Agentic Storefronts, explained for merchants: what Shopify sends to ChatGPT, Gemini, Copilot and Meta, and what it doesn't

Short answer: Agentic Storefronts is Shopify’s name for the set of AI channels it connects your products to. It is active by default for eligible stores. Through Shopify Catalog it sends each product’s title, description, options, images, price and availability to ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta. It does not send your metafields, size charts or compatibility tables, unless you map them into one of five listing fields. Everything below was read on Shopify’s own pages on 12 September 2026 and carries its link.

What Agentic Storefronts is

Shopify describes it in one sentence on the Agentic Storefronts help page (read 12 September 2026): “Agentic storefronts is active by default for eligible stores.” Your products reach the AI channels “through Shopify Catalog or through the Google & YouTube sales channel”, and you “manage which AI channels you want to sell your products on in Sales channels > Agentic”. Using it means agreeing to the Shopify Agentic Storefronts Supplemental Terms of Service.

Default-on dates from the week of 24 March 2026. Shopify’s news post of that day, Millions of merchants can sell in AI chats, put it this way: “Products become discoverable in ChatGPT by default via Agentic Storefronts, with no separate integrations, no apps, and no transaction fees beyond standard processing rates.”

The default is implemented by a single switch. On the Managing agentic storefronts page (read 12 September 2026), “Allow Shopify to manage for me is the default behavior”, and while it is on “you’re automatically enrolled in all future agentic storefronts channels”. Turn it off and you configure each channel by hand.

So if you have not looked, the honest assumption is that your catalog is already in front of these assistants. That is not a problem to fix. It is a reason to know what they were given.

The channels, and which ones can take the order

There are four channels today. They differ in one way that matters commercially: whether the shopper buys inside the assistant or on your checkout.

Channel	How the shopper buys	Who sees it	Extra requirements	Source, read 12 Sep 2026
ChatGPT	On your online store checkout, in a ChatGPT in-app browser (mobile) or a new tab (web). Discovery only.	Customers in the US; your store can be based elsewhere	None beyond Catalog eligibility	Selling on ChatGPT
Google AI Mode and Gemini	“Buy” in the Google interface, Shopify-powered direct checkout; or sent to your store if you turn direct checkout off	Direct checkout only for US customers; store must be US-based	Google Merchant Center account with products synced	Selling on Google AI Mode and Gemini
Microsoft Copilot	Shopify-powered direct checkout “without having to leave their chat conversation”	Direct checkout displays only to US customers	Store policies completed in Settings → Policies	Selling on Microsoft Copilot
Meta	Shopify-powered direct checkout “on Meta surfaces”; falls back to your store with the cart preserved	Direct checkout displays only to US customers	Facebook & Instagram by Meta channel installed with syncing on	Selling on Meta

Three notes on the table.

ChatGPT is a referrer, not a checkout. The main help page calls ChatGPT “a discovery-focused referrer platform”. The ChatGPT page says “ChatGPT users complete their purchase on your online store checkout in a ChatGPT in-app browser, or in a new tab when customers use ChatGPT web.” If you have read older coverage of an in-chat “Instant Checkout” for Shopify merchants, that is stale; since March 2026 the purchase happens with you. There are “no fees associated with selling in ChatGPT. You pay only your standard payment processing fees.”

Direct checkout is US-only on every channel that has it. All three direct-checkout pages use the same wording: the checkout “displays only to customers based in the United States”. Google adds that “your store must be based in the United States”. Copilot and Meta require only that you sell to US customers. Shoppers elsewhere can still see your products in those assistants; they land on your store to buy.

The order still reaches you. Direct checkout is Shopify’s checkout rendered inside the channel. It is not the assistant’s own payment system. No channel charges a fee beyond payment processing (each page states this in the same sentence).

What Shopify Catalog sends

Shopify Catalog is the feed behind all four channels. The Shopify Catalog and product discovery page (read 12 September 2026) lists what it carries: “title, description, options, images, price, availability, and other key attributes”. The data sharing page says the same in slightly different words, “titles, descriptions, images, pricing, and availability information”, and adds what stays private: “AI channels don’t have access to your full order history, orders from other sales channels, or your general customer database.”

Two sentences on the Shopify Catalog overview (read 12 September 2026) set expectations on both sides:

• “You can’t opt out of Shopify Catalog itself.” You can remove a channel’s access to it (more on that below), but the catalog exists regardless.
• Being included in Shopify Catalog, in Shopify’s words, “doesn’t guarantee” that a product “will appear in a specific AI answer, be ranked in a specific position, or be displayed by every channel that’s connected to Shopify Catalog.”

Which products get in is spelled out on the requirements page (read 12 September 2026). A product needs a title and at least one image, “a price greater than zero”, publication to the Online Store, Hydrogen or Headless channel with an identifiable URL, an active (not Unlisted) status, and it “can’t be hidden from search engines”. The store must be on Starter or above and not in private mode. The products page adds that B2B-only products are excluded automatically.

That is a clean, sensible list. Notice what is on it: the fields every product has. Notice what is not: anything you added.

What Catalog Mapping can map, and what it cannot

If your product title or description lives somewhere unusual, Shopify offers Catalog Mapping. The products page says: “If your product data, such as title, description, and category, are stored in custom fields, then you can use Shopify Catalog Mapping to ensure that the product data is correctly sourced for agentic storefronts.”

The mapping page (read 12 September 2026) names the fields you can remap:

• product title
• product description
• product category
• product variant options
• variant grouping (which products count as one listing)

Each of those can be pointed at a product attribute, a product metafield, a metaobject reference, a tag prefix, or a delimiter in the product title; grouping can also follow your Combined Listings setup. It is configured under Sales channels → Agentic, then “Review how your product data is sent to the Shopify Catalog for Agentic Channels”.

This is useful when, for example, your marketing description is a metafield and the default description field holds boilerplate. It is a way to choose the source for five fields. It is not a way to add a sixth. A weight_grams metafield, a compatible_with list, a helmet size chart stored as a metaobject: none of those is a title, a description, a category, an option or a grouping. There is no field on the mapping page for them, and the page does not describe one. As of 12 September 2026, the mapping page also carries no note about how unmapped metafields are treated, which we read as: they are not part of the syndicated record.

The fields Shopify infers for you

Developers who read the catalog from the other side see something merchants do not. Shopify’s Global Catalog extension (version 2026-08-25, read 12 September 2026) documents extra fields on each catalog record: metadata.tech_specs (“Technical specifications”), metadata.top_features (“Top product features”) and metadata.attributes (“Product attributes such as material, style, and occasion”). The page says these are “inferred or enriched by Shopify” and should be “treated as discovery and merchandising signals, not as merchant-authored source text”. There are also inventory signals such as availability.running_low, inferred descriptions and options, and a similarity search.

Read that carefully, because it is the closest thing to a spec field in the whole system. Shopify generates a tech_specs block from what it can see of your product. The extension page does not mention merchant metafields as an input, and the field is labelled as not merchant-authored. If your product’s deciding specification lives in a metafield the theme renders into a table, the inferred specs are a model’s summary of your page, not your data. Often good enough. Not the same thing as the number you maintain.

What the Knowledge Base app covers

Shopify also ships a free app, Shopify Knowledge Base, that sits beside the catalog. Its help page (read 12 September 2026) describes “automatically generated facts about your store” and FAQs that “aren’t displayed directly on your storefront - instead, they serve as a trusted data source for AI platforms to generate accurate responses about your store.” The FAQ management page shows the scope: shipping and delivery settings, return rules, payment options, account features, languages, plus FAQs you write yourself, which are “saved as metaobjects and can be edited”.

A note on the main page is worth quoting because it is the kind of honesty we like: the app “improves the accuracy of AI responses about your store but doesn’t affect how often your store appears in AI platform results.”

So the Knowledge Base answers store-level questions: do you ship to Norway, what is the returns window. Neither page describes product-level facts. “Does the 700c version fit a 28 mm tyre” is not a store FAQ, and the app is not built to hold one per product.

What this means when a shopper asks whether it fits

Put the pieces together and a shape appears. Shopify’s channel knows your products: which exist, what they are called, what they cost, whether they are available, what the options are. That is the majority of what a shopper needs, and on a text-rich catalog it is nearly all of it.

The questions that decide a specialist sale are different. “Which of these pads fits a C2 caliper?” “Is the 15.4-inch sleeve the same as the 15?” “What is the thread pitch?” “I have a 56 cm head, which helmet size?” The data that answers those is a metafield, a metaobject, an option with hundreds of values, or a table in the description. Of the syndicated fields, only description and options can carry it, and only if you have written it there in prose.

We measured how that plays out. In a simulated check on 4 September 2026 across 34 public specialist storefronts (340 questions, ten per store, one shopper model, OpenAI’s GPT-5.4 mini via the API, using only Shopify’s native storefront tools and public pages), 154 answers were fully right (45%), 144 partly right and 42 wrong. Of the 186 answers that were not fully right, 119 failed on a single attribute of a single product: a spec, a variant price, a stock state. The stores were chosen because their catalogs are option-heavy and spec-heavy, so the numbers describe that kind of store, not Shopify stores in general. On a 150-product coffee store two days earlier, the same native tools were enough: Anthropic’s Claude Opus 5 answered 10 of 10.

The related finding is where the missing facts sit. Across 41 storefronts we crawled on 5 September 2026, metafields were invisible from outside on all 41: not in /products.json, not in the storefront tools, not in the catalog endpoint. On one 1,055-product accessories store, a 15-inch versus 15.4-inch fitment comparison stayed partial for every model we tried, because the difference lives in a metafield. That is not a bug in the channel. It is the channel doing exactly what its field list says.

The same limit applies to the ten storefront tools Shopify added to every Liquid theme on 21 August 2026 (changelog, read 12 September 2026). get_product and show_variant read the product record; the tools do not expose metafields, and “agent support is currently limited to Chromium-based browsers through an origin trial.”

Should you turn it off?

Briefly, because a separate article on this site covers the decision (turn-off-agentic-storefronts). The mechanics are on the Managing agentic storefronts page and each channel page, read 12 September 2026:

• Under Sales channels → Agentic, turn off Allow Shopify to manage for me, open a channel, and deactivate Shopify Catalog access (or only Direct checkout, which keeps discovery and sends buyers to your store).
• “It can take up to 7 days before your product data is no longer being shared through Shopify Catalog.”
• “When access is deactivated, your products can still be discovered through web crawling”, in the same way they are listed in Google.
• To hide a single product from AI channels you set it to Unlisted, which the products page notes “also hides the product from sitemaps, search engines such as Google, and your online store search”.

In other words, opting out removes the structured feed and leaves the crawl. It does not make the assistant forget your store, and it does not improve what it says. If the concern is wrong answers about fit or specs, the fix is making that data readable, not removing the data that is already right.

How to see what is being sent

You do not have to guess. From the admin:

1. Go to Sales channels → Agentic. Each channel shows whether Catalog access and direct checkout are on.
2. Open “Review how your product data is sent to the Shopify Catalog for Agentic Channels”. This is where Catalog Mapping lives; it shows the source for title, description, category, options and grouping.
3. The Managing page also describes a search preview tool, “listing quality indicators”, a list of top-ranking products and a list of “products that didn’t rank” (read 12 September 2026). Run a query a shopper would type and see which products come back.
4. For the raw feed, open https://your-store.com/.well-known/ucp and the catalog endpoint at /api/ucp/mcp, which every storefront serves and which has advertised catalog version 2026-08-25 since 4 September 2026 (our reads). That is what an agent reading the catalog receives. Metafields are not in it.

Then ask the real question. Pick a product whose sale depends on one detail, and ask a fresh ChatGPT, Gemini or Copilot session about that detail. Note the answer and the date. If the answer is right, the syndicated fields carried it. If it is vague or generic, you have found the boundary of the feed.

Where StoreKnows fits

Shopify’s channel knows your products. StoreKnows adds what your product data knows: the metafields, metaobjects and option tables that decide the answer. It runs a simulated check on your store with buyer-intent questions, shows you every answer next to the product data behind it, and lets you publish read-only tools that answer “which variant, what price, in stock, does it fit” beside Shopify’s own. Nothing in Agentic Storefronts changes; the assistant simply has a second, merchant-authored place to look. The check is free.

Results in this article come from simulated checks run by StoreKnows on 2, 4 and 5 September 2026 against public storefronts. Questions were answered by OpenAI’s GPT-5.4 mini, GPT-5.5 and GPT-5.6 and Anthropic’s Claude Opus 5, called through their developer APIs, not the consumer apps, and graded by a separate judge model. Platform facts were read on the linked Shopify pages on 12 September 2026. StoreKnows is independently developed and not affiliated with, endorsed by or sponsored by OpenAI, Google, Anthropic or Shopify.