The pitch was simple and a little startling. Ask ChatGPT to buy a specific item, confirm, pay, and the order lands with the shop, all without leaving the chat. The buying moment, the part that used to belong to the shop's own website, happens inside the AI tool.
OpenAI started rolling that out as Instant Checkout on 29 September 2025, with Etsy sellers first and Shopify shops to follow, built on an open standard it published with the payments company Stripe. Then something instructive happened. By spring 2026, the company trying to make one-click AI buying work had quietly stepped back from it. That retreat is more useful to a small business than the launch was.
AI shopping reached checkout and then stalled
The technology underneath is real and it is not going away. OpenAI and Stripe released an open standard, the Agentic Commerce Protocol, that does three plain jobs: it lets a shop publish a machine-readable list of what it sells, it lets an AI agent place an order against that list, and it passes the payment through safely. Crucially, the shop stays the seller. You take the payment, you keep the customer, you handle delivery and returns. The AI tool introduces the sale; it does not take it over.
Google announced its own version, the Agent Payments Protocol, on 16 September 2025 with more than sixty partners, including Mastercard, American Express and PayPal. Its central idea answers the obvious worry, how a shop knows the agent is really acting for the customer: a signed, checkable record of exactly what the customer approved, the item, the amount, and the moment they said yes. The card networks are building their own takes on the same idea.
But by March 2026 OpenAI pulled back the smooth in-chat version of Instant Checkout, saying it "did not offer the level of flexibility that we aspire to provide", and started letting merchants use their own checkout while it focused on helping people discover products instead. Walmart and others built their own shopping experiences inside ChatGPT rather than handing the whole sale to it. The capability did not fail. The frictionless one-click version of it turned out to be harder than the demos suggested.
Product data and checkout caused the delay
Here is the bit worth underlining. The company did not stumble on the clever part, the AI completing a multi-step task. It stumbled on the unglamorous part: onboarding shops, showing accurate product information, handling baskets with more than one item, and connecting loyalty schemes.
If that list sounds familiar, it should. It is the same list that decides whether a normal website sells well. Accurate prices. Real stock. Sensible options. A basket that works. Some of the best-funded companies in the world have just spent a year discovering that the hard part of agentic shopping is the part a small shop can actually control: clean data and a path that does what it says.
The AI was ready before the plumbing was.
Check whether your product data is ready
We have written before about AI tools answering questions instead of sending clicks, and about your website becoming the source those tools quote. Agentic buying is the next turn of the same wheel, and it changes the question again.
Being mentioned is about visibility. Being bought from is about wiring. An AI tool can describe your business beautifully and still be unable to complete an order with you, because describing and transacting are different jobs. The first needs clear, quotable content. The second needs an accurate path from "I want this" to "it is paid for".
For a service business that takes bookings rather than card payments, the same shift applies. The question moves from "will an assistant recommend us" to "can an assistant actually book us", with the right time, the right service, and a real confirmation at the end.
An agent completing that action needs the same three things a hurried human needs. It needs to know what you actually sell: a clean, current list with real prices and real stock, not scattered across the website, a spreadsheet, and somebody's memory. It needs a real path to complete the action, connected to the same prices, stock and rules your website already enforces, so the order it places is one you can fulfil. And it needs you to stay the seller, in control of the sale.
This is the in-house version of work we already do. Modern Printworks, our own online art shop, runs an on-site assistant that answers from the real product catalogue and moves a customer towards a basket configured with the right size and materials. That is the same motion these standards are trying to make universal: an assistant that does not just talk about the product but helps complete the purchase, from real data, with the business still in control.
Most small shops can wait
The retreat we just watched is itself the argument for patience.
The standards are competing, not settled. OpenAI and Stripe have one, Google has another, the card networks have more, and it is not yet clear which a small UK business should back, or when any of this will be available here. Instant Checkout launched for buyers in the United States, not Britain. Picking a side too early is how you end up maintaining wiring that nobody uses.
The trust questions are genuinely unsolved, too. Letting software spend money on a customer's behalf is exactly as serious as it sounds. Earlier this year a US court blocked one company's AI browser from acting inside a major retailer's accounts, on the basis that a customer giving an agent permission is not the same as the retailer allowing it. That is the sort of question still being worked out in public, and it is not closed.
So the sensible position for most small shops is: not yet, but get ready. And getting ready turns out to be useful whether or not agentic buying ever becomes a real channel for you.
Improve product data before adding AI
The work that prepares you for an AI agent is the same work that helps a real person buy from you today. There are three moves, and none of them require betting on a standard.
Make your product data tell the truth. Accurate prices, accurate stock, the options a buyer actually needs, kept current in one place. That helps your website, your search visibility, and any future agent equally.
Shorten the path from wanting to buying. A good test: can a first-time customer get from landing on a product to a completed order in about three taps, on a phone, without emailing you a question first? If a person struggles with that, an agent will fail at it completely. Agents are unforgiving in a way that polite humans are not; they will not phone to clarify.
Decide your rules before anyone connects anything. What is allowed to be bought or booked automatically, what needs a person to confirm, and what an agent must never be able to do on your behalf. These are the same questions we work through on a site audit, and they are easier to answer calmly now than under pressure later.
The buying moment is starting to drift off your website and into the tools your customers already talk to. The companies furthest down that road have just shown that the limit is not the AI; it is the quality of the data and the path behind it. Fix those, and you are ready for whatever wins, while making your shop better at the job it already does. If you want a clear read on where your site stands, that is exactly the sort of thing we look at on a short call.
