The shopping trip that starts and ends in a chat window
Something changed in how people shop online this year, and most small retailers have not caught up with it yet. AI-attributed orders to Shopify stores are now up 11 times since January 2025, according to commerce data cited by Elogic Commerce. That is not a rounding error. That is a new channel appearing almost from nothing in eighteen months.
We have spent the last few weeks building this pipeline you are reading right now, which means we watch AI shopping behaviour closely for our own reasons. What we are seeing matches the data: ChatGPT, Gemini and Perplexity are becoming a real discovery layer for products, sitting somewhere between a search engine and a personal shopper. For a small UK retailer, the practical question is not whether this matters. It is what to actually do about it before your competitors work it out first.
What happened to “buy it inside ChatGPT”
OpenAI launched Instant Checkout in September 2025 with Etsy as the first partner, built on the Agentic Commerce Protocol it developed with Stripe. By February 2026 it had relaunched as “Buy it in ChatGPT” and expanded to more than a million Shopify merchants, including Glossier, Spanx and Vuori, with PayPal added as a second payment rail. OpenAI was charging merchants a 4% transaction fee, which combined with Stripe’s own processing cost pushed the effective take rate to around 9.2%, roughly a third of what Amazon Marketplace charges.
Then, in March 2026, OpenAI pulled back. According to CNBC’s reporting, the company dropped its plans to run checkout directly inside ChatGPT. The reason was not a technical failure. It was shopper behaviour: people were happy to research and compare products in a chat window, but wanted to actually pay somewhere they already had an account, saved card details and an order history, meaning their own website or app rather than a chat interface. We built our own daily content pipeline this month and hit a version of the same pattern: automation is brilliant for the repetitive middle of a job and worse than useless for the part a person actually wants to control themselves.
Why the retreat is good news for a small retailer
This is the bit that gets missed in most of the coverage. OpenAI stepping back from in-chat checkout does not mean agentic commerce is going away. It means the model has settled into something much easier for a small business to act on: discovery happens inside the AI tool, and the sale still closes on your own site. You do not need to integrate a payment protocol, negotiate a transaction fee, or rebuild your checkout flow. You need your product data to be legible to an AI system that is trying to decide whether to recommend you.
Per OpenAI’s own guidance on powering product discovery in ChatGPT, that legibility comes from two things: a structured product feed (CSV, TSV, XML or JSON, refreshed as often as every 15 minutes) and clean Product and Offer schema markup on your website carrying price, availability and review data. Without that, an AI shopping assistant simply cannot see your stock the way it can see a competitor’s who has done the work.
What UK retailers are already doing about it
Some of the UK’s bigger names have already moved. Debenhams became the first UK retailer on PayPal’s agentic commerce rail in February 2026, with a fuller UK rollout planned for later in the year. John Lewis has committed part of its £800 million transformation programme to making its catalogue discoverable through Gemini and ChatGPT rather than building its own checkout integration, a sensible bet given where OpenAI itself has landed. Neither of these examples requires a budget anywhere near a small retailer’s, though. The underlying work, structured data and a clean feed, is the same regardless of turnover.
The UK’s own Small Business Commissioner is now publishing guidance on exactly this, which is as clear a signal as any that it has stopped being a niche concern for large e-commerce operators.
The bit nobody has settled yet
Liability is still genuinely unresolved. If an AI agent buys the wrong size, the wrong quantity, or misreads a price, who is responsible: the shopper, the retailer, or the AI platform that made the recommendation? The Payments Association has flagged this as an open question, and the Competition and Markets Authority’s March 2026 paper made one thing clear in the meantime: existing UK consumer protection law applies whether a purchasing decision is made by a person or by an AI system acting on their behalf. There is no separate, lighter-touch rulebook for agentic sales. Get your returns policy, your pricing accuracy and your stock data right, because none of that gets forgiven just because a bot placed the order.
What to actually do this month
You do not need a developer team or a six-figure budget to be ready for this. Four things matter more than anything else right now.
First, add Product and Offer schema markup to every page that sells something, with accurate price, currency, stock status and review counts. This is the single biggest lever, and it is a few hours of work rather than a project.
Second, if you sell through Shopify, Etsy or a similar platform, check whether a merchant feed to ChatGPT Shopping is already available through your platform’s app store rather than something you have to build from scratch.
Third, do not spend time or money building a bespoke in-chat checkout integration. OpenAI itself has just proven that is not where the model is heading. Put that effort into your own site’s checkout instead, since that is where the sale will actually close.
Fourth, start checking your analytics for AI referral traffic specifically, separate from ordinary search and social. If you cannot see it yet, that is worth fixing before you decide whether any of this is worth further investment.
None of this needs a leap of faith. The 11x growth in AI-attributed orders is already happening to somebody’s shop. The only open question is whether it is happening to yours or to the competitor down the road who got their product feed sorted first.