and delivery](https://autobasket.store/solutions/home-furniture) - [**Beauty and personal care**Skincare, fragrance and cosmetics chosen on ingredients and reviews](https://autobasket.store/solutions/beauty-personal-care) ### By business model - [**Brands selling direct**D2C brands that need to be the source AI answers quote](https://autobasket.store/solutions/brands-selling-direct) - [**Multi-brand retailers**Retailers who win the comparison on price, stock and delivery](https://autobasket.store/solutions/multi-brand-retailers) - [**B2B and wholesale**Suppliers whose buyers delegate reorders to procurement agents](https://autobasket.store/solutions/b2b-wholesale) ## What did our audit of a Swiss chocolatier find? In July 2026 we audited a bean-to-bar chocolatier near Geneva that sells 205 products. AI answer engines named the brand in 58% of the Google AI Overviews and 33% of the Perplexity answers we tested, yet none of the 26 product pages we rendered declared a machine-readable price, so agents could recommend the shop but could not buy from it. We ran 24 prompts per engine, so the 95% confidence intervals run from 36 to 77% for Google and 18 to 53% for Perplexity, and rendered the product pages in a full browser on 13 July 2026. The merchant stays unnamed because the audit used public data only, and the [full audit is published as a use case](https://autobasket.store/use-cases/chocolatier-findable-not-buyable). Named in Google AI Overviews 58% Named in Perplexity answers 33% Product pages with a structured price 0 of 26 ## What does the data say about AI shopping in 2026? AI assistants already send shoppers who buy. Adobe measured traffic from AI tools to US retail sites up 693% over the 2025 holidays, and Shopify reports AI-referred orders up nearly 13 times in a year. In Switzerland, Germany and Austria only 5% of shoppers have used an agent so far, which leaves Swiss shops time to prepare. | What was measured | Source, date and scope | | --- | --- | | Traffic from generative AI tools to US retail sites rose 693.4% year on year in the 2025 holiday season. | [Adobe Analytics](https://news.adobe.com/news/2026/01/adobe-holiday-shopping-season), 7 January 2026. United States, 1 November to 31 December 2025 | | On Shopify, sessions referred by AI chats convert at nearly 50% higher rates than organic search, and AI-referred orders grew nearly 13 times year on year in the first quarter of 2026. | [Shopify](https://www.shopify.com/enterprise/blog/ai-search-insights), 11 May 2026. Shopify merchants, product page sessions | | In March 2026, visits referred by AI tools to US retail sites converted 42% better than other traffic, a year after converting 38% worse. | [Adobe Analytics](https://techcrunch.com/2026/04/16/ai-traffic-to-us-retailers-rose-393-in-q1-and-its-boosting-their-revenue-too/), reported by TechCrunch, 16 April 2026. United States retail | | 5% of online shoppers in Switzerland, Germany and Austria have already used an AI agent to help them shop, and about 18% plan to. | [University of St.Gallen (HSG), survey of just under 3,000 people](https://www.unisg.ch/en/newsdetail/news/ki-agenten-im-onlinehandel-warum-konsumentinnen-und-konsumenten-beim-einkauf-noch-zoegern/), 24 June 2026. Switzerland, Germany and Austria | | Over 80% of people in Switzerland reject a fully automated AI payment without confirmation, while nearly two thirds can imagine an AI assistant recommending how to pay. | [Swiss Payment Monitor 2/2026, University of St.Gallen and ZHAW](https://alexandria.unisg.ch/entities/publication/c1fa42b2-d39d-4420-9ee4-dfe284cbfc14), August 2026. Switzerland, consumer survey, May 2026 | | None of 240 online shops in French-speaking Switzerland scored an A for agent readiness. The median score was 57 out of 100, and 104 shops scored D or F. | [Autobasket audit of 240 Romandie online shops](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness), July 2026. Switzerland, Romandie, 240 shops | | On 117 of 211 shops (55%), the only price an agent could find was in the page text, with no Offer price in the structured data to confirm it. | [Autobasket audit of 240 Romandie online shops](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness), July 2026. Switzerland, Romandie, 211 shops read in full | ## Where would you like to start? Some teams book a conversation after one look at the problem, others want the numbers, the method or the people first. All four routes below lead to the same place, a shop that agents can read and buy from, so pick the one that matches how your team decides. - ### Talk to us this week You already see assistants answering questions in your category. We look at one of your product pages with you and say what we would fix first. [Start selling smarter](https://autobasket.store/contact#session) - ### See the evidence first Read what a real audit measured, with prompts, confidence intervals and the stage where the purchase broke. [Read the chocolatier audit](https://autobasket.store/use-cases/chocolatier-findable-not-buyable) - ### Read the full method The five-stage framework, the twelve checks and the protocols, laid out in tables with their sources. [Go to the framework](#framework) - ### See how we work with you What happens after the first call, who on your side is involved, and what you have in hand at the end. [See the four steps](#working) ## How does the five-stage agentic commerce framework work? At each of the five stages, the person who approves the budget and the agent doing the shopping need different things from your shop. The table shows both, and what Autobasket puts in place so that stage stops losing sales. Read it row by row, or go straight to the stage where you suspect the loss. | Stage | What the decision-maker needs | What the AI agent does | What Autobasket puts in place | | --- | --- | --- | --- | | Discover | To know whether the brand shows up when customers ask an assistant for a recommendation | Searches the web, retrieves passages and ranks sources across ChatGPT, Perplexity, Claude and Gemini | Pages assistants can quote, open access for AI crawlers, and AI referrals counted in your analytics (`llms.txt`, crawler rules) | | Understand | Evidence that the integration is safe, compliant and cheap to run | Extracts the product, its attributes and the merchant's identity from structured data | Product facts that agents read the way your customers do, with every figure sourced (JSON-LD schemas, Markdown mirrors) | | Choose | A comparison with the alternatives on proof, cost and effort | Compares offers on price, availability, delivery time, returns and ratings | Price, delivery and return terms an agent can compare without guessing (Offer, shipping and return markup, HTML comparison tables) | | Buy | A scoped project that can be approved in one meeting | Calls tools to build a cart, then presents a payment mandate the user signed | A checkout agents can complete, which accepts only the orders the shopper approved (MCP tools, UCP or ACP checkout, AP2 mandates) | | Retain | Monthly figures on AI visibility and on orders placed by agents | Reorders under a standing mandate and reads order status through the API | Monthly figures on how often assistants name you and what agents ordered, and reorders that go through on their own (Share of Model, GA4 attribution, reorder endpoints) | ## How can you check whether AI agents can read your shop? The Autobasket readiness check reads one product page the way an AI shopping agent does, in under a minute. It runs 12 checks in three stages: whether agents may reach the page, whether they can read the product and its price, and whether they have enough to decide. Every failure comes with the evidence found and the fix. 1. ### Reach Agents can open the page without a browser and see its content before any JavaScript runs, and your `robots.txt` lets search and shopping agents in. 2. ### Read The agent can tell exactly which product it is, by its GTIN barcode number or SKU, read the price, currency and stock, and find the same price the page shows, with a complete title and preview (Schema.org Product data). 3. ### Decide The agent can read shipping costs, delivery times, return terms and customer ratings, and finds an `llms.txt` file that describes your shop. ## Which open protocols and standards does Autobasket work with? Agents place orders through open protocols that connect to your systems directly. Autobasket implements and maintains the Model Context Protocol, the Universal Commerce Protocol, the Agentic Commerce Protocol, the Agent Payments Protocol and x402, and reads the card networks' agent programmes, so the agents your customers use can query, order and pay. | Protocol | Who stewards it | What it lets an agent do | Where it stands on 12 October 2026 | | --- | --- | --- | --- | | [Model Context Protocol (MCP)](https://modelcontextprotocol.io) | Agentic AI Foundation, Linux Foundation, since December 2025 | Call your tools, such as catalogue search, stock lookup or cart creation, over JSON-RPC 2.0 | Widely supported | | [Universal Commerce Protocol (UCP)](https://shopify.engineering/UCP) | Co-developed by Google and Shopify, January 2026 | Complete checkout with any merchant, over REST, MCP, AP2 or A2A | Live in Google AI Mode | | [Agentic Commerce Protocol (ACP)](https://openai.com/index/buy-it-in-chatgpt/) | OpenAI and Stripe, September 2025 | Send an order and a delegated payment token while you stay merchant of record | Narrowed in ChatGPT, 2026 | | [Agent Payments Protocol (AP2)](https://blog.google/products-and-platforms/platforms/google-pay/agent-payments-protocol-fido-alliance/) | Created by Google, donated to the FIDO Alliance on 28 April 2026 | Prove what the user authorised with signed mandates, including Human Not Present payments in v0.2 | v0.2 published | | Visa Trusted Agent Protocol and Mastercard Agent Pay | The card networks, with Mastercard's Verifiable Intent also going to FIDO | Let a merchant tell a verified shopping agent from a scraping bot, and log what the user approved | Market by market | | x402 | Open specification started by Coinbase | Pay per request in stablecoins such as USDC, using the HTTP 402 status code | Early adoption | [Read how MCP, UCP, ACP and AP2 fit together](https://autobasket.store/learn/agent-commerce-protocols-explained) ## What happens after you get in touch? You talk to the people who will do the work, and you see the problem on your own product pages before you commit to anything. Every engagement starts from what agents actually do on your shop today, ends with fixes your developers can ship, and is measured again afterwards with the same missions. 1. ### A first conversation We open one of your product pages together and read it the way an agent does. You leave with the stage that loses you the most. 2. ### A written proposal Scope, timing and a fixed fee for the audit or the Sprint, based on your catalogue, your platform and your markets. 3. ### The work, with your team The Buyer Journey Check maps where agents drop your shop. The Sprint makes your product pages readable to agents in seven days, on your current platform. 4. ### Measured again We rerun the same missions and the same checks, so you see what changed in figures rather than in a slide. ## What do merchants ask before working with us? These are the questions e-commerce leads, marketing leads and CTOs ask in the first conversation, answered in full. Each answer stands on its own, so you can forward one to a colleague, and the same text is published as structured data for the assistants that read this page. ### What is agentic commerce? Buying done by software on a person’s behalf. An AI assistant such as ChatGPT, Perplexity, Claude or Gemini searches, compares and, where the merchant allows it, places the order. The shop is read by the agent before any person sees it, so what the agent can read decides whether the shop is considered. ### How can AI assistants increase revenue for an online shop? They send shoppers who are close to buying. Shopify reports that sessions referred by AI chats convert at nearly 50% higher rates than organic search, and Adobe measured AI-referred visits to US retailers converting 42% better in March 2026. That revenue goes to the shops an assistant can read and compare, so the first lever is product data it can quote. ### Which AI assistants can complete a purchase today? It depends on the market. In the United States, Google’s AI Mode and the Gemini app complete checkout through the Universal Commerce Protocol for eligible retailers since 11 January 2026, and ChatGPT handles purchases inside apps through the Agentic Commerce Protocol since OpenAI moved Instant Checkout there in March 2026. In Switzerland most assistants still recommend and link, so we check which channels can buy in your market before building anything. ### Is agentic commerce relevant for a Swiss online shop in 2026? For being found, yes, and for being bought, soon. Agent checkout launched in the US first, and only 5% of shoppers in Switzerland, Germany and Austria have used an agent (University of St.Gallen, June 2026). Assistants already recommend Swiss products today, and in our July 2026 audit of 240 Romandie shops none scored an A for agent readiness, so a shop that fixes its data now competes against shops that have not. ### How is this different from SEO? SEO ranks pages for people who click. Agentic commerce prepares pages and data for software that reads, compares and buys without clicking. The two overlap on crawler access and structured data, but agents also need complete offers, machine-readable policies and, for checkout, protocols that SEO never touched. ### Do we need to change our e-commerce platform? No. The Machine Readability Sprint works on top of Shopify, WooCommerce, Salesforce Commerce Cloud, SAP Commerce and custom stacks. We add what agents need to read and order, structured data, Markdown copies of your pages, an llms.txt file and agent endpoints, to the platform you run today. ### How long does it take to become agent-ready? The Machine Readability Sprint makes your product pages readable to agents in seven days. A Buyer Journey Check tells you beforehand which stage loses the most sales, so the first week goes to the fix that matters. Checkout protocols take longer and depend on the channels open in your market. ### What does it cost? The Buyer Journey Check and the Sprint are quoted as fixed fees per shop, after a first conversation about your catalogue and markets. Discover and Convert are monthly retainers. The free readiness check on this page costs nothing and shows where one product page stands today. ## Where can you learn how AI agents choose products? Our guides explain the mechanics in plain language, with the sources named: how assistants retrieve and cite pages, what `llms.txt` does and does not do, and how the payment protocols divide the work. Start with whichever question your team is asking this week. [See all agentic commerce guides](https://autobasket.store/learn), the newest articles on the [agentic commerce blog](https://autobasket.store/blog) or the [glossary](https://autobasket.store/glossary). - [GEO vs SEO, what changes for an online shop? GEO vs SEO for online shops: both start with crawling and indexing, 76.1% of AI Overview citations rank in the top 10, and what AI answers add.](https://autobasket.store/learn/geo-vs-seo-for-online-shops) - [How does agentic commerce work for an online shop in Switzerland? Agentic commerce in Switzerland: CHF prices with VAT in Offer data, TWINT at checkout, Swiss delivery as shippingDetails, and our audit of 240 shops.](https://autobasket.store/learn/agentic-commerce-switzerland) - [UCP vs ACP vs AP2 vs MCP, which protocol does what in a purchase? UCP vs ACP, AP2 and MCP compared for merchants, with owner, launch date, status and who needs each, plus where x402 and A2A fit for a Swiss shop.](https://autobasket.store/learn/agent-commerce-protocols-explained) - [How do you get your products recommended by ChatGPT? How to get your products recommended by ChatGPT: what OpenAI says it ranks on, how product feeds reach it, and what the GEO research measured.](https://autobasket.store/learn/how-ai-assistants-choose-products) ## Make your products the ones agents can buy One conversation on one of your product pages is enough to see which stage loses you sales, so the next agent that asks about your category leaves with your product in its basket. [Start selling smarter](https://autobasket.store/contact#session) [See the Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) --- 1. [Home](https://autobasket.store/) 2. Solutions # How does agentic commerce work by industry and business model? An AI agent reads the product, compares the offer and buys in every category, and the fields it checks change with what you sell: stock per size for bikes, return terms for clothing, allergens for food, GTINs for electronics. Each page below follows one category from the shopper's question to the reorder, across six industries and three business models. Sports Fashion Food Electronics Home Beauty Your Shop ## How do agents shop in each industry? Agents filter on different fields in each industry: frame size and stock per size for bikes, size systems and return terms for clothing, allergens and delivery dates for food, identifiers for electronics, dimensions for furniture and ingredient lists for beauty. Pick your industry to see those fields and what changes for your shop once they are in place. - [ ### Sports and outdoor In our July 2026 audit, 13 of 21 sports and outdoor shops in Romandie lost the most agents at product data. "Find me a gravel bike under CHF 4,000 in size M that can be delivered to Lausanne before the end of the month." ](https://autobasket.store/solutions/sports-outdoor) - [ ### Fashion and apparel Fashion is 16% of Swiss online sales, and 34 of 45 fashion shops in our July 2026 audit lost agents at product data. "A waterproof hiking jacket for women, size S, recycled materials, that I can return for free if it does not fit." ](https://autobasket.store/solutions/fashion-apparel) - [ ### Food and drink 58 of 73 food and drink shops in our July 2026 audit lost the most agents at product data. "A box of dark chocolate pralines from a Swiss chocolatier, under CHF 60, delivered to Zürich by Friday as a gift." ](https://autobasket.store/solutions/food-drink) - [ ### Consumer electronics Home electronics is the largest Swiss online category, 28% of sales in 2025. "Noise-cancelling headphones under CHF 300 that work with my iPhone, in stock in Switzerland, with two years of warranty." ](https://autobasket.store/solutions/consumer-electronics) - [ ### Home and furniture Home and living is 15% of Swiss online sales, and agents compare it on dimensions and delivery dates. "A three-seat sofa under 220 cm wide, washable fabric, delivered to Bern within four weeks." ](https://autobasket.store/solutions/home-furniture) - [ ### Beauty and personal care NielsenIQ counts over 1 billion beauty-related searches a week on ChatGPT. "A fragrance-free vitamin C serum for sensitive skin, under CHF 50, made in Switzerland." ](https://autobasket.store/solutions/beauty-personal-care) ## Where do Swiss online sales sit by category? Home electronics is the largest Swiss online category, at 28% of sales in 2025, ahead of fashion at 16%, home and living at 15% and food at 10%. The figures come from [HANDELSVERBAND.swiss with NIQ/GfK and Swiss Post](https://handelsverband.swiss/news/schweizer-online-konsum-waechst-im-jahr-2025-erneut-um-6-das-wachstum-der-einkaeufe-im-ausland-flacht-ab/), published 12 March 2026, on a market of CHF 15.8 billion, of which CHF 13 billion went to Swiss shops. | Category | Share of Swiss online sales, 2025 | What agents check first | Solution page | | --- | --- | --- | --- | | Home electronics | 28% | Product identifier | [Consumer electronics](https://autobasket.store/solutions/consumer-electronics) | | Fashion | 16% | Size system | [Fashion and apparel](https://autobasket.store/solutions/fashion-apparel) | | Home and living | 15% | Dimensions | [Home and furniture](https://autobasket.store/solutions/home-furniture) | | Food | 10% | Price and pack size | [Food and drink](https://autobasket.store/solutions/food-drink) | ## How does the work change with your business model? The same product is sold differently by a brand, a retailer and a wholesaler, and agents judge each one on different evidence. A brand has to be the source assistants quote, a retailer wins on the terms of its offer, and a supplier keeps recurring orders only if a procurement agent can prove the purchase was approved. - ### [Brands selling direct](https://autobasket.store/solutions/brands-selling-direct) In agentic commerce, a brand selling direct competes with its own retailers for the citation. OpenAI says ChatGPT ranks sellers of the same item partly on whether they are the maker or primary seller. The brand wins the citation and the sale by publishing the most complete version: product data, its seller identity and answers to buyer questions. - ### [Multi-brand retailers](https://autobasket.store/solutions/multi-brand-retailers) A multi-brand retailer sells products other shops sell too, so an agent compares the offer rather than the product: price, stock, delivery time, returns and service. OpenAI says ChatGPT ranks merchants selling the same item on availability, price and quality. A retailer wins when each offer is matched by GTIN and its terms are declared on every page. - ### [B2B and wholesale](https://autobasket.store/solutions/b2b-wholesale) Agentic commerce in B2B starts with repeat purchases: paper, coffee, spare parts, consumables. A procurement agent reorders on a schedule within a budget, and Deloitte found 38% of surveyed B2B buyers already use agentic AI in purchasing. The supplier keeps the order only if the agent can confirm stock, read the contract price and present a signed mandate. ## Where does the purchase usually break in each category? This table puts the nine solutions side by side: the field an agent checks first, the failure we look for first when product data is read, and the engagement we usually start with. These are patterns from our audit method rather than measured rates. For worked purchases traced step by step, read our [agentic commerce examples](https://autobasket.store/use-cases). | Solution | Field an agent checks first | Common failure when the agent reads the product | Where we usually start | | --- | --- | --- | --- | | [Sports and outdoor](https://autobasket.store/solutions/sports-outdoor) | Size and fit per variant | Sizes live in a drop-down built by JavaScript that the agent never runs | [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) | | [Fashion and apparel](https://autobasket.store/solutions/fashion-apparel) | Size system | The size chart is an image | [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) | | [Food and drink](https://autobasket.store/solutions/food-drink) | Price and pack size | Prices appear only in body text, with no `Product` or `Offer` markup | [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) | | [Consumer electronics](https://autobasket.store/solutions/consumer-electronics) | Product identifier | GTIN and model numbers are missing from the markup | [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) | | [Home and furniture](https://autobasket.store/solutions/home-furniture) | Dimensions | Dimensions are drawn on a product photo | [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) | | [Beauty and personal care](https://autobasket.store/solutions/beauty-personal-care) | Ingredients | The ingredient list is only on the packshot | [Autobasket Discover](https://autobasket.store/products/autobasket-discover) | | [Brands selling direct](https://autobasket.store/solutions/brands-selling-direct) | Seller identity | The brand site states less than its retailers do | [Autobasket Discover](https://autobasket.store/products/autobasket-discover) | | [Multi-brand retailers](https://autobasket.store/solutions/multi-brand-retailers) | Identifiers | Identifiers are missing for part of the catalogue | [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) | | [B2B and wholesale](https://autobasket.store/solutions/b2b-wholesale) | Contract prices | Product data lives in PDF catalogues | [Autobasket Convert](https://autobasket.store/products/autobasket-convert) | ## What do shops ask about agentic commerce by industry? These are the questions heads of e-commerce ask before choosing where to start. Each answer stands on its own and names its source where it quotes a figure, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### Is agentic commerce worth it for a mid-sized shop in 2026? For most mid-sized shops the first step is worth it before agent checkout reaches Switzerland, because the product data that lets an agent buy is the same data assistants read when they decide whom to cite today. Only 5% of DACH online shoppers had used an AI agent by June 2026, according to the University of St.Gallen, so the work is preparation, sized to your best-selling products. ### Which industries do AI shopping agents handle most easily? Categories where products are compared on declared facts: electronics, matched by GTIN and compared on specifications, sports gear and fashion, filtered by size and stock, and repeat purchases such as coffee or office supplies. In Switzerland home electronics alone was 28% of online sales in 2025, according to HANDELSVERBAND.swiss, which makes it the largest category exposed to agent comparison. ### Does agentic commerce work the same for a brand, a retailer and a wholesaler? The agent reads the same product data in each case, but it judges each business on different evidence. A brand has to be the source assistants quote, a retailer wins on the terms of its offer against shops selling the same GTIN, and a wholesaler keeps recurring orders only if a procurement agent can confirm stock, read contract prices and prove approval. ## Not sure which page describes your shop? If your shop fits two of these pages, a brand that also sells through retailers or a retailer with a B2B arm, the first conversation sorts out which of them describes the stage to fix first. [Start selling smarter](https://autobasket.store/contact#session) [Read one product page first](https://autobasket.store/#readiness-scan) --- 1. [Home](https://autobasket.store/) 2. [Solutions](https://autobasket.store/solutions) 3. B2B and wholesale # What does agentic commerce in B2B mean for suppliers? Agentic commerce in B2B starts with repeat purchases: paper, coffee, spare parts, consumables. A procurement agent reorders on a schedule within a budget, and Deloitte found 38% of surveyed B2B buyers already use agentic AI in purchasing. The supplier keeps the order only if the agent can confirm stock, read the contract price and present a signed mandate. [Start selling smarter](https://autobasket.store/contact#session) [See where it breaks](#stages) ## Who this page is for - Wholesalers and distributors with recurring orders - Suppliers of office, industrial or hospitality consumables - Manufacturers selling spare parts to business customers Usual starting point: [Autobasket Convert](https://autobasket.store/products/autobasket-convert) ## What does a business buyer ask an assistant? A typical request reads like this one: "Reorder 20 reams of A4 paper and 6 kg of coffee beans on the first working day of each month, under CHF 400, from our current supplier." The agent runs without the buyer present, so it needs stock, the contract price and a way to prove the budget was approved. ## Which product fields decide the purchase in B2B and wholesale? In B2B and wholesale, five fields decide whether an agent can shortlist a product: contract prices, quantity pricing, stock and lead time, mandate verification, order references. The table shows why each one matters to the shopper's request here and, for your developers, how it is declared so an agent can read it. | Field | Why the agent needs it | How it is declared | | --- | --- | --- | | Contract prices | Prices differ per customer and sit behind a login | Authenticated MCP tools for logged-in accounts | | Quantity pricing | Tiered prices decide the order size | `UnitPriceSpecification` with `eligibleQuantity` | | Stock and lead time | The agent switches supplier when stock is unconfirmed | A stock tool returning quantity and delivery date | | Mandate verification | Proof the purchase was approved within limits | [AP2 intent mandates](https://autobasket.store/learn/agent-commerce-protocols-explained) checked before acceptance | | Order references | Invoices must match purchase orders | Order events with the buyer’s reference | ## What does the data say about AI agents in B2B buying? Buyers are ahead of their suppliers: Deloitte found that 38% of surveyed B2B buyers already use agentic AI in purchasing, while only 13% of suppliers have integrated their front and back office, MarketScale reported in September 2026. The payment side has a standard to build on, the Agent Payments Protocol Google announced in 2025. | What was measured | Source, date and scope | | --- | --- | | 38% of surveyed B2B buyers already use agentic AI in purchasing, while only 13% of suppliers have integrated their front and back office. | [Deloitte](https://www.marketscale.com/industries/business-services/38-of-b2b-buyers-already-use-agentic-ai-and-many-suppliers-still-lack-front-to-back-integration), reported by MarketScale, 11 September 2026. B2B buyers and suppliers | | Google announced the Agent Payments Protocol on 16 September 2025 with more than 60 organisations, using cryptographically signed mandates to prove what a person approved. | [Google Cloud](https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol), 16 September 2025. Payments standard | ## Where does an agent purchase break for B2B and wholesale? We audit B2B and wholesale stage by stage, because each stage fails for a different reason. The rows below show the failure we look for first at each stage and what changes once it is fixed, in the order a shopper's request meets them, from the first question to the reorder. 1. ### Discover Where it breaks Catalogue and prices are only visible after login What changes once it is fixed Public product data, with prices served to authenticated agents 2. ### Understand Where it breaks Product data lives in PDF catalogues What changes once it is fixed Catalogue data published as structured product records 3. ### Choose Where it breaks Quantity prices and lead times are not machine-readable What changes once it is fixed The agent can see quantity prices and lead times before it orders, as fields and tools 4. ### Buy Where it breaks Orders from agents are refused or accepted without proof of approval What changes once it is fixed You accept agent orders only with proof the buyer approved them, by checking the [AP2 mandate](https://autobasket.store/use-cases/office-supplies-reorder-mandate) first 5. ### Retain Where it breaks Recurring orders cannot run without a person What changes once it is fixed Recurring orders run on schedule, with an order status the buyer’s systems can read ## Where should B2B suppliers start? We usually start this kind of engagement with the Autobasket Convert. B2B reorders depend on stock and price lookups and on checking the buyer’s approval, which is what Convert builds. From there, the order of the remaining work follows the stage table above, so each step fixes the next place where agents drop your products. [Autobasket Convert](https://autobasket.store/products/autobasket-convert) B2B reorders depend on stock and price lookups and on checking the buyer’s approval, which is what Convert builds. [Start selling smarter](https://autobasket.store/contact#session) [Follow an office reorder under a standing mandate](https://autobasket.store/use-cases/office-supplies-reorder-mandate) ## What should B2B suppliers read next? These guides and worked examples take the questions on this page further for B2B and wholesale. Each link names the topic it covers and says in one line why it matters for a shop like yours, so you can open the one that answers your next question rather than reading all of them in order. - [How an AI purchasing agent reorders supplies](https://autobasket.store/use-cases/office-supplies-reorder-mandate) Follows one monthly office reorder under a signed mandate, from the instruction to the invoice. - [UCP vs ACP vs AP2, explained for merchants](https://autobasket.store/learn/agent-commerce-protocols-explained) What AP2 mandates, MCP tools and checkout protocols each do in a B2B order. - [Agentic commerce glossary](https://autobasket.store/glossary) Plain definitions of intent mandates, MCP and the other terms procurement teams will ask about. - [What is agentic commerce](https://autobasket.store/learn/what-is-agentic-commerce) The definition and examples to share with colleagues before a procurement agent project starts. ## What do B2B suppliers ask about agents? These are the questions teams in B2B and wholesale ask us most often about AI shopping agents, answered in full. Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### How will AI procurement agents change B2B ordering for a wholesaler? Recurring orders move first, because they follow a rule a buyer can write once: the item, the quantity, the schedule and a budget. The agent then checks stock and the contract price and places the order with no person present. A supplier whose stock, prices and order status can only be read by a person on a portal loses those orders to one whose systems answer the agent. ### Should my B2B portal show prices and stock without a login? Show product data and stock publicly where your business allows it, and keep contract prices behind authentication. A buyer’s agent can then find and qualify you without an account, then sign in as the customer’s integration to read that customer’s prices and order. A catalogue hidden entirely behind a login stays out of the agent’s first search for a supplier. ### What is a standing mandate in B2B purchasing? A signed instruction that lets an agent buy specified items on a schedule within a limit, for example a monthly reorder under CHF 400. The supplier checks the signature, the limits and that the mandate has not been revoked before accepting the order. ### Can agents see contract prices behind a login? Yes, when the supplier exposes them through authenticated tools for each business account. The agent signs in as the customer’s integration and receives that customer’s prices, while public pages keep the list price or none at all. ### Why would a supplier lose a recurring order to an agent? Because the agent follows its instruction literally. If it cannot confirm stock or prove approval at the moment it runs, it switches to a supplier that can, and the order often does not come back. ## Other business models - [Brands selling direct](https://autobasket.store/solutions/brands-selling-direct) - [Multi-brand retailers](https://autobasket.store/solutions/multi-brand-retailers) - [All solutions](https://autobasket.store/solutions) ## Get your B2B catalogue ready for agents Bring one product page from your catalogue to the first conversation, and we will show you which of the fields above an agent can read today and which one it is missing. [Start selling smarter](https://autobasket.store/contact#session) [See the Autobasket Convert](https://autobasket.store/products/autobasket-convert) --- 1. [Home](https://autobasket.store/) 2. [Solutions](https://autobasket.store/solutions) 3. Beauty and personal care # How does agentic AI in beauty pick a serum or a shade? Agentic AI in beauty chooses on ingredients, skin type and what other buyers report, and the volume is large: NielsenIQ counts over 1 billion beauty-related searches a week on ChatGPT, Glossy reported in May 2026. When ingredients sit only on a packshot and reviews load in a widget, the assistant answers from retailers and forums instead. [Start selling smarter](https://autobasket.store/contact#session) [See where it breaks](#stages) ## Who this page is for - Skincare and cosmetics brands selling direct - Fragrance houses and perfumeries - Beauty retailers carrying many brands Usual starting point: [Autobasket Discover](https://autobasket.store/products/autobasket-discover) ## What does a beauty shopper ask an assistant? A typical request reads like this one: "A fragrance-free vitamin C serum for sensitive skin, under CHF 50, made in Switzerland." The agent needs the full ingredient list as text, the skin types the product is made for, the price and the origin. ## Which product fields decide the purchase in beauty and personal care? In beauty and personal care, five fields decide whether an agent can shortlist a product: ingredients, skin and hair type, size, reviews, origin. The table shows why each one matters to the shopper's request here and, for your developers, how it is declared so an agent can read it. | Field | Why the agent needs it | How it is declared | | --- | --- | --- | | Ingredients | Read in full when a prompt excludes an ingredient | Visible INCI list and `additionalProperty` | | Skin and hair type | The most common filter in beauty prompts | `audience` or `additionalProperty` | | Size | Price per millilitre decides between similar products | `size` or `weight` with units | | Reviews | Reported results carry weight in the answer | [`AggregateRating`](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema) and `Review` in the HTML | | Origin | Swiss-made is a stated preference in many prompts | `countryOfOrigin` | ## What does the data say about AI shopping for beauty? Beauty is one of the categories people already put to assistants: NielsenIQ counts over 1 billion beauty-related searches a week on ChatGPT. Our July 2026 audit shows the gap on the shop side, with 16 of 20 cosmetics, skincare and perfume shops in Romandie losing the most agents at product data. | What was measured | Source, date and scope | | --- | --- | | Consumers run over 1 billion beauty-related searches a week on ChatGPT. | [NielsenIQ](https://www.glossy.co/beauty/beauty-briefing-what-to-know-about-sephoras-and-ultas-ai-partnerships/), reported by Glossy, 12 May 2026. ChatGPT, global | | Among 20 cosmetics, skincare and perfume shops, 16 lost the most agents at product data, and 11 showed their price only as page text. | [Autobasket audit of 240 Romandie online shops](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness), July 2026. Switzerland, Romandie, 20 beauty shops | | 5% of online shoppers in Switzerland, Germany and Austria have already used an AI agent to help them shop, and about 18% plan to. | [University of St.Gallen (HSG), survey of just under 3,000 people](https://www.unisg.ch/en/newsdetail/news/ki-agenten-im-onlinehandel-warum-konsumentinnen-und-konsumenten-beim-einkauf-noch-zoegern/), 24 June 2026. Switzerland, Germany and Austria | ## Where does an agent purchase break for beauty and personal care? We audit beauty and personal care stage by stage, because each stage fails for a different reason. The rows below show the failure we look for first at each stage and what changes once it is fixed, in the order a shopper's request meets them, from the first question to the reorder. 1. ### Discover Where it breaks Assistants quote retailers and forums because the brand page says little What changes once it is fixed Routine and ingredient answers on the brand’s own pages 2. ### Understand Where it breaks The ingredient list is only on the packshot What changes once it is fixed The agent can check the full ingredient list against what the shopper wants to avoid, as text on the page 3. ### Choose Where it breaks Reviews load in a widget that agents do not run What changes once it is fixed Your reviews count in the comparison, rendered in the HTML and declared in markup 4. ### Buy Where it breaks Bundles and sets have no single offer to buy What changes once it is fixed The agent can buy a set as one product with its own price, and each shade as a [declared variant](https://autobasket.store/learn/sizes-and-variants-for-agents) 5. ### Retain Where it breaks Refills and repeat purchases need a person to log in What changes once it is fixed Refills the assistant can reorder on the customer’s schedule, with no login in the way ## Where should beauty brands start? We usually start this kind of engagement with the Autobasket Discover. Beauty buyers ask assistants about ingredients and routines before they visit a shop, so becoming the source those answers cite comes first. From there, the order of the remaining work follows the stage table above, so each step fixes the next place where agents drop your products. [Autobasket Discover](https://autobasket.store/products/autobasket-discover) Beauty buyers ask assistants about ingredients and routines before they visit a shop, so becoming the source those answers cite comes first. [Start selling smarter](https://autobasket.store/contact#session) ## What should beauty brands read next? These guides and worked examples take the questions on this page further for beauty and personal care. Each link names the topic it covers and says in one line why it matters for a shop like yours, so you can open the one that answers your next question rather than reading all of them in order. - [How to get your products recommended by ChatGPT](https://autobasket.store/learn/how-ai-assistants-choose-products) How assistants judge which page to cite, and why the brand page has to state more than its retailers. - [Measuring your share of AI answers](https://autobasket.store/learn/share-of-model-explained) How to measure how often assistants name your brand for the routines and ingredients you sell. - [Are Swiss online shops ready for AI agents? 240 shops audited](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness) The audit behind the beauty figures on this page, with the 20 cosmetics and perfume shops grouped by stage. - [GEO vs SEO for online shops](https://autobasket.store/learn/geo-vs-seo-for-online-shops) What a beauty product page needs to be cited by assistants as well as ranked by Google. ## What do beauty brands ask about agents? These are the questions teams in beauty and personal care ask us most often about AI shopping agents, answered in full. Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### How do beauty brands show ingredients and shades to AI assistants? Publish the full INCI list as text on the product page, not only on the packshot, and declare each shade as a variant with its own name, colour, image and stock. Skin type and finish go in visible text and in `additionalProperty` fields. An assistant asked for a fragrance-free serum can then read the list itself rather than rely on a retailer’s summary. ### How do efficacy claims affect what an AI assistant says about a product? An assistant repeats what it reads, so a claim on your page can reach the shopper in the assistant’s words without the qualifiers around it. Keep the claims on the page to what your product file supports, and place them in text next to the evidence for them. What an agent repeats about your product remains your responsibility. ### Why do AI assistants quote retailers instead of beauty brands? Because the retailer page often states more: the full ingredient list, the reviews, the price per size. When the brand’s own page carries that information as text and markup, assistants have a reason to cite the brand directly. ### Do review widgets count for AI agents? Only if the reviews end up in the HTML. Many widgets load reviews after the page, and agents that do not run the script never see them. Rendering a summary and a few reviews server-side, with `AggregateRating` markup, fixes it. ### Can skincare subscriptions work with AI agents? Reorders under a standing instruction are a natural purchase to hand to an agent. They require a reorder endpoint and, for payment without the person present, verification of a signed mandate, the customer’s recorded approval of what the agent may buy, as defined in the Agent Payments Protocol (AP2). ## Other industries - [Sports and outdoor](https://autobasket.store/solutions/sports-outdoor) - [Fashion and apparel](https://autobasket.store/solutions/fashion-apparel) - [Food and drink](https://autobasket.store/solutions/food-drink) - [Consumer electronics](https://autobasket.store/solutions/consumer-electronics) - [Home and furniture](https://autobasket.store/solutions/home-furniture) - [All solutions](https://autobasket.store/solutions) ## Get your beauty catalogue ready for agents Bring one product page from your catalogue to the first conversation, and we will show you which of the fields above an agent can read today and which one it is missing. [Start selling smarter](https://autobasket.store/contact#session) [See the Autobasket Discover](https://autobasket.store/products/autobasket-discover) --- 1. [Home](https://autobasket.store/) 2. [Solutions](https://autobasket.store/solutions) 3. Brands selling direct # What does agentic commerce mean for brands selling direct? In agentic commerce, a brand selling direct competes with its own retailers for the citation. OpenAI says ChatGPT ranks sellers of the same item partly on whether they are the maker or primary seller. The brand wins the citation and the sale by publishing the most complete version: product data, its seller identity and answers to buyer questions. [Start selling smarter](https://autobasket.store/contact#session) [See where it breaks](#stages) ## Who this page is for - Brands with their own shop and a presence on retailers or marketplaces - Founders and marketing leads who need to be named in AI answers - Brands with a few hundred products or fewer Usual starting point: [Autobasket Discover](https://autobasket.store/products/autobasket-discover) ## What does a brand's customer ask an assistant? A typical request reads like this one: "What is the best Swiss-made merino base layer for ski touring, and where can I buy it directly?" For the brand to be named and linked, its own page has to be the clearest source on the product, the material and where to buy it. ## Which product fields decide the purchase in brands selling direct? In brands selling direct, five fields decide whether an agent can shortlist a product: seller identity, complete product data, answers to buyer questions, reviews, where to buy. The table shows why each one matters to the shopper's request here and, for your developers, how it is declared so an agent can read it. | Field | Why the agent needs it | How it is declared | | --- | --- | --- | | Seller identity | Tells the agent the brand itself is selling | `Organization` with `sameAs` links, linked as `seller` | | Complete product data | The most complete page tends to be the one cited | `Product` and `Offer` with every field filled, including the [GTIN](https://autobasket.store/learn/why-agents-need-a-gtin) | | Answers to buyer questions | Assistants retrieve passages that answer the prompt | Question headings with short answers beneath | | Reviews | First-party reviews compete with retailer reviews | `AggregateRating` rendered in the HTML | | Where to buy | Assistants are asked where to buy, not only what | Clear purchase path and store availability | ## What does the data say about AI shopping for direct brands? AI-referred shoppers are worth winning: Shopify reported in May 2026 that sessions from AI chats convert at nearly 50% higher rates than organic search, and that AI-referred orders grew nearly 13 times year on year in the first quarter. These are Shopify and US figures, so read them as a direction rather than a Swiss forecast. | What was measured | Source, date and scope | | --- | --- | | On Shopify, sessions referred by AI chats convert at nearly 50% higher rates than organic search, and AI-referred orders grew nearly 13 times year on year in the first quarter of 2026. | [Shopify](https://www.shopify.com/enterprise/blog/ai-search-insights), 11 May 2026. Shopify merchants, product page sessions | | 30 to 45% of US consumers use generative AI to research and compare products, and shoppers trust a retailer’s own on-site agent three times more than a third-party agent. | [Bain & Company, Consumer Lab](https://www.bain.com/insights/agentic-ai-commerce-the-next-retail-revolution-is-here/), 2 March 2026. United States consumers | | ChatGPT reads structured metadata such as price and product description, and ranks merchants selling the same item on availability, price, quality and whether they are the maker or primary seller. | [OpenAI Help Center, Shopping with ChatGPT Search](https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search), fetched 16 October 2026. ChatGPT shopping results | ## Where does an agent purchase break for brands selling direct? We audit brands selling direct stage by stage, because each stage fails for a different reason. The rows below show the failure we look for first at each stage and what changes once it is fixed, in the order a shopper's request meets them, from the first question to the reorder. 1. ### Discover Where it breaks Retailers and review sites are cited for the brand’s own products What changes once it is fixed You see each month how often assistants name you and which pages they cite instead, measured as your [share of AI answers](https://autobasket.store/learn/share-of-model-explained) 2. ### Understand Where it breaks The brand site states less than its retailers do What changes once it is fixed Your own site becomes the most complete source on your products 3. ### Choose Where it breaks No first-party reviews an agent can read What changes once it is fixed Your own reviews count in the comparison, rendered in the page and declared in markup 4. ### Buy Where it breaks Checkout is built only for people with a browser What changes once it is fixed Assistants can complete the purchase on your own shop, through [agentic checkout](https://autobasket.store/products/autobasket-convert) on the channels open in your market 5. ### Retain Where it breaks Repeat purchases go back through a retailer What changes once it is fixed Repeat orders come back to your own shop ## Where should direct brands start? We usually start this kind of engagement with the Autobasket Discover. Being the cited source is the first thing a direct brand stands to win or lose. From there, the order of the remaining work follows the stage table above, so each step fixes the next place where agents drop your products. [Autobasket Discover](https://autobasket.store/products/autobasket-discover) Being the cited source is the first thing a direct brand stands to win or lose. [Start selling smarter](https://autobasket.store/contact#session) ## What should direct brands read next? These guides and worked examples take the questions on this page further for brands selling direct. Each link names the topic it covers and says in one line why it matters for a shop like yours, so you can open the one that answers your next question rather than reading all of them in order. - [How to get your products recommended by ChatGPT](https://autobasket.store/learn/how-ai-assistants-choose-products) How assistants decide which page to cite when the brand and its retailers describe the same product. - [Measuring your share of AI answers](https://autobasket.store/learn/share-of-model-explained) The method behind measuring how often assistants name your brand, and how to read the variation. - [Do you need a GTIN for AI shopping agents](https://autobasket.store/learn/why-agents-need-a-gtin) Why the same GTIN on your shop and your retailers’ decides who an agent treats as the maker. - [UCP vs ACP vs AP2, explained for merchants](https://autobasket.store/learn/agent-commerce-protocols-explained) Which checkout protocols a direct brand will need once agent checkout opens in its market. ## What do direct brands ask about agents? These are the questions teams in brands selling direct ask us most often about AI shopping agents, answered in full. Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### How can a direct-to-consumer brand sell through ChatGPT without Amazon? By making its own shop the clearest source on its products, with complete `Product` and `Offer` data, the brand declared as `seller`, and a product feed where the channel accepts one. OpenAI says ChatGPT ranks merchants selling the same item partly on whether they are the maker. Checkout inside ChatGPT now runs through apps connected by the Agentic Commerce Protocol, opened in the US first. ### My products are sold by retailers too, who gets the AI recommendation? Whichever page states the facts most completely and can confirm price and stock. When a retailer publishes the GTIN, the full specification and reviews while the brand page shows a mood image, the retailer gets the link even for the brand’s own product. The same GTIN, complete data and the brand declared as maker on your own page give the assistant a reason to cite you. ### Why does an AI assistant send shoppers to a retailer instead of the brand? It cites the page that answers the question most completely and links to where it can confirm price and stock. If a retailer publishes more structured data than the brand, the retailer wins the link even for the brand’s own product. ### How do I measure how often AI answers name my brand? By asking a fixed set of buyer prompts repeatedly across assistants such as ChatGPT, Perplexity, Claude and Gemini, and counting the share of answers that name you, often called share of model. We report it with confidence intervals, because answers vary from one run to the next. ### Does a small brand need agent checkout protocols? Not first. Most small brands gain more from complete product data and being cited correctly. Checkout protocols come once the channels open in your market can complete a purchase, and Shopify merchants receive part of that layer from the platform. ## Other business models - [Multi-brand retailers](https://autobasket.store/solutions/multi-brand-retailers) - [B2B and wholesale](https://autobasket.store/solutions/b2b-wholesale) - [All solutions](https://autobasket.store/solutions) ## Get your brand catalogue ready for agents Bring one product page from your catalogue to the first conversation, and we will show you which of the fields above an agent can read today and which one it is missing. [Start selling smarter](https://autobasket.store/contact#session) [See the Autobasket Discover](https://autobasket.store/products/autobasket-discover) --- 1. [Home](https://autobasket.store/) 2. [Solutions](https://autobasket.store/solutions) 3. Consumer electronics # How do AI shopping agents compare consumer electronics? Consumer electronics are the easiest products for AI shopping agents to compare: the same model and GTIN, sold by many shops. Home electronics made up 28% of Swiss online sales in 2025, according to HANDELSVERBAND.swiss. An agent matches the model by identifier, then picks on price, stock, delivery and warranty, so a shop without a GTIN is left out. [Start selling smarter](https://autobasket.store/contact#session) [See where it breaks](#stages) ## Who this page is for - Electronics retailers selling the same models as their competitors - Brands selling devices and accessories direct - Shops selling spare parts and accessories by compatibility Usual starting point: [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) ## What does an electronics shopper ask an assistant? A typical request reads like this one: "Noise-cancelling headphones under CHF 300 that work with my iPhone, in stock in Switzerland, with two years of warranty." The agent needs an identifier to match the model across shops, then price, stock, warranty and compatibility as fields it can compare. ## Which product fields decide the purchase in consumer electronics? In consumer electronics, five fields decide whether an agent can shortlist a product: product identifier, specifications, compatibility, warranty, energy label. The table shows why each one matters to the shopper's request here and, for your developers, how it is declared so an agent can read it. | Field | Why the agent needs it | How it is declared | | --- | --- | --- | | Product identifier | Agents match the same model across shops by GTIN | [`gtin13`](https://autobasket.store/learn/why-agents-need-a-gtin), `mpn`, `model`, `brand` | | Specifications | Battery life, connectivity or resolution decide the shortlist | `additionalProperty` with `unitCode` | | Compatibility | Accessories are bought for a named device | `isAccessoryOrSparePartFor` | | Warranty | A tiebreaker between two shops at the same price | `warranty` as a `WarrantyPromise`, and in text | | Energy label | Required in the EU for many appliance categories | `hasEnergyConsumptionDetails` | ## What does the data say about AI shopping for electronics? Electronics buyers already ask assistants: Criteo’s Consumer Electronics Pulse 2026 found that 48% of home appliance and electronics buyers worldwide used an AI chatbot or assistant during part of their purchase. In Germany, 24% of consumers use digital assistants for product advice, and the GTIN is the identifier both UCP and ACP match on. | What was measured | Source, date and scope | | --- | --- | | 48% of home appliance and electronics buyers worldwide used an AI chatbot or assistant during part of their purchase. | [Criteo, Consumer Electronics Pulse 2026](https://www.venturesquare.net/en/1118708/), reported by Venture Square, 10 October 2026. Global electronics buyers | | Home electronics made up 28% of Swiss online sales in 2025, fashion 16%, home and living 15% and food 10%. | [HANDELSVERBAND.swiss with NIQ/GfK and Swiss Post](https://handelsverband.swiss/news/schweizer-online-konsum-waechst-im-jahr-2025-erneut-um-6-das-wachstum-der-einkaeufe-im-ausland-flacht-ab/), 12 March 2026. Switzerland, share by retailers’ main assortment | | 24% of German consumers already use digital assistants for product research or purchase advice, 29% among people over 55. | [GFU Consumer and Home Electronics, YouGov survey of 2,002 people](https://gfu.de/en/?p=3342), 26 February 2026. Germany | | At the W3C and GS1 workshop on e-commerce for humans and AI agents in Zurich on 8 and 9 September 2026, the industry converged on the GTIN as the primary identifier for physical goods, which both UCP and ACP rely on. | [W3C and GS1 workshop](https://forkast.news/what-agent-commerce-needs-from-product-data-lessons-from-the-w3c-gs1-workshop/), reported by Forkast, 10 September 2026. Industry standards | ## Where does an agent purchase break for consumer electronics? We audit consumer electronics stage by stage, because each stage fails for a different reason. The rows below show the failure we look for first at each stage and what changes once it is fixed, in the order a shopper's request meets them, from the first question to the reorder. 1. ### Discover Where it breaks Product pages repeat the manufacturer text found on fifty other shops What changes once it is fixed Answers to the questions your customers ask, which the manufacturer text does not cover 2. ### Understand Where it breaks GTIN and model numbers are missing from the markup What changes once it is fixed Your offer matched to the right model in every comparison, with GTIN and model number declared on each product and variant 3. ### Choose Where it breaks Price and stock in the markup lag the page after a promotion What changes once it is fixed The agent sees the same price and stock as the shopper, because the markup is rendered from the record behind the visible price 4. ### Buy Where it breaks Bot protection blocks verified shopping agents along with scrapers What changes once it is fixed Verified shopping agents get through while scrapers stay blocked, using the card networks’ [agent programmes](https://autobasket.store/learn/agent-commerce-protocols-explained) where available 5. ### Retain Where it breaks Accessories are not linked to the device a customer bought What changes once it is fixed The agent reorders the right accessory or part for the device the customer owns ## Where should electronics retailers start? We usually start this kind of engagement with the Buyer Journey Check. Comparison missions show exactly where a competitor wins the same model from you. From there, the order of the remaining work follows the stage table above, so each step fixes the next place where agents drop your products. [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) Comparison missions show exactly where a competitor wins the same model from you. [Start selling smarter](https://autobasket.store/contact#session) ## What should electronics retailers read next? These guides and worked examples take the questions on this page further for consumer electronics. Each link names the topic it covers and says in one line why it matters for a shop like yours, so you can open the one that answers your next question rather than reading all of them in order. - [Do you need a GTIN for AI shopping agents](https://autobasket.store/learn/why-agents-need-a-gtin) The full rule on GTINs per model and per variant, and what an agent does when one is missing. - [How to get your products recommended by ChatGPT](https://autobasket.store/learn/how-ai-assistants-choose-products) How assistants shortlist products on specs and price before they name a shop. - [UCP vs ACP vs AP2, explained for merchants](https://autobasket.store/learn/agent-commerce-protocols-explained) Which checkout and payment protocols an electronics shop will meet as agent checkout reaches Europe. - [Agentic commerce in Switzerland: TWINT, CHF and VAT](https://autobasket.store/learn/agentic-commerce-switzerland) How CHF prices, VAT and delivery in Switzerland are read by an agent comparing Swiss and foreign shops. ## What do electronics retailers ask about agents? These are the questions teams in consumer electronics ask us most often about AI shopping agents, answered in full. Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### How can a consumer electronics shop win AI comparisons on specs? Declare each specification buyers filter on as a field with its unit, battery life in hours, weight in grams, resolution in pixels, and show the same values in a table on the page. Then keep the price and stock in the markup identical to the page. An agent comparing three shops on one model drops the offer whose numbers it cannot read or confirm. ### Is there a ChatGPT for shopping, and does it cover electronics? ChatGPT answers product questions with shopping results. OpenAI’s help page says it reads structured metadata such as price and product description, and ranks merchants selling the same item on availability, price, quality and whether they are the maker or primary seller. For an electronics retailer, that makes the accuracy of price and stock the deciding fields. ### Why does a GTIN matter for AI shopping agents? An agent comparing offers for one model needs to know that two pages describe the same product, and the GTIN, the number behind the barcode on the box, is how it tells. Without it, your offer may be treated as a different product, or left out of a comparison that you would have won on price. ### Should an electronics shop block AI bots? Blocking every automated visitor also blocks the shopping agents that bring orders. The better approach separates verified agents from scrapers, with crawler rules for answer engines and, where your payment provider supports them, the card networks’ agent verification programmes. ### What happens when the markup price differs from the page price? An agent that sees two prices trusts neither and often moves to another shop. The gap usually opens after a promotion, when the visible price changes and the structured data still carries the old one, and it is fixed by rendering the structured data from the same record that fills the visible price. ## Other industries - [Sports and outdoor](https://autobasket.store/solutions/sports-outdoor) - [Fashion and apparel](https://autobasket.store/solutions/fashion-apparel) - [Food and drink](https://autobasket.store/solutions/food-drink) - [Home and furniture](https://autobasket.store/solutions/home-furniture) - [Beauty and personal care](https://autobasket.store/solutions/beauty-personal-care) - [All solutions](https://autobasket.store/solutions) ## Get your electronics catalogue ready for agents Bring one product page from your catalogue to the first conversation, and we will show you which of the fields above an agent can read today and which one it is missing. [Start selling smarter](https://autobasket.store/contact#session) [See the Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) --- 1. [Home](https://autobasket.store/) 2. [Solutions](https://autobasket.store/solutions) 3. Fashion and apparel # What does agentic commerce change for fashion and apparel shops? Agentic commerce in fashion is decided by size, material and the cost of a wrong fit. Before recommending a jacket, an assistant checks that the size exists, what the fabric is and how returns work. In our July 2026 audit, 34 of 45 fashion shops in Romandie lost the most agents at that product data stage. [Start selling smarter](https://autobasket.store/contact#session) [See where it breaks](#stages) ## Who this page is for - Fashion brands selling direct, with collections that change every season - Multi-brand clothing and shoe retailers - Accessories and jewellery shops where material and origin matter Usual starting point: [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) ## What does a fashion shopper ask an assistant? A typical request reads like this one: "A waterproof hiking jacket for women, size S, recycled materials, that I can return for free if it does not fit." The agent needs the size in stock, the material composition and the return terms as fields, because those three decide whether it recommends the jacket. ## Which product fields decide the purchase in fashion and apparel? In fashion and apparel, five fields decide whether an agent can shortlist a product: size system, colour and material, variants, returns, care and origin. The table shows why each one matters to the shopper's request here and, for your developers, how it is declared so an agent can read it. | Field | Why the agent needs it | How it is declared | | --- | --- | --- | | Size system | EU, UK and US sizes differ, and the agent has to match the person’s size | `size` with `SizeSpecification` and `sizeSystem` | | Colour and material | Shoppers name both in their prompts | `color`, `material` | | Variants | Each colour and size combination has its own stock | [`ProductGroup`](https://autobasket.store/learn/sizes-and-variants-for-agents) with `variesBy` | | Returns | The cost of a wrong size is part of the decision | `MerchantReturnPolicy` with `merchantReturnDays` and `returnFees` | | Care and origin | Recurring questions in follow-up prompts | `countryOfOrigin`, `additionalProperty` | ## What does the data say about AI shopping for fashion? Shoppers will hand clothing to an agent but want a say: in the Global Payments 2026 Agentic Commerce Report, 69% of 16,000 consumers in 7 markets would let an agent spend up to USD 100 on clothing and footwear, and 59% would approve every purchase. The Swiss rows below show where shops lose those agents. | What was measured | Source, date and scope | | --- | --- | | Among 45 fashion, footwear and accessories shops, 34 lost the most agents at product data, and 23 showed their price only as page text. | [Autobasket audit of 240 Romandie online shops](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness), July 2026. Switzerland, Romandie, 45 fashion shops | | In a survey of 16,000 consumers in 7 markets, 69% would let an AI agent spend up to USD 100 on clothing and footwear, and 59% would require approval for every purchase. | [Global Payments, 2026 Agentic Commerce Report](https://fashionunited.uk/news/retail/us-consumers-would-let-ai-agents-buy-clothes-but-not-without-a-say/2026092990613), reported by FashionUnited, September 2026. 7 markets, 16,000 consumers | | Home electronics made up 28% of Swiss online sales in 2025, fashion 16%, home and living 15% and food 10%. | [HANDELSVERBAND.swiss with NIQ/GfK and Swiss Post](https://handelsverband.swiss/news/schweizer-online-konsum-waechst-im-jahr-2025-erneut-um-6-das-wachstum-der-einkaeufe-im-ausland-flacht-ab/), 12 March 2026. Switzerland, share by retailers’ main assortment | ## Where does an agent purchase break for fashion and apparel? We audit fashion and apparel stage by stage, because each stage fails for a different reason. The rows below show the failure we look for first at each stage and what changes once it is fixed, in the order a shopper's request meets them, from the first question to the reorder. 1. ### Discover Where it breaks Collection pages carry a mood image and two lines, with nothing an assistant can quote What changes once it is fixed Fit, fabric and care written as text an assistant can cite 2. ### Understand Where it breaks The size chart is an image What changes once it is fixed A size chart the agent can match to the shopper’s size, as an HTML table and size fields in the markup 3. ### Choose Where it breaks Return terms are only in the footer PDF What changes once it is fixed Return terms the agent can quote on every product, declared as a return policy 4. ### Buy Where it breaks Colour and size selection only works with clicks in a browser What changes once it is fixed The agent can put the chosen colour and size in the cart, using an ID and a [GTIN](https://autobasket.store/learn/why-agents-need-a-gtin) for each variant 5. ### Retain Where it breaks The next season replaces URLs, and earned citations point to dead pages What changes once it is fixed Citations earned last season keep sending shoppers to you, through stable URLs and redirects when a style changes ## Where should fashion brands start? We usually start this kind of engagement with the Buyer Journey Check. Fashion purchases fail at different stages per category, so the audit shows which collections to fix first. From there, the order of the remaining work follows the stage table above, so each step fixes the next place where agents drop your products. [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) Fashion purchases fail at different stages per category, so the audit shows which collections to fix first. [Start selling smarter](https://autobasket.store/contact#session) ## What should fashion brands read next? These guides and worked examples take the questions on this page further for fashion and apparel. Each link names the topic it covers and says in one line why it matters for a shop like yours, so you can open the one that answers your next question rather than reading all of them in order. - [ProductGroup schema for sizes and variants](https://autobasket.store/learn/sizes-and-variants-for-agents) The markup for size systems, colours and size charts that lets an agent match a shopper’s size. - [How to get your products recommended by ChatGPT](https://autobasket.store/learn/how-ai-assistants-choose-products) How assistants retrieve and judge product pages, and what makes them quote a fashion shop by name. - [Are Swiss online shops ready for AI agents? 240 shops audited](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness) The audit behind the fashion figures on this page, with all 45 fashion shops grouped by stage. - [GEO vs SEO for online shops](https://autobasket.store/learn/geo-vs-seo-for-online-shops) What changes in a collection page when it has to be cited by an assistant as well as ranked by Google. ## What do fashion brands ask about agents? These are the questions teams in fashion and apparel ask us most often about AI shopping agents, answered in full. Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### How can my fashion e-shop get recommended by ChatGPT? Give the assistant the facts shoppers put in their prompts as fields: size with its size system, colour, material, stock per variant and return terms. OpenAI’s help page says ChatGPT reads structured metadata such as price and description and ranks sellers of the same item on availability, price and quality. Fit notes written as text give it something to quote. ### How can a fashion brand reduce returns from AI-assisted purchases? By giving the agent enough to pick the right size the first time: a size chart as an HTML table, the size system on every variant, fit notes such as "runs small" in the text, and the fabric composition. An agent that can match the shopper’s measurements to a declared size has less to guess, and the return policy it quotes is your own. ### Can an AI assistant recommend clothing in the right size? It can when the size is declared with its size system and each size has its own availability. Without that, the assistant either guesses or names a shop that publishes the information, which is why size markup comes first in our apparel plans. ### Do return policies affect AI recommendations? Assistants answer follow-up questions about returns, and shopping agents compare return terms when two offers are close. When your return terms are declared in the structured data as a `MerchantReturnPolicy`, they answer from your terms rather than from a forum post about your shop. ### What happens to AI citations when a collection changes? If the old product URLs return errors, the citations an assistant had for them are lost. Keeping stable URLs, or redirecting each retired style to its successor, preserves the visibility earned in previous seasons. ## Other industries - [Sports and outdoor](https://autobasket.store/solutions/sports-outdoor) - [Food and drink](https://autobasket.store/solutions/food-drink) - [Consumer electronics](https://autobasket.store/solutions/consumer-electronics) - [Home and furniture](https://autobasket.store/solutions/home-furniture) - [Beauty and personal care](https://autobasket.store/solutions/beauty-personal-care) - [All solutions](https://autobasket.store/solutions) ## Get your fashion catalogue ready for agents Bring one product page from your catalogue to the first conversation, and we will show you which of the fields above an agent can read today and which one it is missing. [Start selling smarter](https://autobasket.store/contact#session) [See the Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) --- 1. [Home](https://autobasket.store/) 2. [Solutions](https://autobasket.store/solutions) 3. Food and drink # What does agentic commerce mean for food and drink shops? Food and drink is chosen on origin, ingredients and delivery date, and a gift adds a deadline. Our July 2026 audit found a chocolatier near Geneva named in 58% of the Google AI Overviews we tested, while none of its 26 rendered product pages declared a price an agent could read, so the recommendation ended at the product page. [Start selling smarter](https://autobasket.store/contact#session) [See where it breaks](#stages) ## Who this page is for - Chocolatiers, coffee roasters and confectioners selling online - Wine merchants and speciality drink producers - Delicatessen and regional food shops shipping across Switzerland Usual starting point: [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) ## What does a food and drink shopper ask an assistant? A typical request reads like this one: "A box of dark chocolate pralines from a Swiss chocolatier, under CHF 60, delivered to Zürich by Friday as a gift." The agent needs the price, the delivery date to that postcode, gift options and allergens, and it needs to confirm the seller is the chocolatier itself. ## Which product fields decide the purchase in food and drink? In food and drink, five fields decide whether an agent can shortlist a product: price and pack size, ingredients and allergens, origin, delivery date, seller identity. The table shows why each one matters to the shopper's request here and, for your developers, how it is declared so an agent can read it. | Field | Why the agent needs it | How it is declared | | --- | --- | --- | | Price and pack size | Gifts are compared on price per box or per bottle | `Offer.price`, `priceCurrency`, `weight` | | Ingredients and allergens | A follow-up question in almost every food conversation | Visible text near the top and `additionalProperty` | | Origin | Swiss origin, cocoa origin or appellation is often in the prompt | `countryOfOrigin`, `additionalProperty` | | Delivery date | Gifts are bought against an occasion | `shippingDetails` with `deliveryTime` | | Seller identity | The agent checks that it is buying from the maker | `Organization` linked as `seller` | ## What does the data say about AI shopping for food and drink? In our July 2026 audit of 73 food and drink shops in Romandie, 58 lost the most agents at product data and 47 showed their price only as page text. Food is 10% of Swiss online sales, and TWINT holds two thirds of mobile distance-selling transactions, which shapes how an agent will eventually pay. | What was measured | Source, date and scope | | --- | --- | | Among 73 food and drink shops (chocolatiers, wine producers, coffee roasters, breweries and fine food), 58 lost the most agents at product data, and 47 showed their price only as page text. | [Autobasket audit of 240 Romandie online shops](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness), July 2026. Switzerland, Romandie, 73 food and drink shops | | A chocolatier near Geneva was named in 58% of Google AI Overviews and 33% of Perplexity answers we tested, yet none of its 26 rendered product pages declared a price an agent could read. | [Autobasket audit](https://autobasket.store/use-cases/chocolatier-findable-not-buyable), July 2026. Switzerland, one shop, 24 prompts per engine | | Home electronics made up 28% of Swiss online sales in 2025, fashion 16%, home and living 15% and food 10%. | [HANDELSVERBAND.swiss with NIQ/GfK and Swiss Post](https://handelsverband.swiss/news/schweizer-online-konsum-waechst-im-jahr-2025-erneut-um-6-das-wachstum-der-einkaeufe-im-ausland-flacht-ab/), 12 March 2026. Switzerland, share by retailers’ main assortment | | Mobile payment accounts for 74.4% of Swiss distance-selling transactions, and TWINT holds 67.0% of those mobile transactions. | [Swiss Payment Monitor 1/2026, University of St.Gallen and ZHAW](https://fsi.unisg.ch/fileadmin/user_upload/HSG_ROOT/Institut_FSI/Dokumente/Aktuelle_Forschung/Report_Swiss_Payment_Monitor_2026-1_final_EN.pdf), February 2026. Switzerland, domestic distance selling, by number of transactions | ## Where does an agent purchase break for food and drink? We audit food and drink stage by stage, because each stage fails for a different reason. The rows below show the failure we look for first at each stage and what changes once it is fixed, in the order a shopper's request meets them, from the first question to the reorder. 1. ### Discover Where it breaks The shop is named in guides and press, but the product pages do not answer gifting questions What changes once it is fixed Gift, shipping and allergen answers on the pages assistants retrieve 2. ### Understand Where it breaks Prices appear only in body text, with no `Product` or `Offer` markup What changes once it is fixed A price and availability the agent can read on every product, generated from the catalogue as [`Product` and `Offer` markup](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema) 3. ### Choose Where it breaks No ratings or seller identity an agent can read What changes once it is fixed The agent can confirm it is buying from the maker and read your ratings, through an `Organization` block and ratings in the markup 4. ### Buy Where it breaks A cart exists, but an agent cannot confirm the offer it is buying What changes once it is fixed The agent can confirm the exact box and price it is buying, then complete the checkout where the channel allows it, with prices in [CHF including VAT](https://autobasket.store/learn/agentic-commerce-switzerland) 5. ### Retain Where it breaks No reorder or subscription path for coffee or wine What changes once it is fixed Coffee and wine reorders the assistant can place on a schedule the customer approved once ## Where should food producers start? We usually start this kind of engagement with the Machine Readability Sprint. In our chocolatier audit the gap was almost entirely structured data, which the Sprint covers. From there, the order of the remaining work follows the stage table above, so each step fixes the next place where agents drop your products. [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) In our chocolatier audit the gap was almost entirely structured data, which the Sprint covers. [Start selling smarter](https://autobasket.store/contact#session) [Read the chocolatier audit](https://autobasket.store/use-cases/chocolatier-findable-not-buyable) ## What should food producers read next? These guides and worked examples take the questions on this page further for food and drink. Each link names the topic it covers and says in one line why it matters for a shop like yours, so you can open the one that answers your next question rather than reading all of them in order. - [Why AI recommends a chocolatier that agents cannot buy from](https://autobasket.store/use-cases/chocolatier-findable-not-buyable) The case behind the 58% figure: a shop assistants already name, and the product data that stopped agents at checkout. - [Agentic commerce in Switzerland: TWINT, CHF and VAT](https://autobasket.store/learn/agentic-commerce-switzerland) How CHF prices, VAT and TWINT look to an agent, which matters for every Swiss food and wine order. - [Are Swiss online shops ready for AI agents? 240 shops audited](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness) The full audit of 240 shops, including the 73 food and drink shops summarised above. - [How to get your products recommended by ChatGPT](https://autobasket.store/learn/how-ai-assistants-choose-products) How assistants pick the passage they quote, so gifting and allergen answers end up in it. ## What do food producers ask about agents? These are the questions teams in food and drink ask us most often about AI shopping agents, answered in full. Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### How do I get my wine shop recommended by AI assistants? Publish each wine as a product with its price, bottle size, vintage, grape, appellation and stock as fields, and write tasting and pairing notes as text near the top of the page. Assistants retrieve passages that answer the prompt, so "a Chasselas under CHF 25 for a fondue" is answered from pages that state those facts. Age and shipping rules then decide which channels can complete the sale. ### How can a food shop handle shipping and freshness questions from AI agents? Declare the delivery time per region in `shippingDetails`, and state the conditions that matter for perishables in visible text: chilled shipping, the days you dispatch and the shelf life on arrival. An agent asked for a gift by Friday compares a stated delivery time against that date, and it skips an offer that only promises fast delivery. ### Why would an AI assistant recommend a shop it cannot buy from? Because recommendation and purchase rely on different evidence. Assistants name shops that appear in guides and press, while a shopping agent needs a structured price and availability on the product page. Our chocolatier audit found the first without the second, so the recommendation ended on a page an agent could not use. ### How should a food shop declare allergens for AI agents? In visible text near the top of the product page, so it appears in the passage an assistant retrieves, and as `additionalProperty` fields in the structured data. Allergen statements must match the packaging, because what an agent repeats about your product remains your responsibility. ### Can wine and spirits be sold through AI agents? Age checks and alcohol shipping rules differ by country and canton, and agent checkout channels handle them unevenly. We map which assistants can complete such a purchase in your market before building anything, and often the right first step is recommendation rather than checkout. ## Other industries - [Sports and outdoor](https://autobasket.store/solutions/sports-outdoor) - [Fashion and apparel](https://autobasket.store/solutions/fashion-apparel) - [Consumer electronics](https://autobasket.store/solutions/consumer-electronics) - [Home and furniture](https://autobasket.store/solutions/home-furniture) - [Beauty and personal care](https://autobasket.store/solutions/beauty-personal-care) - [All solutions](https://autobasket.store/solutions) ## Get your food and drink catalogue ready for agents Bring one product page from your catalogue to the first conversation, and we will show you which of the fields above an agent can read today and which one it is missing. [Start selling smarter](https://autobasket.store/contact#session) [See the Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) --- 1. [Home](https://autobasket.store/) 2. [Solutions](https://autobasket.store/solutions) 3. Home and furniture # What does AI in furniture retail change for home shops? AI shopping agents buy furniture on dimensions, materials and delivery: will the sofa fit through the door, and when does it arrive. Home and living made up 15% of Swiss online sales in 2025, according to HANDELSVERBAND.swiss. Agents answer only from declared width, height, depth and lead time, so dimensions drawn on a photo keep a product out. [Start selling smarter](https://autobasket.store/contact#session) [See where it breaks](#stages) ## Who this page is for - Furniture brands and retailers with made-to-order products - Lighting, rug and homeware shops - Retailers with long or variable delivery lead times Usual starting point: [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) ## What does a furniture shopper ask an assistant? A typical request reads like this one: "A three-seat sofa under 220 cm wide, washable fabric, delivered to Bern within four weeks." The agent needs the dimensions in centimetres, the fabric and care, and a delivery time it can compare against four weeks. ## Which product fields decide the purchase in home and furniture? In home and furniture, five fields decide whether an agent can shortlist a product: dimensions, material and care, lead time, delivery and assembly, configurations. The table shows why each one matters to the shopper's request here and, for your developers, how it is declared so an agent can read it. | Field | Why the agent needs it | How it is declared | | --- | --- | --- | | Dimensions | The first filter for anything that has to fit a room | `width`, `height`, `depth` as `QuantitativeValue` | | Material and care | Washable, solid wood or outdoor use appear in prompts | `material`, `additionalProperty` | | Lead time | Made-to-order items are compared on delivery date | `shippingDetails` with `deliveryTime` | | Delivery and assembly | Part of the real price of a large item | `shippingRate` and stated assembly options | | Configurations | Fabric and size options change price and lead time | [`ProductGroup`](https://autobasket.store/learn/sizes-and-variants-for-agents) with `variesBy` | ## What does the data say about AI shopping for furniture? Furniture can already be bought inside an assistant in the US: Ashley, North America’s largest furniture store brand, launched a fully transactional shopping experience in Perplexity in November 2025, with PayPal handling payment. In Switzerland, home and living is 15% of online sales, and 5% of DACH shoppers have used an agent so far. | What was measured | Source, date and scope | | --- | --- | | Ashley, North America’s largest furniture store brand, launched a fully transactional shopping experience inside Perplexity, with PayPal handling payment. | [Ashley](https://www.woodworkingnetwork.com/furniture/ashley-announces-partnership-ai-powered-answer-engine), reported by Woodworking Network, 30 November 2025. United States | | Home electronics made up 28% of Swiss online sales in 2025, fashion 16%, home and living 15% and food 10%. | [HANDELSVERBAND.swiss with NIQ/GfK and Swiss Post](https://handelsverband.swiss/news/schweizer-online-konsum-waechst-im-jahr-2025-erneut-um-6-das-wachstum-der-einkaeufe-im-ausland-flacht-ab/), 12 March 2026. Switzerland, share by retailers’ main assortment | | 5% of online shoppers in Switzerland, Germany and Austria have already used an AI agent to help them shop, and about 18% plan to. | [University of St.Gallen (HSG), survey of just under 3,000 people](https://www.unisg.ch/en/newsdetail/news/ki-agenten-im-onlinehandel-warum-konsumentinnen-und-konsumenten-beim-einkauf-noch-zoegern/), 24 June 2026. Switzerland, Germany and Austria | ## Where does an agent purchase break for home and furniture? We audit home and furniture stage by stage, because each stage fails for a different reason. The rows below show the failure we look for first at each stage and what changes once it is fixed, in the order a shopper's request meets them, from the first question to the reorder. 1. ### Discover Where it breaks Inspiration pages carry images and little text an assistant can cite What changes once it is fixed Room and use-case guides written as answers 2. ### Understand Where it breaks Dimensions are drawn on a product photo What changes once it is fixed The agent can check the piece against the shopper’s measurements, with dimensions as [product schema fields](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema) and in a table 3. ### Choose Where it breaks Delivery reads "on request", so the agent has no date to compare What changes once it is fixed A delivery date the agent can compare for each fabric and size 4. ### Buy Where it breaks Configurators work only with clicks and previews What changes once it is fixed The agent can order the exact configuration, using an ID for each valid combination 5. ### Retain Where it breaks Covers, spare parts and matching pieces are not linked What changes once it is fixed Covers, spare parts and matching pieces the agent can find from the item the customer bought ## Where should furniture retailers start? We usually start this kind of engagement with the Machine Readability Sprint. Dimensions and lead times are catalogue data that the Sprint turns into markup in a week. From there, the order of the remaining work follows the stage table above, so each step fixes the next place where agents drop your products. [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) Dimensions and lead times are catalogue data that the Sprint turns into markup in a week. [Start selling smarter](https://autobasket.store/contact#session) ## What should furniture retailers read next? These guides and worked examples take the questions on this page further for home and furniture. Each link names the topic it covers and says in one line why it matters for a shop like yours, so you can open the one that answers your next question rather than reading all of them in order. - [ProductGroup schema for sizes and variants](https://autobasket.store/learn/sizes-and-variants-for-agents) How fabric, colour and size options are declared so each configuration has its own price and lead time. - [llms.txt, Markdown and product schema for online shops](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema) Which product schema fields agents read, including dimensions and delivery details for large items. - [How AI shopping agents compare products, traced on a gravel bike](https://autobasket.store/use-cases/gravel-bike-under-chf-4000) A worked comparison showing how a declared delivery time decides between two shops at different prices. - [Agentic commerce in Switzerland: TWINT, CHF and VAT](https://autobasket.store/learn/agentic-commerce-switzerland) How Swiss delivery, CHF prices and VAT are presented to an agent comparing home shops. ## What do furniture retailers ask about agents? These are the questions teams in home and furniture ask us most often about AI shopping agents, answered in full. Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### How do I get furniture with many variants recommended by AI? Group the configurations under one `ProductGroup`, declare what they vary by, such as fabric, colour or size, and give each valid combination its own identifier, price, dimensions and delivery time. An agent asked for a grey three-seater under 220 cm then filters the variants itself. A configurator that builds the combination only in the browser leaves the agent one generic product to judge. ### How do AI agents handle delivery dates for bulky furniture? They compare the delivery time an offer declares against the date the shopper gives, so a range per configuration in `shippingDetails`, such as three to five weeks, is what they can use. State the delivery and assembly cost too, since it is part of the real price. "On request" gives the agent no date, and it moves to an offer that has one. ### Can an AI assistant check whether furniture fits a room? It can compare declared dimensions against the measurements a person gives. It cannot read numbers drawn on a photo, so a sofa whose width is only in an image is either skipped or described with a guess. ### How should a shop declare made-to-order delivery times? As a delivery time range per configuration in the structured data, and the same range in visible text. "On request" gives an agent nothing to compare, so it tends to choose an offer that states a date. ### Do product configurators work with AI agents? Not when the configuration only exists inside a visual tool. Each valid configuration needs an identifier and a price that an agent can request, either as variants in the catalogue or through a tool the agent can call over MCP, the protocol agents use to query a shop. ## Other industries - [Sports and outdoor](https://autobasket.store/solutions/sports-outdoor) - [Fashion and apparel](https://autobasket.store/solutions/fashion-apparel) - [Food and drink](https://autobasket.store/solutions/food-drink) - [Consumer electronics](https://autobasket.store/solutions/consumer-electronics) - [Beauty and personal care](https://autobasket.store/solutions/beauty-personal-care) - [All solutions](https://autobasket.store/solutions) ## Get your furniture catalogue ready for agents Bring one product page from your catalogue to the first conversation, and we will show you which of the fields above an agent can read today and which one it is missing. [Start selling smarter](https://autobasket.store/contact#session) [See the Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) --- 1. [Home](https://autobasket.store/) 2. [Solutions](https://autobasket.store/solutions) 3. Multi-brand retailers # How does agentic commerce for retailers decide who gets the sale? A multi-brand retailer sells products other shops sell too, so an agent compares the offer rather than the product: price, stock, delivery time, returns and service. OpenAI says ChatGPT ranks merchants selling the same item on availability, price and quality. A retailer wins when each offer is matched by GTIN and its terms are declared on every page. [Start selling smarter](https://autobasket.store/contact#session) [See where it breaks](#stages) ## Who this page is for - Retailers carrying many brands, online or online and in store - E-commerce teams managing large catalogues and frequent price changes - Retailers with store stock that agents could reserve or collect Usual starting point: [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) ## What does a retail customer ask an assistant? A typical request reads like this one: "Where can I buy these trail running shoes in size 43 today in Geneva, or with delivery by tomorrow?" The agent needs to match the exact model, then compare stock in size 43 per store, delivery times and price across the shops that carry it. ## Which product fields decide the purchase in multi-brand retailers? In multi-brand retailers, five fields decide whether an agent can shortlist a product: identifiers, store stock, delivery and returns, price accuracy, catalogue scale. The table shows why each one matters to the shopper's request here and, for your developers, how it is declared so an agent can read it. | Field | Why the agent needs it | How it is declared | | --- | --- | --- | | Identifiers | Matching the same product across shops | [`gtin13`](https://autobasket.store/learn/why-agents-need-a-gtin), `mpn`, `brand` | | Store stock | Same-day and click-and-collect prompts | `availableAtOrFrom` with store locations | | Delivery and returns | Where retailers actually differ from each other | `shippingDetails`, `MerchantReturnPolicy` | | Price accuracy | Frequent changes create gaps between page and markup | Markup rendered from the live price | | Catalogue scale | Fixes have to apply by template, not page by page | Template-level JSON-LD generation | ## What does the data say about AI shopping for retailers? In Switzerland the comparison already runs against marketplaces: 13 of the 30 largest B2C online shops are marketplaces, according to HANDELSVERBAND.swiss, and Galaxus alone reported CHF 3.764 billion in platform sales for 2025. In the US, AI-referred visits converted 42% better than other traffic in March 2026, which makes the comparison worth winning. | What was measured | Source, date and scope | | --- | --- | | 13 of the 30 largest Swiss B2C online shops are marketplaces. | [HANDELSVERBAND.swiss, citing a Carpathia analysis](https://handelsverband.swiss/news/schweizer-online-konsum-waechst-im-jahr-2025-erneut-um-6-das-wachstum-der-einkaeufe-im-ausland-flacht-ab/), 12 March 2026. Switzerland, top 30 B2C online shops | | Galaxus reported CHF 3.764 billion in platform sales for 2025, up 17%, with 10 million products on sale. | [Digitec Galaxus AG](https://www.digitec.ch/en/page/galaxus-grows-by-17-per-cent-in-2025-40855), 7 January 2026. Switzerland and EU, company figures | | ChatGPT reads structured metadata such as price and product description, and ranks merchants selling the same item on availability, price, quality and whether they are the maker or primary seller. | [OpenAI Help Center, Shopping with ChatGPT Search](https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search), fetched 16 October 2026. ChatGPT shopping results | | In March 2026, visits referred by AI tools to US retail sites converted 42% better than other traffic, a year after converting 38% worse. | [Adobe Analytics](https://techcrunch.com/2026/04/16/ai-traffic-to-us-retailers-rose-393-in-q1-and-its-boosting-their-revenue-too/), reported by TechCrunch, 16 April 2026. United States retail | ## Where does an agent purchase break for multi-brand retailers? We audit multi-brand retailers stage by stage, because each stage fails for a different reason. The rows below show the failure we look for first at each stage and what changes once it is fixed, in the order a shopper's request meets them, from the first question to the reorder. 1. ### Discover Where it breaks Category pages are thin lists of products with no answers What changes once it is fixed Buying guides per category that assistants can cite 2. ### Understand Where it breaks Identifiers are missing for part of the catalogue What changes once it is fixed Every product matched to the right model, with missing identifiers measured and filled by template 3. ### Choose Where it breaks Delivery and return terms differ from what the markup says What changes once it is fixed The agent reads the same delivery and return terms your customers see, declared once and applied to every offer 4. ### Buy Where it breaks Store stock is visible to people but not to agents What changes once it is fixed The agent can check store stock and reserve for collection, through tools exposed over MCP 5. ### Retain Where it breaks Agent orders land in analytics as direct traffic What changes once it is fixed You see which orders came from agents, attributed in GA4 through [agentic checkout](https://autobasket.store/products/autobasket-convert) order events ## Where should retailers start? We usually start this kind of engagement with the Buyer Journey Check. Comparison missions show which competitor wins your categories and on which term. From there, the order of the remaining work follows the stage table above, so each step fixes the next place where agents drop your products. [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) Comparison missions show which competitor wins your categories and on which term. [Start selling smarter](https://autobasket.store/contact#session) ## What should retailers read next? These guides and worked examples take the questions on this page further for multi-brand retailers. Each link names the topic it covers and says in one line why it matters for a shop like yours, so you can open the one that answers your next question rather than reading all of them in order. - [How AI shopping agents compare products, traced on a gravel bike](https://autobasket.store/use-cases/gravel-bike-under-chf-4000) A worked comparison between shops selling similar bikes, and the fields that decided the shortlist. - [Do you need a GTIN for AI shopping agents](https://autobasket.store/learn/why-agents-need-a-gtin) How agents match the same product across retailers, and what happens to offers without an identifier. - [Agentic commerce in Switzerland: TWINT, CHF and VAT](https://autobasket.store/learn/agentic-commerce-switzerland) The Swiss market context for retailers: marketplaces, TWINT, CHF prices and VAT as an agent reads them. - [What is agentic commerce](https://autobasket.store/learn/what-is-agentic-commerce) The definition and examples to share with colleagues who are new to agent-led buying. ## What do retailers ask about agents? These are the questions teams in multi-brand retailers ask us most often about AI shopping agents, answered in full. Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### How can a Swiss shop compete with Galaxus when agents compare prices? On the terms an agent can verify where a specialist often beats a marketplace: stock in the size or colour asked for, delivery time to the postcode, store pickup, return terms and advice. Declare each of them on every offer, matched to the product by GTIN. Galaxus lists 10 million products, so a smaller shop rarely wins on range and wins instead on a complete, accurate offer. ### Should I sell on Galaxus or invest in my own shop for AI agents? Often both, for different jobs. A marketplace listing puts your offers where many comparisons already start, while your own shop is where you control the data an agent reads, the return terms and the customer relationship. Agents compare offers by GTIN across both, so the data has to be complete and identical wherever the product appears, or one listing undercuts the other. ### How do AI agents choose between retailers selling the same product? They match the product by identifier, then compare the offers on price, availability, delivery time, return terms and ratings. A retailer wins by declaring all five accurately, since an offer with a missing field is often dropped before the comparison starts. ### Can AI agents use in-store stock? They can when store availability is declared per location or exposed through a tool the agent can call. That opens prompts such as "available today in Geneva", which a shop with online stock only cannot answer. ### How do you fix structured data across a large catalogue? By template rather than by page. We measure coverage across the catalogue, fix the templates that generate the markup, and validate after each catalogue change so new products ship with complete data. ## Other business models - [Brands selling direct](https://autobasket.store/solutions/brands-selling-direct) - [B2B and wholesale](https://autobasket.store/solutions/b2b-wholesale) - [All solutions](https://autobasket.store/solutions) ## Get your retail catalogue ready for agents Bring one product page from your catalogue to the first conversation, and we will show you which of the fields above an agent can read today and which one it is missing. [Start selling smarter](https://autobasket.store/contact#session) [See the Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) --- 1. [Home](https://autobasket.store/) 2. [Solutions](https://autobasket.store/solutions) 3. Sports and outdoor # How do AI shopping agents choose sporting goods and outdoor gear? AI shopping agents choose sporting goods on specifications and fit: frame size, weight, waterproof rating and stock in the size asked for. In our July 2026 audit of 21 sports and outdoor shops in Romandie, 13 lost the most agents at product data. When stock per size is unreadable, the agent compares on guesswork or leaves the shop out. [Start selling smarter](https://autobasket.store/contact#session) [See where it breaks](#stages) ## Who this page is for - Bike shops and brands selling complete bikes, frames and components - Ski, running, climbing and camping retailers with size-dependent stock - Brands whose products are compared on measured specifications Usual starting point: [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) ## What does a sports shopper ask an assistant? A typical request reads like this one: "Find me a gravel bike under CHF 4,000 in size M that can be delivered to Lausanne before the end of the month." To answer, the agent needs the price, the size, stock for that size and a delivery time it can compare, all as fields rather than prose. ## Which product fields decide the purchase in sports and outdoor? In sports and outdoor, five fields decide whether an agent can shortlist a product: size and fit per variant, stock per size, specifications, compatibility, delivery time. The table shows why each one matters to the shopper's request here and, for your developers, how it is declared so an agent can read it. | Field | Why the agent needs it | How it is declared | | --- | --- | --- | | Size and fit per variant | Agents filter on the size the person asked for before anything else | [`ProductGroup`](https://autobasket.store/learn/sizes-and-variants-for-agents) with `hasVariant` and `variesBy: size` | | Stock per size | An agent drops a product whose requested size it cannot confirm | `Offer.availability` on each variant | | Specifications | Weight, gearing or waterproof rating decide the comparison | `additionalProperty` with units | | Compatibility | Components and spare parts are bought for a specific model | `isAccessoryOrSparePartFor` | | Delivery time | Seasonal purchases are bought against a date | `shippingDetails` with `deliveryTime` | ## What does the data say about AI shopping for sports gear? The most useful number for a sports shop comes from our July 2026 audit: 13 of 21 sports and outdoor shops in Romandie lost the most agents at product data, and 12 showed their price only as page text. Agent checkout opened in the US first, and JD Sports shows where the channel is heading. | What was measured | Source, date and scope | | --- | --- | | Among 21 sports and outdoor shops (ski and outdoor apparel, cycling and sports nutrition), 13 lost the most agents at product data, and 12 showed their price only as page text. | [Autobasket audit of 240 Romandie online shops](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness), July 2026. Switzerland, Romandie, 21 sports shops | | JD Sports let US customers search and buy through AI platforms, starting with Microsoft Copilot and then Gemini and ChatGPT, built on commercetools and Stripe. | [Reuters](https://in.marketscreener.com/news/britain-s-jd-sports-launches-ai-shopping-for-us-customers-ce7e58dad08bfe22), reported by MarketScreener, 12 January 2026. United States | | 5% of online shoppers in Switzerland, Germany and Austria have already used an AI agent to help them shop, and about 18% plan to. | [University of St.Gallen (HSG), survey of just under 3,000 people](https://www.unisg.ch/en/newsdetail/news/ki-agenten-im-onlinehandel-warum-konsumentinnen-und-konsumenten-beim-einkauf-noch-zoegern/), 24 June 2026. Switzerland, Germany and Austria | ## Where does an agent purchase break for sports and outdoor? We audit sports and outdoor stage by stage, because each stage fails for a different reason. The rows below show the failure we look for first at each stage and what changes once it is fixed, in the order a shopper's request meets them, from the first question to the reorder. 1. ### Discover Where it breaks Category guides are long single pages, so the passage about one bike competes with eleven others What changes once it is fixed One page per question, with the answer in its opening lines 2. ### Understand Where it breaks Sizes live in a drop-down built by JavaScript that the agent never runs What changes once it is fixed Every size visible to the agent as its own option, rendered in the HTML and declared as a `ProductGroup` 3. ### Choose Where it breaks Specifications sit in an image or a PDF What changes once it is fixed Specifications the agent can compare line by line, as table rows and `additionalProperty` fields, with a [GTIN](https://autobasket.store/learn/why-agents-need-a-gtin) per size where the maker assigns one 4. ### Buy Where it breaks The agent cannot confirm stock for the size before building a cart What changes once it is fixed The agent can check that the size is in stock before it builds the cart, through a stock tool exposed over MCP 5. ### Retain Where it breaks Servicing and spare parts are not linked to the model sold What changes once it is fixed Servicing and spare parts the agent can match to the bike or ski the customer already owns ## Where should sports retailers start? We usually start this kind of engagement with the Machine Readability Sprint. Variant and stock markup has to be in place before an agent can match a size, and it fits in seven days. From there, the order of the remaining work follows the stage table above, so each step fixes the next place where agents drop your products. [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) Variant and stock markup has to be in place before an agent can match a size, and it fits in seven days. [Start selling smarter](https://autobasket.store/contact#session) [Trace a gravel bike purchase from prompt to checkout](https://autobasket.store/use-cases/gravel-bike-under-chf-4000) ## What should sports retailers read next? These guides and worked examples take the questions on this page further for sports and outdoor. Each link names the topic it covers and says in one line why it matters for a shop like yours, so you can open the one that answers your next question rather than reading all of them in order. - [How AI shopping agents compare products, traced on a gravel bike](https://autobasket.store/use-cases/gravel-bike-under-chf-4000) Follows one bike request from the prompt to the shortlist and shows why the shop with structured delivery data won. - [ProductGroup schema for sizes and variants](https://autobasket.store/learn/sizes-and-variants-for-agents) Shows how frame and shoe sizes are declared so an agent can match the size the shopper asked for. - [Do you need a GTIN for AI shopping agents](https://autobasket.store/learn/why-agents-need-a-gtin) Explains when a bike or component needs a GTIN per size and what the agent does without one. - [Are Swiss online shops ready for AI agents? 240 shops audited](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness) The full audit behind the sports figures on this page, with the method and every category. ## What do sports retailers ask about agents? These are the questions teams in sports and outdoor ask us most often about AI shopping agents, answered in full. Each answer stands on its own, so you can forward one to a colleague, and an assistant asked the same question can quote it as it is. ### How can my sports shop get recommended when people ask ChatGPT for gear? Publish the facts the prompt filters on as fields an assistant can read: size per variant, stock for that size, measured specifications with units and a delivery time. OpenAI’s shopping help page says ChatGPT reads structured metadata such as price and description, so a shop whose specifications sit in a PDF gives it nothing to compare and is usually left off the shortlist. ### How should a bike shop structure product data for AI assistants? One product page per model, with each frame size declared as a variant in a `ProductGroup`, its own availability and, where the manufacturer assigns one, its own GTIN. Geometry, weight and gearing go in a table on the page and in `additionalProperty` fields with units, and compatible parts link to the model with `isAccessoryOrSparePartFor`. ### Which sports retailers already sell through AI assistants? JD Sports is the clearest example. Reuters reported in January 2026 that it let US customers search and buy through AI platforms, starting with Microsoft Copilot and then Gemini and ChatGPT, built on commercetools and Stripe. Agent checkout has opened in the US first, so a Swiss sports shop is preparing its data ahead of the rollout. ### Do AI assistants handle products sold in many sizes? They handle them when each size is a declared variant with its own availability. When the page shows one product with a size selector built in JavaScript, the assistant usually sees one price and no sizes, and it cannot tell whether the size the person asked for is in stock. ### Should a bike shop publish its specifications as text or as a table? Both, from the same source. A table on the page serves people and the agents that read the page, and the same values as `additionalProperty` fields in the structured data serve agents that compare offers without reading any prose. ### Where should a sports retailer start? With the twenty products that bring the most revenue in the coming season. Fixing their variants, stock and delivery data first gives the fastest read on whether agent traffic and citations move. ## Other industries - [Fashion and apparel](https://autobasket.store/solutions/fashion-apparel) - [Food and drink](https://autobasket.store/solutions/food-drink) - [Consumer electronics](https://autobasket.store/solutions/consumer-electronics) - [Home and furniture](https://autobasket.store/solutions/home-furniture) - [Beauty and personal care](https://autobasket.store/solutions/beauty-personal-care) - [All solutions](https://autobasket.store/solutions) ## Get your sports catalogue ready for agents Bring one product page from your catalogue to the first conversation, and we will show you which of the fields above an agent can read today and which one it is missing. [Start selling smarter](https://autobasket.store/contact#session) [See the Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) --- 1. [Home](https://autobasket.store/) 2. Buyer Journey Check # How does an agentic commerce readiness check find where agents drop your shop? The Buyer Journey Check is our agentic commerce readiness check and audit. We give ChatGPT, Perplexity, Claude and a test agent the shopping tasks your customers would, record where each one stops naming, comparing or buying your products, from the first question to the reorder, and hand your developers the fixes in order of the sales they recover. [Book a Buyer Journey Check](https://autobasket.store/contact#buyer-journey-check) [See what is tested](#diagnostic-framework) Discover Understand Choose Buy Retain Check ## What are the five stages evaluated during the audit? The audit follows the five stages of a purchase made through an assistant: whether it finds your shop, understands the product, can compare your offer, can complete checkout and can come back for the next order. Each stage gets a score, the evidence behind it and a list of fixes ordered by the sales they recover. 1. ### Discover - Search and shopping assistants are allowed to read your pages - Your products show without JavaScript having to run - Assistants name you when customers ask about your category 2. ### Understand - Each product, its offer and your company are described as data (Product, Offer and Organization schema) - Every product carries its barcode number (GTIN), brand and model, so it matches reviews elsewhere - The price an agent reads is the price a person sees 3. ### Choose - Shipping cost and delivery time are stated as fields an agent can compare - Your return terms can be read by an agent - Your ratings can be read by an agent 4. ### Buy - An agent can get from cart to paid order - The checkout connections open in your market are in place - Text on your product pages cannot be used to hijack an agent (prompt injection) 5. ### Retain - An agent can check the status of an order - An agent can reorder or renew a subscription - Agent orders show up in your analytics ### What did the same checks find on 240 Swiss online shops? In July 2026 we ran the readiness checks on 240 online shops in French-speaking Switzerland. Most lost agents at the Understand stage, because price, stock and identity were written for people only. The [full Romandie audit](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness) breaks the results down by stage and by sector. | What was measured | Source, date and scope | | --- | --- | | None of 240 online shops in French-speaking Switzerland scored an A for agent readiness. The median score was 57 out of 100, and 104 shops scored D or F. | [Autobasket audit of 240 Romandie online shops](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness), July 2026. Switzerland, Romandie, 240 shops | | For 157 of 211 shops (74%), the stage that lost the most agents was reading the product itself: price, stock and identity were not declared as structured data. | [Autobasket audit of 240 Romandie online shops](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness), July 2026. Switzerland, Romandie, 211 shops read in full | | On 117 of 211 shops (55%), the only price an agent could find was in the page text, with no Offer price in the structured data to confirm it. | [Autobasket audit of 240 Romandie online shops](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness), July 2026. Switzerland, Romandie, 211 shops read in full | | About 34% of individual retail product pages cannot be properly accessed by AI tools. | [Adobe, AI Content Visibility Checker](https://techcrunch.com/2026/04/16/ai-traffic-to-us-retailers-rose-393-in-q1-and-its-boosting-their-revenue-too/), reported by TechCrunch, 16 April 2026. Retail product pages tested by Adobe | ## What makes an agent drop one bike shop and buy from another? Asked for a gravel bike under CHF 4,000 for alpine rides, available in Switzerland, an agent checks four things on every shop. It drops shops that load the price with JavaScript, show the geometry as an image or describe delivery vaguely, and it recommends and buys from shops that state those facts as data it can read. | What the agent checks | Shop that is dropped | Shop that is bought from | | --- | --- | --- | | Price under CHF 4,000 | Price loaded by JavaScript, absent from the page itself | Price stated in the page code as `price: 3890.00`, `priceCurrency: "CHF"` | | Suited to long climbs | Geometry chart as an image | Geometry and gearing in an HTML table | | The same model as reviewed elsewhere | No barcode number (GTIN) | Barcode number, brand and model declared | | In Switzerland within a week | "Fast delivery" in the footer | Two to three days to Switzerland, declared in `shippingDetails` | [Follow the gravel bike mission step by step](https://autobasket.store/use-cases/gravel-bike-under-chf-4000) ## What is included in the audit package? You receive four documents: a one-page scorecard per stage for decision-makers, a report on how often assistants name you and who they name instead, an analysis of which agent checkout routes your market offers and which you lack, and a 30-day Technical Roadmap. Every finding comes with the change that fixes it, down to the code. - ### Readiness scorecard Where you stand at each stage, the evidence behind it and what to fix first. - ### AI visibility report How often each assistant names you when customers ask, and who it names instead, with the margin of error on every rate, measured as described in [how to measure share of model](https://autobasket.store/learn/share-of-model-explained). - ### Protocol gap analysis Which ways of letting an agent check out your channels need, from MCP and UCP to ACP and AP2, and which ones your shop is missing today, explained in [UCP vs ACP vs AP2 vs MCP](https://autobasket.store/learn/agent-commerce-protocols-explained). - ### 30-day technical roadmap Tickets your developers can pick up, each with the code change it needs. ## How do you commission a Buyer Journey Check? Send us your shop address and the three categories that matter most to your revenue. We confirm the scope, the missions we will run and a fixed fee, then run the audit on public data, with no access to your systems needed, and walk your team through the findings when the report is ready. 1. You send the shop address and your priority categories, which follow your [industry or business model](https://autobasket.store/solutions). 2. We reply with the missions, the assistants tested, the timing and a fixed fee. 3. We run the audit from the outside, using public pages only. 4. We present the findings to your team and hand over the four documents. ## What do merchants ask before booking the audit? E-commerce and marketing leads usually ask four things before booking: what a buying mission is, what they hold at the end, whether we need access to their systems and how the fee is set. The answers below give you the shape of the audit before we speak. ### What is a buying mission? A task a customer would plausibly give an assistant, such as finding a gravel bike under CHF 4,000 in size M delivered before a given date. We run each mission through ChatGPT, Perplexity, Claude and a test agent connected through MCP, the standard agents use to query a shop directly, and record where it stops. ### What do we receive at the end? A score per stage with the evidence behind it, the missions we ran and where each one stopped, and a list of fixes ordered by the sales each one recovers, written so your developers can work from it directly. ### Is my website ready for AI shopping agents? Three signals give a first answer in minutes: your robots.txt lets search and shopping crawlers in, each product page declares a Product with an Offer carrying the price, currency and availability, and that price matches the one a person sees. In our July 2026 audit of 240 Romandie shops, none passed every stage, and 117 of the 211 we read in full showed their price only as page text. ### How is the Buyer Journey Check different from a free readiness score? A free score, such as the product page check on our home page, reads the code of one page and lists what blocks an agent. The Buyer Journey Check runs your customers' shopping tasks through the assistants themselves, measures how often they name you and your competitors with a margin of error, tests the checkout routes open in your market and ranks the fixes by the sales they recover. ### Do you need access to our systems? No. The audit reads your shop from the outside, the way an agent does, using public pages and the assistants themselves. Access is needed only later, if you go on to the Machine Readability Sprint. ### How is the fee set? As a fixed fee per shop, based on the number of categories and markets we test. You receive it in writing before anything starts. ## See your shop the way an agent does The free check reads one product page in a minute. The Buyer Journey Check covers your categories, your competitors and every stage of the purchase, so your developers fix the problems that cost you the most sales first. [Book a Buyer Journey Check](https://autobasket.store/contact#buyer-journey-check) [Run the free check first](https://autobasket.store/#readiness-scan) --- 1. [Home](https://autobasket.store/) 2. Machine Readability Sprint # How does a seven-day sprint make your online shop agent-ready? An agent-ready shop states what each product is, what it costs, whether it is in stock and how it ships as data an AI assistant reads without guessing. The Machine Readability Sprint adds that layer in seven days on top of the platform you already run, tests it with real assistants and leaves your checkout untouched. [Reserve a sprint slot](https://autobasket.store/contact#sprint) [Check your platform](#compatibility-list) Data Pages Index Checkout Test Sprint ## What can an agent do with your shop after the sprint? After the sprint, an agent can get a list of your pages and their full text in one request, read each product's price, stock, shipping and return terms as data, open your content pages without banners or scripts, and ask your shop directly for search results and stock. Every piece arrives as code or CDN settings, tested before handover. | Deliverable | Where it lives | What it changes | | --- | --- | --- | | `llms.txt` and `llms-full.txt`, an index of your site for AI tools, see [llms.txt for ecommerce](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema) | Site root | An agent finds all your pages and reads their full text in one request | | JSON-LD `@graph`, your product facts as structured data | Every product, category and policy page | An agent reads product, offer, seller, shipping and returns as fields | | Markdown mirrors, plain-text copies of your pages | Same path as each content page, with `.md` | An agent reads the page text without layout, scripts or banners | | HTTP Link headers, pointers to those copies | CDN or server configuration | An agent finds the plain-text copy from the page response itself | | MCP tools, the standard agents use to query a shop, see [the protocols compared](https://autobasket.store/learn/agent-commerce-protocols-explained) | Hosted endpoint, reading your product API | An agent searches your catalogue, checks stock and reads offers | | UCP or ACP routing, the protocols agents use to check out | Your checkout, where your channels use them | An agent hands a ready order to your checkout, the first step of [agentic checkout with Convert](https://autobasket.store/products/autobasket-convert) | | Validation report | Handed over on day seven | You see every test we ran, with the assistant and the result | ## How does the sprint work without replatforming? Your shop keeps running as it is: the checkout, the product database and the theme stay untouched, and removing the layer returns everything to its previous state. On most platforms the layer is a theme snippet or app plus a small worker at your CDN that serves the plain-text copies and headers. Shopify and Shopify Plus A theme snippet for the product data, an app proxy for the plain-text copies and llms.txt WooCommerce A plugin that turns your product data into JSON-LD and serves the plain-text copies Adobe Commerce (Magento) A module for the product data, with the plain-text copies served from the CDN Salesforce Commerce Cloud A cartridge for the product data, an edge worker for the plain-text copies and headers SAP Commerce (Hybris) Edge worker in front of the storefront, fed by the product export Headless and custom stacks Edge worker or middleware, reading your product API ## Which product data does an agent-ready shop declare? An agent compares offers on a handful of fields, and each one has a Schema.org property that states it without guessing. For a [Swiss shop](https://autobasket.store/learn/agentic-commerce-switzerland) that means the price in CHF with VAT included, the delivery time to a Swiss address and the return terms, declared on the page a person sees, so data and text agree. | What the agent checks | Schema.org property | Swiss detail to declare | | --- | --- | --- | | Price | `Offer.price`, `priceCurrency`, `PriceSpecification.valueAddedTaxIncluded` | CHF, with VAT included, as on the page | | Stock | `Offer.availability` on each variant | Stock for the size or colour asked for, not for the product as a whole | | Identity | `gtin`, `brand`, `mpn` | The [GTIN](https://autobasket.store/learn/why-agents-need-a-gtin) from the manufacturer or your GS1 Switzerland number | | Variants | `ProductGroup` with `hasVariant` and `variesBy` | One declared variant per size and colour, see [sizes and variants for agents](https://autobasket.store/learn/sizes-and-variants-for-agents) | | Delivery | `shippingDetails` with `deliveryTime` and `shippingDestination` | Delivery days to an address in Switzerland, and the carrier you ship with | | Returns | `hasMerchantReturnPolicy` with `merchantReturnDays` and `returnFees` | Return window and who pays for the return | | Seller | `Organization` with name, address and contact | Your company and its Swiss address | ## What access is needed to start the sprint? Your team gives us read access and a kickoff session, and we do the rest: the build, validation, deployment and tests, with a progress report at the end of each day. That means read-only access to your product feed or catalogue API, permission to add DNS or CDN rules and one session with your technical owner. - Read-only access to the product feed or catalogue API - Permission to add rules at your CDN or DNS provider - A staging environment, or approval to deploy behind a feature flag - One kickoff session with your technical owner - A named person to approve the final deployment ## What do teams ask before booking the sprint? Developers and e-commerce leads usually ask four things before reserving a Sprint: what happens during the seven days, which platforms it runs on, what their own team has to do and how it relates to Discover. The answers below let the kickoff start with the access already in place. ### What happens during the seven days? We turn your catalogue into structured product data, publish llms.txt and plain-text copies of your pages, and set up the agent tools listed in the deliverables table, then check everything against the readiness tests before handover. ### Which platforms does the sprint support? Shopify, WooCommerce, Adobe Commerce, Salesforce Commerce Cloud, SAP Commerce and custom stacks, as listed in the compatibility section, all without moving your shop to a new platform. ### How much does it cost to make an online shop agent-ready? The Sprint is a fixed fee per shop, quoted after a first conversation, because the work depends on the platform, the number of products and variants, and the languages and markets you sell in. A Buyer Journey Check beforehand tells you which stage loses the most sales, so the seven days go to the fix that pays back first. ### Do we need a product feed for ChatGPT? It depends on your platform. OpenAI states that Shopify merchants are included in ChatGPT shopping through Shopify Catalog without extra work, and that other merchants can apply to send a product feed directly. The Sprint makes sure the feed, the structured data on the page and the visible page carry the same price, stock and delivery terms, since a mismatch is a reason for an agent to drop the offer. ### Should we let GPTBot and other AI crawlers into the shop? For product and policy pages, usually yes, since an assistant cannot name or compare what its crawler cannot read. The Sprint reviews your robots.txt crawler by crawler, separating search and shopping crawlers from training crawlers, so you decide each one on purpose. Our guide to llms.txt, Markdown and schema for online shops explains what each crawler reads. ### What does our team need to do? Grant the access listed above and name one developer who can review and merge changes. The rest of the work runs on our side. ### How does the sprint relate to Discover? The Sprint builds the readable layer once. Discover is the monthly retainer that keeps it current as the catalogue changes and measures whether assistants name you more often. ## Book the week your shop becomes readable If you already know where agents drop you, the sprint fixes it in a week. If you do not, the Buyer Journey Check tells you which parts of the sprint to run first. [Reserve a sprint slot](https://autobasket.store/contact#sprint) [Audit before you build](https://autobasket.store/services/buyer-journey-check) --- 1. [Home](https://autobasket.store/) 2. Blog # Agentic commerce blog, newest first This is the dated feed of everything we publish, sorted by the day each article was written or last updated, so you can see what changed since your last visit. Protocols and channels move every few months, and each entry names its sources. For the library by topic, go to the [agentic commerce guides](https://autobasket.store/learn). Guides Use Cases Glossary Blog Learn Topic 13 articles - [Use cases Updated 16 October 2026 5 min read A chocolatier AI agents could recommend but not buy from July 2026 audit: a chocolatier near Geneva was named in 58% of Google AI Overviews, yet none of 26 product pages declared a price an agent could read.](https://autobasket.store/use-cases/chocolatier-findable-not-buyable) - [Discover Updated 16 October 2026 4 min read Do AI shopping agents need a GTIN to compare your product? Do you need a GTIN for AI shopping agents? Merchant Center lists it as recommended, OpenAI's feed as optional, and agents use it to match products.](https://autobasket.store/learn/why-agents-need-a-gtin) - [Discover 16 October 2026 4 min read GEO vs SEO, what changes for an online shop? GEO vs SEO for online shops: both start with crawling and indexing, 76.1% of AI Overview citations rank in the top 10, and what AI answers add.](https://autobasket.store/learn/geo-vs-seo-for-online-shops) - [Use cases Updated 16 October 2026 3 min read How an AI purchasing agent reorders office supplies under a standing mandate Worked example: a Zurich office's AI purchasing agent reorders paper and coffee under CHF 400, and what a supplier must expose, up to AP2 mandate checks.](https://autobasket.store/use-cases/office-supplies-reorder-mandate) - [Use cases Updated 16 October 2026 4 min read How an AI shopping agent compares gravel bikes under CHF 4,000 How AI shopping agents compare products, in a worked example: the bike shop whose price, stock and delivery are structured fields beats a cheaper rival.](https://autobasket.store/use-cases/gravel-bike-under-chf-4000) - [Discover Updated 16 October 2026 4 min read How do you get your products recommended by ChatGPT? How to get your products recommended by ChatGPT: what OpenAI says it ranks on, how product feeds reach it, and what the GEO research measured.](https://autobasket.store/learn/how-ai-assistants-choose-products) - [Foundations 16 October 2026 5 min read How does agentic commerce work for an online shop in Switzerland? Agentic commerce in Switzerland: CHF prices with VAT in Offer data, TWINT at checkout, Swiss delivery as shippingDetails, and our audit of 240 shops.](https://autobasket.store/learn/agentic-commerce-switzerland) - [Discover Updated 16 October 2026 3 min read How does ProductGroup schema declare sizes and variants for agents? How to mark up sizes and colours with ProductGroup, hasVariant and variesBy, with a JSON-LD example and the six variesBy values Google supports.](https://autobasket.store/learn/sizes-and-variants-for-agents) - [Use cases 16 October 2026 8 min read How ready are Swiss online shops for AI shopping agents? Are Swiss online shops ready for AI agents? Our July 2026 audit of 240 shops in French-speaking Switzerland found none scoring an A and a median of 57.](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness) - [Convert Updated 16 October 2026 5 min read UCP vs ACP vs AP2 vs MCP, which protocol does what in a purchase? UCP vs ACP, AP2 and MCP compared for merchants, with owner, launch date, status and who needs each, plus where x402 and A2A fit for a Swiss shop.](https://autobasket.store/learn/agent-commerce-protocols-explained) - [Discover Updated 16 October 2026 4 min read What does llms.txt do for ecommerce, next to Markdown and JSON-LD? llms.txt for ecommerce explained: Google says AI Overviews ignore it, who does read it, and why JSON-LD product data comes first for shopping agents.](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema) - [Foundations Updated 16 October 2026 5 min read What is agentic commerce, and how is it different from ecommerce? What is agentic commerce? A plain definition, a comparison with ecommerce, named examples from ChatGPT, Google and Perplexity, and what a shop fixes first.](https://autobasket.store/learn/what-is-agentic-commerce) - [Discover Updated 16 October 2026 3 min read What is share of model, and how do you measure it? What share of model means, the two ways to calculate it, how many prompt runs a reliable rate needs, and why every AI visibility figure needs an interval.](https://autobasket.store/learn/share-of-model-explained) No article matches that search yet. Try a broader word, or [ask us the question directly](https://autobasket.store/contact#session). --- 1. [Home](https://autobasket.store/) 2. Guides # Agentic commerce guides for online shops These guides explain, from the merchant's side, how AI assistants find, compare and buy products, and what a shop changes so they can. Each one answers its question in the first lines and names its sources. Foundations covers the basics, Discover covers being named in AI answers, and Convert covers how agents order and pay. Guides Use Cases Glossary Blog Learn ## Where should a merchant new to agentic commerce start? Read the definition first, then what it means for a Swiss shop, then how assistants pick products. The technical guides on protocols, llms.txt, GTINs and variants come after, and measurement last. 1. [What is agentic commerce, and how is it different from ecommerce? The definition, a comparison with ecommerce and the five stages of an agent purchase.](https://autobasket.store/learn/what-is-agentic-commerce) 2. [How does agentic commerce work for an online shop in Switzerland? What is live in Switzerland today, and how CHF, VAT, TWINT and delivery reach an agent.](https://autobasket.store/learn/agentic-commerce-switzerland) 3. [How do you get your products recommended by ChatGPT? What OpenAI says ChatGPT ranks products on, and the order of work for a shop.](https://autobasket.store/learn/how-ai-assistants-choose-products) 4. [GEO vs SEO, what changes for an online shop? Which SEO work still counts for AI answers, and what assistants reward on top of it.](https://autobasket.store/learn/geo-vs-seo-for-online-shops) 5. [UCP vs ACP vs AP2 vs MCP, which protocol does what in a purchase? Owner, launch date and status of each buying protocol, and which ones your channels need.](https://autobasket.store/learn/agent-commerce-protocols-explained) 6. [What does llms.txt do for ecommerce, next to Markdown and JSON-LD? What Google says about llms.txt, who reads it, and why JSON-LD comes first.](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema) 7. [Do AI shopping agents need a GTIN to compare your product? What Merchant Center and the ChatGPT feed ask for, and what to do without a GTIN.](https://autobasket.store/learn/why-agents-need-a-gtin) 8. [How does ProductGroup schema declare sizes and variants for agents? A JSON-LD example for sizes and colours, with the properties Google supports.](https://autobasket.store/learn/sizes-and-variants-for-agents) 9. [What is share of model, and how do you measure it? How to count how often assistants name you, and how many runs a reliable rate needs.](https://autobasket.store/learn/share-of-model-explained) ## Foundations What agentic commerce is and how an agent purchase unfolds. - [How does agentic commerce work for an online shop in Switzerland? Agentic commerce in Switzerland: CHF prices with VAT in Offer data, TWINT at checkout, Swiss delivery as shippingDetails, and our audit of 240 shops.](https://autobasket.store/learn/agentic-commerce-switzerland) - [What is agentic commerce, and how is it different from ecommerce? What is agentic commerce? A plain definition, a comparison with ecommerce, named examples from ChatGPT, Google and Perplexity, and what a shop fixes first.](https://autobasket.store/learn/what-is-agentic-commerce) ## Discover How assistants find, judge and cite pages, and what a shop can publish for them. - [GEO vs SEO, what changes for an online shop? GEO vs SEO for online shops: both start with crawling and indexing, 76.1% of AI Overview citations rank in the top 10, and what AI answers add.](https://autobasket.store/learn/geo-vs-seo-for-online-shops) - [How do you get your products recommended by ChatGPT? How to get your products recommended by ChatGPT: what OpenAI says it ranks on, how product feeds reach it, and what the GEO research measured.](https://autobasket.store/learn/how-ai-assistants-choose-products) - [How does ProductGroup schema declare sizes and variants for agents? How to mark up sizes and colours with ProductGroup, hasVariant and variesBy, with a JSON-LD example and the six variesBy values Google supports.](https://autobasket.store/learn/sizes-and-variants-for-agents) - [What does llms.txt do for ecommerce, next to Markdown and JSON-LD? llms.txt for ecommerce explained: Google says AI Overviews ignore it, who does read it, and why JSON-LD product data comes first for shopping agents.](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema) - [What is share of model, and how do you measure it? What share of model means, the two ways to calculate it, how many prompt runs a reliable rate needs, and why every AI visibility figure needs an interval.](https://autobasket.store/learn/share-of-model-explained) - [Do AI shopping agents need a GTIN to compare your product? Do you need a GTIN for AI shopping agents? Merchant Center lists it as recommended, OpenAI's feed as optional, and agents use it to match products.](https://autobasket.store/learn/why-agents-need-a-gtin) ## Convert How agents place orders and pay, and what a merchant has to set up so they can. - [UCP vs ACP vs AP2 vs MCP, which protocol does what in a purchase? UCP vs ACP, AP2 and MCP compared for merchants, with owner, launch date, status and who needs each, plus where x402 and A2A fit for a Swiss shop.](https://autobasket.store/learn/agent-commerce-protocols-explained) ## Where can you see a buying mission from start to finish? The use cases trace one buying mission or one audit end to end, field by field, from the shopper's request to the checkout. Each one says whether its data was measured on a real shop or constructed to show the mechanics, so you know how far to trust every figure in it. [See all use cases](https://autobasket.store/use-cases), look up a term in the [glossary](https://autobasket.store/glossary) or read every article by date on the [agentic commerce blog](https://autobasket.store/blog). - [Audit, Online retail, French-speaking Switzerland How ready are Swiss online shops for AI shopping agents?](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness) - [Worked example, Sports retail, Switzerland How an AI shopping agent compares gravel bikes under CHF 4,000](https://autobasket.store/use-cases/gravel-bike-under-chf-4000) - [Worked example, B2B supplies, DACH How an AI purchasing agent reorders office supplies under a standing mandate](https://autobasket.store/use-cases/office-supplies-reorder-mandate) - [Audit, Food and confectionery, Switzerland A chocolatier AI agents could recommend but not buy from](https://autobasket.store/use-cases/chocolatier-findable-not-buyable) ## Test what you just read on your own shop The free readiness check applies the same rules as these guides to one of your product pages and lists what to fix, in under a minute. [Run the free check](https://autobasket.store/#readiness-scan) --- 1. [Home](https://autobasket.store/) 2. [Guides](https://autobasket.store/learn) 3. UCP vs ACP vs AP2 vs MCP, which protocol does what in a purchase? # UCP vs ACP vs AP2 vs MCP, which protocol does what in a purchase? The four protocols stack. MCP lets an agent call a shop's tools, UCP and ACP define how it places an order, and AP2 proves what the person authorised it to pay. UCP launched on 11 January 2026 with Google AI Mode, ACP in September 2025 with ChatGPT. A merchant needs only those used by channels open in its market. Guide, Convert, published 12 October 2026, updated 16 October 2026 ## What is the difference between UCP and ACP? Both define how an agent places an order with a merchant, and they come from different camps. ACP was published by OpenAI and Stripe for purchases inside ChatGPT. UCP was launched by Google with Shopify, Etsy, Wayfair, Target and Walmart, for any agent and any commerce stack, starting with AI Mode and the Gemini app. | Protocol | Owner | Launched | What it does | Who needs it | Status in October 2026 | | --- | --- | --- | --- | --- | --- | | Model Context Protocol (MCP) | Anthropic, stewarded by the Agentic AI Foundation under the Linux Foundation since 9 December 2025 | 25 November 2024 | Lets an agent discover and call a shop’s tools, such as catalogue search or stock lookup | Shops that want agents to query the catalogue directly instead of scraping pages | Supported by ChatGPT, Claude, Gemini, Microsoft Copilot and VS Code | | Universal Commerce Protocol (UCP) | Google, co-developed with Shopify, Etsy, Wayfair, Target and Walmart | 11 January 2026 | Standard checkout an agent can complete with any merchant, over REST, MCP, AP2 or A2A | Shops whose buyers use Google AI Mode or Gemini | Checkout live for eligible US retailers first | | Agentic Commerce Protocol (ACP) | OpenAI and Stripe, open source under Apache 2.0 | 29 September 2025 | Lets ChatGPT pass an order and a delegated payment token to the merchant, who stays merchant of record | Shops with a large ChatGPT audience | Instant Checkout moved into apps inside ChatGPT in March 2026 | | Agent Payments Protocol (AP2) | Google, donated to the FIDO Alliance on 28 April 2026 | 16 September 2025 | Signed mandates that prove what the person authorised the agent to buy | Shops that accept payments from agents, above all recurring or unattended ones | Version 0.2 released 28 April 2026 | | x402 | Coinbase, with an A2A x402 extension launched alongside AP2 | Extension announced 16 September 2025 | Payment per request in stablecoins over the HTTP 402 status code | APIs and data sellers charging per call | Niche for retail goods | A2A, the Agent2Agent protocol, is the messaging layer between agents. UCP and AP2 both state compatibility with it, so a merchant rarely implements it directly. ## Do I need both UCP and ACP? Usually not at the same time. A shop needs the protocol of each channel where its buyers actually ask, so the choice follows the audience, and both protocols read the same product data underneath. A Shopify merchant gets much of UCP from the platform. A merchant on Salesforce Commerce Cloud, SAP Commerce or a custom stack has to implement it. ACP matters if ChatGPT sends real buyers: in March 2026 an OpenAI spokesperson said Instant Checkout was moving to apps, with ACP “serving as the infrastructure that connects users to merchants” ([Digital Commerce 360, 6 March 2026](https://www.digitalcommerce360.com/2026/03/06/openai-shifts-checkout-plans-agentic-commerce-strategy/)). Both launched in the United States, so a Swiss shop prepares the data now and the protocol when the channel opens here. We build that layer as part of [agentic checkout with Autobasket Convert](https://autobasket.store/products/autobasket-convert). ## What is AP2, and what is a mandate? AP2 is the proof layer for payments made by software. A mandate is a cryptographically signed record of what the person approved, which the merchant and the payment network can verify before money moves. At launch Google described two mandates ([Google Cloud, 16 September 2025](https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol)). An intent mandate records the instruction and its limits, for example “reorder printer paper every month, up to CHF 400”. A cart mandate binds approval to one basket and one amount. More than 60 organisations backed the launch, among them Adyen, American Express, Coinbase, Etsy, Mastercard, PayPal, Revolut, Salesforce and Worldpay. Version 0.2, released when Google donated AP2 to the FIDO Alliance on 28 April 2026, adds Human Not Present payments, where the agent completes the purchase later under a mandate signed earlier. Verifiable Intent, co-developed with Mastercard, went to FIDO at the same time. The spec is still moving, so check the current mandate names on [ap2-protocol.org](https://ap2-protocol.org/) before you build. The [office reorder use case](https://autobasket.store/use-cases/office-supplies-reorder-mandate) shows a mandate check step by step. Human Not Present payments matter most in B2B, where a buyer’s procurement agent reorders on a schedule. What a supplier has to expose for that, from contract prices to stock without a login, is set out in [agentic commerce for B2B suppliers](https://autobasket.store/solutions/b2b-wholesale). ## How is MCP different from A2A? MCP connects an agent to tools, and A2A connects an agent to other agents. For a shop, MCP is the one that matters: a server that exposes a handful of typed tools, such as search the catalogue under a maximum price or return stock for a SKU in a postcode. At its handover to the Agentic AI Foundation, MCP counted more than 97 million monthly SDK downloads and about 10,000 active servers ([MCP blog, 9 December 2025](https://blog.modelcontextprotocol.io/posts/2025-12-09-mcp-joins-agentic-ai-foundation/)). The merchant decides what is exposed and at what rate, and the agent no longer has to parse the product page. An MCP server does not replace structured data on the page, since assistants that search the open web still read the page. ## Which protocols does a Swiss merchant actually need? Fewer than the table suggests, and in a fixed order. First complete structured data, because every protocol reads the same product, price and stock. Then the protocol of the channel your buyers use, and AP2 checks before you accept orders placed without the customer present. | Situation | What to build first | | --- | --- | | Customers ask Google, then buy on your site | Structured data now, UCP when AI Mode checkout opens in Switzerland | | Large ChatGPT audience | Structured data and an OpenAI product feed, ACP through a ChatGPT app | | B2B customers reorder under standing rules | AP2 mandate checks and an MCP server for catalogue and stock | | Most payments run through TWINT | Check with your payment provider, since we found no TWINT support in the published protocol documents | A protocol built for a channel that is not open in your market costs money without bringing orders. The [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) does the data step, and the [Switzerland guide](https://autobasket.store/learn/agentic-commerce-switzerland) covers TWINT and CHF. ## Where to read next - [Agentic checkout with Autobasket Convert](https://autobasket.store/products/autobasket-convert)How we build and maintain the UCP, ACP and AP2 layer your channels need. - [An AI purchasing agent reorders under a mandate](https://autobasket.store/use-cases/office-supplies-reorder-mandate)The five protocol steps of a recurring B2B order, with the cart total checked against the mandate. - [Agentic commerce in Switzerland](https://autobasket.store/learn/agentic-commerce-switzerland)What the protocols mean for a shop paid in CHF and mostly through TWINT. - [Agentic commerce glossary](https://autobasket.store/glossary#ap2)One-line definitions of mandates, Human Not Present and x402. ## Sources 1. [Anthropic, Introducing the Model Context Protocol, 25 November 2024](https://www.anthropic.com/news/model-context-protocol) 2. [Model Context Protocol, joins the Agentic AI Foundation, 9 December 2025](https://blog.modelcontextprotocol.io/posts/2025-12-09-mcp-joins-agentic-ai-foundation/) 3. [Google, NRF 2026 remarks on the Universal Commerce Protocol, 11 January 2026](https://blog.google/company-news/inside-google/message-ceo/nrf-2026-remarks/) 4. [Agentic Commerce Protocol, agenticcommerce.dev](https://www.agenticcommerce.dev/) 5. [OpenAI, Buy it in ChatGPT, 29 September 2025](https://openai.com/index/buy-it-in-chatgpt/) 6. [Digital Commerce 360, OpenAI shifts checkout plans, 6 March 2026](https://www.digitalcommerce360.com/2026/03/06/openai-shifts-checkout-plans-agentic-commerce-strategy/) 7. [Google Cloud, Announcing the Agent Payments Protocol, 16 September 2025](https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol) 8. [Google, Donating AP2 to the FIDO Alliance, 28 April 2026](https://blog.google/products-and-platforms/platforms/google-pay/agent-payments-protocol-fido-alliance/) ## See how your own product pages score Bring one of your product pages to a first conversation and we read it with you the way an agent does, which is the fastest way to turn what you just read into a ticket for your developers. [Start selling smarter](https://autobasket.store/contact#session) [Read one product page first](https://autobasket.store/#readiness-scan) --- 1. [Home](https://autobasket.store/) 2. [Guides](https://autobasket.store/learn) 3. How does agentic commerce work for an online shop in Switzerland? # How does agentic commerce work for an online shop in Switzerland? Agentic commerce reaches Swiss shops through recommendations first, since agent checkout launched in the United States. 5% of DACH online shoppers had used an AI agent by June 2026. A Swiss shop prepares by stating CHF prices with VAT, Swiss delivery and stock as structured data. None of the 240 Romandie shops we audited in July 2026 scored an A. Guide, Foundations, published 16 October 2026 ## Is agentic commerce available in Switzerland yet? Recommendations are, and agent checkout is not yet, in the main channels. ChatGPT, Gemini, Perplexity and Google AI Overviews already name and compare Swiss shops, while UCP checkout in Google AI Mode launched for eligible US retailers and ChatGPT’s Instant Checkout started with US sellers. Swiss shoppers are early. 5% of online shoppers in Switzerland, Germany and Austria had used an AI agent to help them shop, and about 18% planned to, in a University of St.Gallen survey of just under 3,000 people ([HSG, 24 June 2026](https://www.unisg.ch/en/newsdetail/news/ki-agenten-im-onlinehandel-warum-konsumentinnen-und-konsumenten-beim-einkauf-noch-zoegern/)). The market they would act on is sizeable: Swiss consumers spent CHF 15.8 billion online in 2025, CHF 13 billion of it at Swiss shops ([HANDELSVERBAND.swiss, 12 March 2026](https://handelsverband.swiss/news/schweizer-online-konsum-waechst-im-jahr-2025-erneut-um-6-das-wachstum-der-einkaeufe-im-ausland-flacht-ab/)). So the work for a Swiss shop now is being read correctly at the recommendation stage, ahead of checkout opening here. ## Are Swiss online shops ready for AI shopping agents? Most are not yet. None of the 240 online shops we audited in French-speaking Switzerland in July 2026 scored an A for agent readiness. The median score was 57 out of 100, and 104 shops scored D or F ([Romandie audit](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness)). The weak point is the product itself. For 157 of 211 shops read in full (74%), the stage that lost the most agents was reading price, stock and identity, and on 117 of them (55%) the only price was in the page text, with no `Offer` price in the structured data. An agent comparing three shops on price cannot include one whose price it cannot read as a field. ## How do I show CHF prices with VAT to AI agents? Declare the price as a number with `priceCurrency` set to `CHF`, and state that VAT is included with a `UnitPriceSpecification` and `valueAddedTaxIncluded`. Prices shown to Swiss consumers include VAT, and an agent comparing a Swiss shop with a foreign one needs to know which kind of figure it is reading. The example below is for an invented shop. Replace its price and delivery days with the values you actually commit to, rendered from the same record as the visible price: ``` { "@context": "https://schema.org", "@type": "Product", "name": "Gravel bike GR 2, size M", "gtin13": "7610000000031", "brand": { "@type": "Brand", "name": "Example Cycles" }, "offers": { "@type": "Offer", "price": "3890.00", "priceCurrency": "CHF", "priceSpecification": { "@type": "UnitPriceSpecification", "price": "3890.00", "priceCurrency": "CHF", "valueAddedTaxIncluded": true }, "availability": "https://schema.org/InStock", "shippingDetails": { "@type": "OfferShippingDetails", "shippingDestination": { "@type": "DefinedRegion", "addressCountry": "CH" }, "shippingRate": { "@type": "MonetaryAmount", "value": "0.00", "currency": "CHF" }, "deliveryTime": { "@type": "ShippingDeliveryTime", "handlingTime": { "@type": "QuantitativeValue", "minValue": 0, "maxValue": 1, "unitCode": "DAY" }, "transitTime": { "@type": "QuantitativeValue", "minValue": 1, "maxValue": 2, "unitCode": "DAY" } } } } } ``` ## How do I tell agents that I deliver with Swiss Post? Declare delivery as `shippingDetails` with a Swiss destination, a rate and a delivery time in days, as in the example above. Agents compare delivery as a number, so “delivered by Swiss Post in one to two working days” in a footer or an image does not enter the comparison. If you deliver abroad, add one `OfferShippingDetails` per destination country with its own rate and transit time. The [gravel bike use case](https://autobasket.store/use-cases/gravel-bike-under-chf-4000) shows a shortlist decided on structured Swiss stock and delivery. ## Can AI agents pay with TWINT? Not through the published agent protocols yet, as far as we could find. ACP launched with Stripe’s Shared Payment Token, AP2 with card networks and wallets among its partners, and none of the protocol documents and announcements we read for this guide names TWINT. That gap matters in Switzerland. Mobile payment accounts for 74.4% of domestic distance-selling transactions, and TWINT holds 67.0% of those ([Swiss Payment Monitor 1/2026](https://fsi.unisg.ch/fileadmin/user_upload/HSG_ROOT/Institut_FSI/Dokumente/Aktuelle_Forschung/Report_Swiss_Payment_Monitor_2026-1_final_EN.pdf)), and TWINT reports that more than 84% of Swiss online shops accept it ([TWINT](https://www.twint.ch/en/about-us/)). Until an agent channel supports it, an agent-referred buyer finishing on your own checkout still pays with TWINT, so keep it there and ask your payment provider when it reaches agent checkouts. ## Do Swiss shoppers trust an AI agent to pay for them? Most do not want it to pay alone. Over 80% of people in Switzerland reject a fully automated AI payment without confirmation, while nearly two thirds can imagine an assistant recommending how to pay ([Swiss Payment Monitor 2/2026](https://alexandria.unisg.ch/entities/publication/c1fa42b2-d39d-4420-9ee4-dfe284cbfc14)). For a consumer shop, that points to the agent preparing the order and the person confirming it. For B2B, a standing mandate approved once by the buyer fits better, as the [office reorder use case](https://autobasket.store/use-cases/office-supplies-reorder-mandate) shows. ## How does a Swiss shop compete with Galaxus when agents compare? On data it controls and Galaxus cannot copy: complete identifiers, Swiss stock, delivery times and service terms declared as fields. Galaxus reported CHF 3.764 billion in platform sales for 2025, up 17%, with 10 million products on sale ([Digitec Galaxus, 7 January 2026](https://www.digitec.ch/en/page/galaxus-grows-by-17-per-cent-in-2025-40855)), and 13 of the 30 largest Swiss B2C online shops are marketplaces (HANDELSVERBAND.swiss). When an agent compares the same GTIN across shops, OpenAI says merchants are ranked on availability, price, quality and whether they are the maker or primary seller ([OpenAI Help Center](https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search)). A smaller shop with faster delivery or a better return policy wins only if those facts are machine-readable. The [multi-brand retailers](https://autobasket.store/solutions/multi-brand-retailers) page goes further. ## What does a Swiss shop need that a generic guide misses? | Swiss specific | What the agent needs | Field | | --- | --- | --- | | Prices in CHF, shown VAT-inclusive | A numeric price, its currency and whether tax is included | `price`, `priceCurrency`, `valueAddedTaxIncluded` | | Swiss Post or courier delivery | Destination, rate and delivery days as numbers | `shippingDetails` with `addressCountry: CH` | | Stock held in Switzerland | Availability per product and variant | `availability` on each `Offer` | | French, German and Italian pages | The same product identified in every language | `gtin13`, `sku`, one `@id` per product | | TWINT as the main mobile method | A checkout the agent can hand to the person | Your own checkout, until agent channels support TWINT | | Galaxus listing the same product | A match on the same identifier, then a better offer | `gtin13`, `brand`, `mpn` | The industry’s direction is set: at the W3C and GS1 workshop on e-commerce for humans and AI agents in Zurich on 8 and 9 September 2026, participants converged on the GTIN as the primary identifier for physical goods ([Forkast, 10 September 2026](https://forkast.news/what-agent-commerce-needs-from-product-data-lessons-from-the-w3c-gs1-workshop/)). The [GTIN guide](https://autobasket.store/learn/why-agents-need-a-gtin) covers identifiers in detail. ## What should a Swiss shop do first? Fix the product data, which pays off in every assistant today, and leave checkout protocols until a channel opens in Switzerland. In order: let shopping crawlers in, declare CHF price with VAT, stock and Swiss delivery on every product, add GTINs, then measure how often assistants name you. The [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) runs that diagnosis on your catalogue, the [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) fixes the data in seven days, and [agentic checkout with Autobasket Convert](https://autobasket.store/products/autobasket-convert) prepares the protocol layer for when Swiss checkout opens. ## Where to read next - [240 Swiss online shops audited for AI agents](https://autobasket.store/use-cases/romandie-online-shops-agent-readiness)The full Romandie data release behind the figures on this page, by sector and by stage. - [Agent-ready online shop in seven days](https://autobasket.store/services/machine-readability-sprint)CHF, VAT, delivery and stock declared as structured data on your existing platform. - [Agentic commerce for retailers](https://autobasket.store/solutions/multi-brand-retailers)How a Swiss multi-brand shop competes with Galaxus when agents compare prices. - [UCP vs ACP vs AP2 vs MCP](https://autobasket.store/learn/agent-commerce-protocols-explained)The checkout and payment protocols, their owners and where each is live. ## Sources 1. [University of St.Gallen, AI agents in online retail, 24 June 2026](https://www.unisg.ch/en/newsdetail/news/ki-agenten-im-onlinehandel-warum-konsumentinnen-und-konsumenten-beim-einkauf-noch-zoegern/) 2. [HANDELSVERBAND.swiss with NIQ/GfK and Swiss Post, Swiss online consumption 2025, 12 March 2026](https://handelsverband.swiss/news/schweizer-online-konsum-waechst-im-jahr-2025-erneut-um-6-das-wachstum-der-einkaeufe-im-ausland-flacht-ab/) 3. [Swiss Payment Monitor 1/2026, University of St.Gallen and ZHAW, February 2026](https://fsi.unisg.ch/fileadmin/user_upload/HSG_ROOT/Institut_FSI/Dokumente/Aktuelle_Forschung/Report_Swiss_Payment_Monitor_2026-1_final_EN.pdf) 4. [Swiss Payment Monitor 2/2026, University of St.Gallen and ZHAW, August 2026](https://alexandria.unisg.ch/entities/publication/c1fa42b2-d39d-4420-9ee4-dfe284cbfc14) 5. [TWINT, About us, fetched 16 October 2026](https://www.twint.ch/en/about-us/) 6. [Digitec Galaxus, Galaxus grows by 17 per cent in 2025, 7 January 2026](https://www.digitec.ch/en/page/galaxus-grows-by-17-per-cent-in-2025-40855) 7. [Forkast, lessons from the W3C and GS1 workshop in Zurich, 10 September 2026](https://forkast.news/what-agent-commerce-needs-from-product-data-lessons-from-the-w3c-gs1-workshop/) 8. [Google, the Universal Commerce Protocol, 11 January 2026](https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/) 9. [OpenAI Help Center, Shopping with ChatGPT Search](https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search) ## See how your own product pages score Bring one of your product pages to a first conversation and we read it with you the way an agent does, which is the fastest way to turn what you just read into a ticket for your developers. [Start selling smarter](https://autobasket.store/contact#session) [Read one product page first](https://autobasket.store/#readiness-scan) --- 1. [Home](https://autobasket.store/) 2. [Guides](https://autobasket.store/learn) 3. GEO vs SEO, what changes for an online shop? # GEO vs SEO, what changes for an online shop? GEO builds on SEO. Google's AI answers use the same index, and Ahrefs found 76.1% of pages cited in AI Overviews also rank in the top 10. What changes is the unit: answers quote passages, favour third-party sources, and reward sourced numbers, while keyword stuffing scored below baseline in the GEO research. Guide, Discover, published 16 October 2026 ## Is SEO dead with AI search? No. Google’s AI Overviews and AI Mode pick their supporting links from pages that are indexed and eligible to show a snippet in Search, and Google says there are “no additional requirements” to appear in them ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)). The overlap is measurable. Ahrefs analysed 1.9 million citations from 1 million AI Overviews and found that 76.1% of cited pages ranked in the top 10 organic results for the query ([Ahrefs, updated 2 July 2026](https://ahrefs.com/blog/search-rankings-ai-citations/)). That figure covers Google. ChatGPT and Perplexity run their own retrieval, so a top-10 ranking helps there less directly, but crawlable pages and clear topics still gate them. ## What is the same in GEO and SEO? The foundations: a crawler must be allowed in, the content must be in the HTML, the page must be indexed, and it must be about one clear topic. Generative engine optimization, GEO, adds nothing that replaces these, because an assistant cannot quote a page it never retrieved. For a shop that means the SEO backlog still comes first: crawl rules, server-rendered product content, one product or one question per URL, internal links with descriptive anchors, and structured data that matches the visible text, which Google lists among its fundamentals for AI features. ## What is different in GEO for an online shop? Three things change. Answers quote passages rather than ranking whole pages, they lean on third-party sources more than search results do, and they reward evidence inside the passage. AI search engines show “a systematic and overwhelming bias towards Earned media”, meaning third-party authoritative sources, over brand-owned content, unlike Google’s more balanced mix ([Chen et al., arXiv, September 2025](https://arxiv.org/abs/2509.08919)). In the foundational GEO study, adding cited sources, quotations and statistics raised a source’s visibility in generative answers by 30 to 40%, and by 115.1% for a source ranked fifth ([Aggarwal et al., KDD 2024](https://arxiv.org/abs/2311.09735)). Keyword stuffing went the other way: on GEO-bench it scored 17.8 against a 19.5 baseline on the paper’s position-adjusted word count. The boundary is in the 2026 survey by Olivier Martinez, which reviewed 45 GEO studies and found no technique with a stable, cross-platform effect on discoverability or traffic, and that citation-aimed rewrites can reduce retrieval when they dilute the topic ([arXiv 2607.14035](https://arxiv.org/abs/2607.14035)). The GEO gains are real for passages already being read, and unproven beyond that. | | SEO | GEO | | --- | --- | --- | | Unit that competes | The page, for a ranking position | The passage, for a citation or a product mention | | Main input | Query match, links, page quality | Retrieved passages, structured product data, third-party mentions | | What helps | Relevance, internal links, earned links | The same, plus the answer in the first sentences, sourced numbers and quotations | | What hurts | Thin or duplicate pages | The same, plus keyword stuffing and diluted topics | | Outside sources | Backlinks as a ranking signal | Reviews, guides and press quoted directly in the answer | | Success metric | Rankings, clicks, organic revenue | How often assistants name you, and orders arriving from agents | | Measurement | Stable enough to read weekly | Varies from run to run, so repeated prompts and intervals | ## Does Google need special files or markup for AI Overviews? No. Google states that you do not need to create new machine-readable files, AI text files or markup to appear in AI Overviews or AI Mode. An `llms.txt` file is ignored by Google Search, and structured data is recommended for SEO rather than as an AI requirement. Shopping agents and product feeds are a different case, because they compare offers on fields. The [llms.txt guide](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema) separates what Google needs from what agents read. ## What should a shop do first? Fix the SEO foundations, then make each product and guide quotable, then work on the sources that name you. In that order, because each step depends on the one before. - Crawl, render and index: product content in the HTML, crawlers allowed, one topic per URL. - Product data as fields: price, stock, GTIN and delivery in JSON-LD and the product feed, through the [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint). - Answer first: each section opens with the answer in one or two sentences, with a number and its source where one exists. - Earned mentions: reviews, buying guides and press in your market, which assistants cite more than your own pages. - Measure across repeated prompts per assistant, as the [share of model guide](https://autobasket.store/learn/share-of-model-explained) explains. The [Buyer Journey Check](https://autobasket.store/services/buyer-journey-check) shows where your shop stands on each step, and [Autobasket Discover](https://autobasket.store/products/autobasket-discover) runs the work and the measurement as a retainer. ## Where to read next - [GEO for ecommerce with Autobasket Discover](https://autobasket.store/products/autobasket-discover)The retainer that measures how often assistants name your products and fixes what drops them. - [How to get your products recommended by ChatGPT](https://autobasket.store/learn/how-ai-assistants-choose-products)What OpenAI says ChatGPT ranks products on, and the order of work for a shop. - [llms.txt for ecommerce](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema)Why Google needs no special files, and which machine-readable layers still matter for agents. - [AI shopping agents for consumer electronics](https://autobasket.store/solutions/consumer-electronics)A category where spec tables and identifiers decide AI comparisons. ## Sources 1. [Ahrefs, 76% of AI Overview citations pull from the top 10, updated 2 July 2026](https://ahrefs.com/blog/search-rankings-ai-citations/) 2. [Google Search Central, AI features and your website, updated 10 December 2025](https://developers.google.com/search/docs/appearance/ai-features) 3. [Aggarwal et al., GEO, Generative Engine Optimization, KDD 2024](https://arxiv.org/abs/2311.09735) 4. [Chen et al., Generative Engine Optimization, How to Dominate AI Search, arXiv 2509.08919](https://arxiv.org/abs/2509.08919) 5. [Martinez, A Critical Survey of Generative Engine Optimization (2023 to 2026), arXiv 2607.14035](https://arxiv.org/abs/2607.14035) ## See how your own product pages score Bring one of your product pages to a first conversation and we read it with you the way an agent does, which is the fastest way to turn what you just read into a ticket for your developers. [Start selling smarter](https://autobasket.store/contact#session) [Read one product page first](https://autobasket.store/#readiness-scan) --- 1. [Home](https://autobasket.store/) 2. [Guides](https://autobasket.store/learn) 3. How do you get your products recommended by ChatGPT? # How do you get your products recommended by ChatGPT? ChatGPT picks products from structured metadata such as price and description, plus third-party content, and OpenAI says the results are not ads. Among merchants selling the same item it ranks on availability, price, quality and whether the shop is the maker. Being readable, cited elsewhere and stocked does more than any rewrite of the copy. Guide, Discover, published 12 October 2026, updated 16 October 2026 ## How does ChatGPT decide which products to recommend? ChatGPT combines structured metadata from first-party and third-party providers, such as price and product description, with other third-party content and the model’s own knowledge. Among merchants selling the same item, it ranks on availability, price, quality and whether the merchant is the maker or primary seller ([OpenAI Help Center](https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search)). The request shapes the weighting. OpenAI’s example: “If you specify a budget of $30, ChatGPT will focus more on price.” A shopper’s constraint becomes a filter, and a product whose price or stock is not machine-readable cannot pass it. The [gravel bike use case](https://autobasket.store/use-cases/gravel-bike-under-chf-4000) shows a cheaper shop dropped for exactly that reason. So the first lever is data the assistant can read, and the second is what other sites say about you. ## Are ChatGPT product recommendations paid ads? No. OpenAI states that product results “are selected independently by ChatGPT and are not ads”. A merchant cannot buy a place in them, so visibility comes from the inputs above, which is why the work sits in product data and in third-party coverage. ## Does a product feed help with ChatGPT shopping? Yes, it is the most direct input a merchant controls. Shopify merchants are included through Shopify Catalog with no extra work, according to OpenAI, and other merchants can apply to send a direct product feed. The feed specification marks `brand`, `price` and `availability` as required, `gtin` as optional, and a group ID as conditionally required for variants ([OpenAI product feed spec](https://developers.openai.com/commerce/specs/feed)). A feed that disagrees with the page creates two prices, and the [GTIN guide](https://autobasket.store/learn/why-agents-need-a-gtin) compares its identifier rules with Google Merchant Center. On Google, AI Overviews and AI Mode draw on the same index plus Merchant Center data, and may run a “query fan-out”, several related searches across subtopics for one question ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)). ## Why does ChatGPT recommend my competitors and not my shop? Usually because the competitor is easier to read or is named by more third-party sources. AI search engines show “a systematic and overwhelming bias towards Earned media”, meaning third-party authoritative sources, over brand-owned content, according to [Chen and colleagues (arXiv, September 2025)](https://arxiv.org/abs/2509.08919). Our July 2026 audit shows both causes on one shop. A chocolatier near Geneva was named in 58% of the Google AI Overviews we tested, with guides, press and Reddit among the cited sources, yet none of its 26 product pages declared a price an agent could read ([chocolatier use case](https://autobasket.store/use-cases/chocolatier-findable-not-buyable)). Coverage got it named, and missing data stopped the sale. Check both: whether reviews, guides and press in your market name you, and whether your price, stock and identifiers are fields. When several shops sell the same product, the comparison moves to the offer itself: availability, price and delivery. That is the case for [agentic commerce for multi-brand retailers](https://autobasket.store/solutions/multi-brand-retailers), where the same GTIN is listed by many shops at once. ## Do reviews and ratings matter for AI recommendations? They count as part of the third-party content assistants read, and their effect differs by model. A study of AI shopping agents found that “sensitivities to price, ratings, and reviews vary sharply across models”, and that agents concentrate demand on a few products ([arXiv 2508.02630, August 2025](https://arxiv.org/abs/2508.02630)). Publish ratings in structured data where you have genuine reviews, and do not expect one rating to move every assistant the same way. ## What does the GEO research say actually helps? Adding cited sources, quotations and statistics raised a source’s visibility in generative engine answers by 30 to 40% in the foundational GEO study, and by 115.1% for a source ranked fifth in search, while the top-ranked source lost 30.3% on average ([Aggarwal et al., KDD 2024](https://arxiv.org/abs/2311.09735)). Keyword stuffing scored below the unoptimised baseline. That rank-five result is the useful one for a mid-market shop, since evidence on the page helps most when the page is not already the leader. The boundary comes from the 2026 survey by Olivier Martinez, which reviewed 45 studies and found no technique with a stable, cross-platform effect on discoverability or traffic, and that rewrites aimed at citation can reduce how often a page is retrieved when they dilute its topic ([arXiv 2607.14035, 15 July 2026](https://arxiv.org/abs/2607.14035)). The GEO gains were measured on passages already in the model’s context, so they say nothing about being retrieved in the first place. ## What should a shop do, in order? | Step | What it fixes | Where it is covered | | --- | --- | --- | | Let OpenAI’s and Google’s crawlers in, with product content in the HTML | Being retrieved at all | [llms.txt and schema guide](https://autobasket.store/learn/llms-txt-markdown-mirrors-schema) | | Declare price, stock, GTIN and delivery as structured data, and send a product feed | Passing the shopper’s constraints | [Machine Readability Sprint](https://autobasket.store/services/machine-readability-sprint) | | Keep one product or one question per page, with the answer in the first sentences | The passage that gets retrieved carries the answer | [GEO vs SEO guide](https://autobasket.store/learn/geo-vs-seo-for-online-shops) | | Add sourced numbers and named examples to guides and category pages | Being cited once read | This guide | | Earn mentions in reviews, guides and press | The earned media bias | [Autobasket Discover](https://autobasket.store/products/autobasket-discover) | | Measure across repeated prompts | Telling progress from noise | [share of model guide](https://autobasket.store/learn/share-of-model-explained) | Nobody, ourselves included, can promise a fixed place in an AI answer. What can be done is to remove the reasons a product is dropped and to measure whether the share of answers naming you moves over several months. ## Where to read next - [GEO for ecommerce with Autobasket Discover](https://autobasket.store/products/autobasket-discover)Monthly measurement of how often ChatGPT and Gemini name your products, and the fixes behind it. - [How to measure share of model](https://autobasket.store/learn/share-of-model-explained)The method for counting how often assistants name you, with confidence intervals. - [How an AI agent shortlists a gravel bike](https://autobasket.store/use-cases/gravel-bike-under-chf-4000)A budget request traced constraint by constraint, where the cheaper shop loses. - [Agentic commerce for brands](https://autobasket.store/solutions/brands-selling-direct)Who gets the recommendation when your product is also sold by retailers. ## Sources 1. [OpenAI Help Center, Shopping with ChatGPT Search, fetched 16 October 2026](https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search) 2. [OpenAI developers, product feed specification](https://developers.openai.com/commerce/specs/feed) 3. [Aggarwal et al., GEO, Generative Engine Optimization, KDD 2024](https://arxiv.org/abs/2311.09735) 4. [Martinez, A Critical Survey of Generative Engine Optimization (2023 to 2026), arXiv 2607.14035](https://arxiv.org/abs/2607.14035) 5. [Chen et al., Generative Engine Optimization, How to Dominate AI Search, arXiv 2509.08919](https://arxiv.org/abs/2509.08919) 6. [Google Search Central, AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) ## See how your own product pages score Bring one of your product pages to a first conversation and we read it with you the way an agent does, which is the fastest way to turn what you just read into a ticket for your developers. [Start selling smarter](https://autobasket.store/contact#session) [Read one product page first](https://autobasket.store/#readiness-scan) --- 1. [Home](https://autobasket.store/) 2. [Guides](https://autobasket.store/learn) 3. What does llms.txt do for ecommerce, next to Markdown and JSON-LD? # What does llms.txt do for ecommerce, next to Markdown and JSON-LD? An llms.txt file lists a site's pages for AI systems, a Markdown mirror gives them clean page text, and JSON-LD states product, price and stock as fields. Google said in July 2026 that Search ignores llms.txt and needs no special files for AI Overviews. For a shop, JSON-LD comes first, because agents compare offers on fields. Guide, Discover, published 12 October 2026, updated 16 October 2026 ## What is llms.txt? `llms.txt` is a Markdown file at the root of a site that lists the pages an AI system should read, with one line on each. Jeremy Howard proposed it on 3 September 2024 at [llmstxt.org](https://llmstxt.org/). The only required part is an H1 with the site name, followed by an optional summary and lists of links. A shop’s file can be this short, shown here for an invented bike shop: ``` # Velo Atelier > Bike shop in Lausanne. Road, gravel and e-bikes, delivered across Switzerland. ## Policies - [Delivery](https://example.ch/delivery.md): carriers, costs and delivery times in Switzerland - [Returns](https://example.ch/returns.md): 30-day returns, conditions and refunds ## Buying guides - [Gravel bike sizes](https://example.ch/guides/gravel-sizes.md): frame size by rider height ``` Ours is at [/llms.txt](https://autobasket.store/llms.txt), with the full text in [/llms-full.txt](https://autobasket.store/llms-full.txt). `llms-full.txt` is a community convention and not part of the proposal. ## Does Google use llms.txt? No. Google’s guide to generative AI in Search, updated on 10 July 2026, says you do not need machine-readable files, AI text files or Markdown to appear in Search, that it is “completely fine” to keep an llms.txt file, and that Search ignores it ([Google Search Central](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide)). Google’s page on AI features says the same for AI Overviews and AI Mode: “There are no additional requirements”, and a page must be indexed and eligible for a snippet ([Google Search Central, updated 10 December 2025](https://developers.google.com/search/docs/appearance/ai-features)). So llms.txt does nothing for Google rankings, and a shop should never buy it as a Google tactic. The [GEO vs SEO guide](https://autobasket.store/learn/geo-vs-seo-for-online-shops) covers what does count for Google’s AI answers. ## Who does read llms.txt? Tools and agents that fetch it on purpose. Chrome’s Lighthouse checks the file in its agentic browsing audits, and flags it only when the server returns an error; a missing file is marked not applicable ([Chrome for Developers, 5 May 2026](https://developer.chrome.com/docs/lighthouse/agentic-browsing/llms-txt)). Beyond that, an agent browsing on a user’s behalf can use the file as a table of contents if it chooses to request it. We have seen no assistant vendor document that its crawler reads it, and no published study measuring its effect on citations. That is why we treat it as cheap and optional: it costs an hour, keeps an inventory of what the site publishes, and helps the agents that look. ## What is the difference between llms.txt and robots.txt? `robots.txt` tells crawlers what they may fetch. `llms.txt` suggests what is worth reading. Only the first one controls access, and it is the one that decides whether a shopping assistant sees the shop at all. | | robots.txt | llms.txt | JSON-LD | | --- | --- | --- | --- | | What it says | Which crawlers may fetch which paths | Which pages matter, with a line on each | What the product is, its price, stock, delivery and returns | | Who reads it | Every well-behaved crawler, including GPTBot and OAI-SearchBot | Agents and tools that request it | Google, Bing, Merchant Center and AI crawlers | | Effect if missing | Everything is allowed | None for Google | The offer can drop out of agent comparisons | | Priority for a shop | Check first | Optional, last | First | A crawler blocked in `robots.txt` cannot fetch the product pages, whatever llms.txt lists, so check those rules before anything else. ## Which product schema fields do agents read? JSON-LD in a `