What is agentic commerce in simple terms?

Agentic commerce is a purchase in which an AI agent does the legwork for a person: it searches, compares offers and, where the channel allows, builds the basket and pays within limits the person set. The person still decides what they want and how much they will spend.

A shopper types “a gravel bike under CHF 4,000 for long alpine rides, available in Switzerland” into ChatGPT, Perplexity, Claude or Gemini. The assistant runs searches, opens product pages, compares prices, stock and delivery, then names two or three products. The gravel bike use case traces that mission field by field. What the agent reads is the markup, the Schema.org Product and Offer fields, the shipping and return policies and the stock status. When those fields are missing it moves on to a shop where they exist.

How is agentic commerce different from ecommerce?

In ecommerce a person browses the shop and decides on the page. In agentic commerce software reads several shops at once and arrives with a shortlist, so the shop competes on data an agent can compare rather than on layout.

QuestionEcommerceAgentic commerce
Who searchesThe person, in Google or on the shopAn agent, running several searches for one request
Who comparesThe person, across browser tabsThe agent, on price, stock, delivery and returns read as fields
What the shop is judged onPhotos, layout, brand, page copyStructured product data, policies, identifiers such as the GTIN
Where the purchase happensThe shop’s own checkoutThe shop’s checkout after a recommendation, or a protocol checkout inside the assistant
Who confirms paymentThe person, in the checkoutThe person, or a signed mandate the agent carries
What the shop measuresSessions and conversion rateHow often assistants name it, and orders arriving from agents

Most purchases still end on the merchant’s own site, so for now the agent changes how the shortlist is made before the person arrives, more than where they pay.

What are examples of agentic commerce today?

Named examples exist, and almost all of them launched in the United States first:

  • OpenAI launched Instant Checkout in ChatGPT on 29 September 2025 with US Etsy sellers, on the Agentic Commerce Protocol. In March 2026 OpenAI moved Instant Checkout into apps inside ChatGPT, according to an OpenAI spokesperson quoted by Digital Commerce 360, and kept the protocol as the link to merchants.
  • Google launched the Universal Commerce Protocol on 11 January 2026 for checkout in AI Mode and the Gemini app, for eligible US retailers, who stay the seller of record.
  • Ashley, the furniture brand, launched a transactional shopping experience inside Perplexity with PayPal handling payment (Woodworking Network, 30 November 2025).
  • JD Sports let US customers search and buy through AI platforms, starting with Microsoft Copilot (Reuters, 12 January 2026).

The traffic follows. Adobe Analytics measured a 693.4% year-on-year rise in visits from generative AI tools to US retail sites in the 2025 holiday season (Adobe, 7 January 2026). That is a US figure. For worked cases on Swiss shops, see the agentic commerce examples.

What are the five stages of an agent purchase?

Every agent purchase passes five stages, and each fails for a different reason:

StageWhat the agent doesWhy a shop drops out
DiscoverSearches and retrieves candidate pagesIts crawler is blocked in robots.txt, or the catalogue renders only with JavaScript
UnderstandExtracts what the product isNo GTIN, brand or model name in structured data, so it cannot match the product (GTIN guide)
ChooseCompares offers on price, stock, delivery and returnsThe price sits only in page text, the delivery time only in an image
BuyBuilds a cart and pays through a protocol such as UCP or ACPThe checkout works only with a mouse and a keyboard
RetainReorders under a standing mandateNo machine-readable order or account path

Our audit of 240 online shops in French-speaking Switzerland found the second stage to be the weak one: for 157 of 211 shops read in full, the stage that lost the most agents was reading the product itself (Romandie audit, July 2026).

Which protocols power agentic commerce?

Four protocols carry most of the work. MCP lets an agent call a shop’s tools, UCP and ACP define how it places an order, and AP2 records what the person authorised it to pay. Each covers a different step, so they stack, and a shop needs only the ones its channels use. The protocol comparison gives the owner, launch date and status of each.

Is agentic commerce available in Switzerland and Europe yet?

Recommendations are live everywhere ChatGPT, Gemini and Perplexity run, while agent checkout launched in the United States first. Adoption in the region is early: 5% of online shoppers in Switzerland, Germany and Austria had used an AI agent to help them shop, and about 18% planned to (University of St.Gallen, 24 June 2026).

So the first two stages, being found and being understood, pay off in Switzerland today, and the buying stages pay off as each channel opens here. The guide to agentic commerce in Switzerland covers CHF, VAT and TWINT.

What should a merchant change first?

Start with the stages that pay off in every market:

  • Let the search and shopping crawlers in, and make sure product content exists in the HTML without JavaScript.
  • Declare every product as Schema.org Product with an Offer carrying price, currency, availability, shipping details and a return policy.
  • Check that the price in the markup is the price on the page, since a mismatch gets an offer dropped.
  • Treat llms.txt as optional, for agents that fetch it, after the structured data. The llms.txt guide explains why Google does not need it.

Checkout work makes sense only after that, because an agent that cannot read an offer never reaches the cart. The free readiness check tests one product page in under a minute, the Buyer Journey Check runs all five stages through real assistants, and the Machine Readability Sprint fixes the product data in seven days.