Your Shopify catalogue is already in ChatGPT

For about a year the advice to merchants was to get ready for AI checkout. Two things happened instead. The in-chat checkout was cancelled, and your product data was switched on for AI surfaces without anyone asking you.

Almost everything written about agentic commerce is still about the first thing. The second is the one with work attached.

The in-chat checkout experiment, and how it ended

OpenAI launched Instant Checkout on 29 September 2025, with Etsy as the first US partner, running on Stripe and the newly published Agentic Commerce Protocol. Shopify was announced next, and the figure that travelled was a million eligible merchants.

By early March 2026, around thirty Shopify merchants were actually live through it.

On 24 March 2026 OpenAI published a post titled Powering Product Discovery in ChatGPT and deprioritised the feature. Its own explanation was that the first version "did not offer the level of flexibility that we aspire to provide". Reporting around the decision described merchant onboarding as arduous and product information in chat as frequently stale or wrong, and one account of Walmart's results had checkout inside ChatGPT converting three times worse than Walmart's own site.

The replacement model is the one now in production: discovery happens in the assistant, the purchase happens on your site, with your account linking, your loyalty programme and your checkout intact. For most merchants that is a better outcome than it sounds, because the hard parts stay where you control them.

Worth being precise about one thing: that million was eligibility, not usage. It described how many stores could connect, and it was repeated everywhere as though it were adoption.

What is actually switched on

This is the part that affects you whether or not you have read anything about protocols.

Around March 2026 Shopify began enabling Agentic Storefronts by default for eligible merchants, with no opt-in and no app to install. The admin setting is called "Allow Shopify to manage for me" and it is on. With it on, AI channels get access to your products through Shopify Catalog, direct checkout is activated for the relevant channels, and your store is auto-enrolled in new agentic channels as they are added. Meta was added on 8 September 2026, again by default.

What goes out through Shopify Catalog is not a teaser: titles, descriptions, images, pricing, availability, variants and store policies. The channels include ChatGPT, Google AI Mode, Microsoft Copilot and Perplexity.

By selling through these channels you accept the Agentic Storefronts Supplemental Terms. Nobody signed anything.

I am describing this from secondary reporting rather than from Shopify's own help pages, and these settings move. Go and look in your own admin, under Sales channels, at the Agentic section, and see what your store has on. If you take one thing from this article, take that.

What you can turn off, and what you cannot

Three things here surprise people.

There is no global off switch for Shopify Catalog. You can block individual channels. You cannot remove your catalogue from the mechanism that feeds them.

For ChatGPT specifically, opting out removes purchasing, not discoverability. Products stay findable; what you turn off is the built-in checkout.

Blocking AI crawlers does nothing about this. This is the one that catches technical teams. Disallowing GPTBot and the rest in robots.txt, or blocking them at the network edge, affects open-web crawling only. Your product data still reaches activated channels, because Shopify pushes it through Catalog rather than waiting to be crawled. If you added those robots.txt rules believing you had opted out of AI surfaces, you have not.

Two more practical notes. Changes take up to seven days to propagate, so this is not a switch you flip the morning a problem appears. And Shopify excludes B2B-only products when it can identify them as such, which means wholesale pricing that is modelled loosely is worth checking rather than assuming.

The protocols, and why you can mostly ignore them

There are two, and they have split along predictable lines.

	Agentic Commerce Protocol	Universal Commerce Protocol
Behind it	OpenAI, Stripe, Salesforce	Google, Shopify, Mastercard, Visa
Launched	September 2025	January 2026

If you are on Shopify, you are already inside whichever of these Shopify supports, without integrating anything. A protocol decision is a platform decision, and the platform has made it. Custom and headless builds are the case where this becomes a real engineering question, and even there it is early enough that the sensible move is to keep the product data clean and wait for one to win.

Forrester's March 2026 consumer work put purchasing inside an AI assistant as the least-adopted use case it measured. The money is not in the checkout integration. It is in whether the assistant describes your product correctly when someone asks.

So the work is product data, which is unglamorous

Here is the shift. For a decade, SEO for a store meant a page a human would land on. An assistant does not land on your page. It receives structured fields and answers from them, and it will not go and check.

What that changes:

• The product title is a data field, not a headline. "Classic Tee, 100% Organic Cotton, Unisex, Navy" answers a question. "The Only Tee You'll Ever Need" does not. An assistant filtering for organic cotton cannot infer it from a slogan.
• Descriptions have to contain the attributes, in words. Material, dimensions, fit, compatibility, care, what is in the box. If it only exists in a spec image, it does not exist.
• Structured attributes beat prose. Metafields and metaobjects are what Catalog can map cleanly. Facts buried in a paragraph are what gets paraphrased wrong.
• Availability and price have to be true right now. Stale product data was one of the stated reasons the checkout experiment struggled. An assistant that recommends an out-of-stock product at last month's price produces an annoyed customer and a return you pay for.
• Variants need to be modelled, not implied. If colour and size live in the title rather than as options, the assistant cannot filter on them.
• Policies are part of the answer. Returns, shipping times and destinations get quoted at the point of recommendation. Vague policy pages turn into vague answers.
• Your own site still has to be the better destination. Discovery in the assistant, purchase on your site means the handover lands on a product page that must load fast and work on a phone. That part has not changed.

None of this is new work for a well-run catalogue. It is the same product data hygiene that always mattered, now with a consumer reading it through a machine that will not squint at a photo to work out the material.

What to do this week

Open your admin and look at Sales channels, Agentic. Find out what is on. Most merchants reading this will discover a channel list they have never seen, including Meta.

Decide deliberately, per channel. Discovery is almost always worth having. Direct checkout in a channel is a real decision, because it affects your attribution, your customer relationship and your returns handling. Deciding by default is the only clearly wrong answer.

Pick your twenty best-selling products and read their data as an assistant would. Title, description, variants, metafields, availability. Ask whether a machine could answer "is this waterproof" or "does this fit a 2019 model" from the fields alone. For most catalogues the answer is no, and that is the whole project.

Fix the top sellers first. Catalogue-wide data cleanup is a programme. The products that earn the money are a fortnight.

---

This is the work we do. Smart-UI builds and supports Shopify stores, and structured product data is the part of that we would argue matters most right now. On Fine Matter the product record is the single source of truth for 94 independent makers, with every field traceable to a source, which is exactly the shape this needs.

If you want to know what your catalogue looks like to a machine, the fixed-price audit covers structured data and product modelling alongside performance, SEO and accessibility. For the storefront work itself, Shopify development is quoted within 48 hours of a brief.

If you are replatforming anyway, do this at the same time: the catalogue is already being remodelled, and that is the cheapest hour you will ever spend on it. We went through what else that involves in Magento to Shopify migration.

If you have just discovered a channel list you did not know about, send us the store and we will tell you what your product data looks like from the other side.

Sources: CNBC on OpenAI's shopping pivot, TechCrunch, Modern Retail on what went wrong, Fast Company on the protocol race, Axios on the original launch.