Nathan Smith · 18 June 2026
ChatGPT Product Feed: How OpenAI's New Merchant Tool Puts Your Products Inside ChatGPT
OpenAI's new merchant programme lets retailers feed their full catalogue straight into ChatGPT, so your products can be the answer when someone asks what to buy, with a buy button attached. Here's how the ChatGPT product feed works, how Instant Checkout and the Agentic Commerce Protocol fit together, and what we'd do first.
For twenty years, getting found by a shopper meant ranking on Google and, increasingly, paying to sit at the top of it. The shop window was a results page. This week OpenAI quietly moved the shop window somewhere new: inside the conversation itself. Its merchant programme at chatgpt.com/merchants lets retailers feed their full product catalogue straight into ChatGPT, so when someone asks "what's the best waterproof jacket for under £150", your products can be the answer, with a buy button attached.
We've been watching this one closely (it sits right at the intersection of AI, search and ecommerce, which is more or less our favourite corner of the internet). Here is what OpenAI has actually shipped, how the product feed works, and what we'd do about it if we sold things online.
What OpenAI has actually launched
Two things, really, and it helps to keep them separate.
The first is product discovery: ChatGPT can now recommend specific, real, in-stock products inside a normal chat, complete with images, prices and a link through to your site. The second is Instant Checkout: for eligible products, the shopper can complete the purchase without ever leaving ChatGPT. Both are powered by the same thing, a structured product feed that you send to OpenAI.
The crucial change is the direction of travel. ChatGPT is not crawling your website and guessing. You push a clean, structured file of your catalogue to OpenAI, and that is what it shows. For anyone who has spent years fighting to get a crawler to understand their product pages, that is a genuinely different model, and a much more controllable one.

How the product feed actually works
If you've ever set up a Google Shopping feed, this will feel familiar, because it is deliberately close to it. You provide a structured file describing every product, and OpenAI uses it to power search, the product cards, and checkout.
A few specifics worth knowing:
You push it, OpenAI doesn't pull it. The feed is sent over encrypted HTTPS to an allow-listed endpoint OpenAI gives you. There's a validation sample first to check it's compliant before you go live.
Formats are flexible. CSV, TSV, XML and JSON are all accepted, so most existing catalogue exports can be adapted rather than rebuilt.
You can refresh as often as every 15 minutes. That matters for price and stock. If something sells out, you don't want ChatGPT cheerfully recommending it for another day.
The required fields are the usual ecommerce suspects, and getting them right is the whole game: a stable unique id, title, a plain-text description (no HTML), link, image_link, price with an ISO currency, availability, inventory_quantity, brand and condition. You also supply your policies, the return policy, privacy policy and terms, because OpenAI wants the shopper to see who they're actually buying from.
The two flags that decide everything
Two boolean fields control how each product behaves, and they're the bit to understand before you touch anything else:
enable_search: can this product appear in ChatGPT's recommendations at all?enable_checkout: can a shopper buy it inside ChatGPT via Instant Checkout?
They stack: you can't enable checkout on a product that isn't enabled for search. So you can start cautiously, surface your range in discovery, learn what gets pulled, and switch on in-chat buying for your best sellers when you're ready.

The bit underneath: the Agentic Commerce Protocol
Instant Checkout runs on something OpenAI calls the Agentic Commerce Protocol (ACP), an open standard it co-developed with Stripe and has open-sourced. The jargon is doing a lot of work there, so here is the plain version, and the three things a retailer should actually care about.
You stay the merchant of record. ChatGPT never becomes the seller. It's a facilitating layer that passes order and payment details securely between the shopper and you. The customer is still your customer, the sale is still your sale.
Stripe is the easy path, not the only one. The protocol was built with Stripe, so if you're already on it the lift is smaller, but it supports other processors too.
Placement is organic, not paid. OpenAI is clear that recommendations are unsponsored and ranked on relevance to the shopper, not on who paid. This is not Shopping ads with a new coat of paint. You can't currently buy your way to the top, you earn it by being the genuinely best answer.
OpenAI does take a small commission on completed Instant Checkout sales, which is the part that will get a finance director's attention. But weigh it against the alternative: a sale you would otherwise never have seen, from a shopper who asked an AI for a recommendation and was handed yours.
The catch (for now)
Two honest caveats. First, this is live in the US only at the time of writing, across ChatGPT Free, Plus and Pro, and currently for single-item purchases. UK and wider rollout will follow, but it isn't here yet. Second, it's partner-led so far: Etsy sellers went first, Shopify merchants are rolling out, and applications are open for everyone else at chatgpt.com/merchants. OpenAI has said a self-serve merchant platform is coming later this year.
So for most UK retailers this is a "get the house in order now" moment rather than a "switch it on this afternoon" one. Which, frankly, is the best kind of moment, because it means you have time.
What we'd do first
If you sell online, here's the practical list we're working through with clients, in order:
Audit your product data. This is the single highest-leverage thing. Clean titles, accurate plain-text descriptions, real-time stock and price, proper brand and condition fields. The retailers who win in ChatGPT will be the ones whose feed is genuinely excellent, the same way clean structured data has always quietly won at search.
Treat your existing shopping feed as the starting point. If your Google Shopping feed is in good shape, you're most of the way to a ChatGPT one. If it isn't, fixing it pays off in both places at once.
Think about how people ask, not just what they type. Search keywords are short. Conversations are long and specific ("a gift for my dad who likes hiking, under £80"). The products that get recommended are the ones whose data answers real questions, so descriptions written for humans, not for a 60-character ad headline, start to matter again.
Get on the waitlist and watch the rollout. Register your interest now so you're ready the moment it opens to UK merchants, rather than starting from scratch when your competitors are already in.
The short version
OpenAI now lets retailers feed their catalogue straight into ChatGPT for product discovery and in-chat Instant Checkout.
It's a feed you push (CSV/TSV/XML/JSON), refreshable every 15 minutes, with
enable_searchandenable_checkoutflags controlling visibility and buying.You stay the merchant of record, placement is organic not paid, and OpenAI takes a small commission on completed sales.
US-only and partner-led for now, with self-serve and wider rollout expected later this year. The job today is getting your product data spotless.

The bigger picture is the one we keep coming back to: people are starting more and more of their buying journeys by asking an AI instead of typing into a search box. Google Shopping didn't kill SEO, it changed where the work went. This is the same shift, one layer further on. The retailers who treat their product data as a first-class asset, not an afterthought, are the ones who'll be the answer when someone asks.
If you'd like a hand getting your feed and product data ready for ChatGPT (and squeezing more out of your Google Shopping setup while we're at it), come and talk to us. It's exactly the sort of thing our AI and paid media teams enjoy far too much.