Most store owners found out their products were being sold inside AI assistants the way they find out about most Shopify changes: a line in the changelog and a new entry under Sales channels. Shopify's agentic storefronts, in its own words, "lets customers discover and purchase your products in AI channels, such as ChatGPT, Google AI Mode and Gemini, Microsoft Copilot, and Meta", and the feature is "active by default for eligible stores".
On 8 September 2026 Meta joined the list, and on 28 September Shopify extended WebMCP, the protocol that lets AI agents operate a page, to the checkout. So the question for a store is no longer whether to be in the AI channels. It is whether the product data those channels are reading is good enough to be recommended on, and whether the store wants the sale to close inside the assistant or on its own pages. This article is the checklist.
Agentic storefronts are Shopify's feature for selling in AI channels such as ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta, with products shared through Shopify Catalog, or Google Merchant Center for Google's channels. For eligible stores it is on by default, and Shopify's changelog of 8 September 2026 says products are shared with Meta by default too. Each channel can be switched off under Sales channels, then Agentic, in the admin, and direct checkout is a separate control. The sections below are what to check before leaving it on.
The channels run on Shopify Catalog, Shopify's product data index, and in one case on Google Merchant Center. Shopify's help page: products are made available "through Shopify Catalog" or, for Google AI Mode and Gemini, through the Google & YouTube sales channel or Merchant Center. The Meta entry of 8 September says it plainly: "Meta now appears alongside your other AI channels in Agentic Storefronts in your admin", with products shared by default through Shopify Catalog, and Catalog access, direct checkout and performance for Meta managed in the agentic settings.
Two purchase models sit behind the channels. Shopify describes ChatGPT as "a discovery-focused referrer platform": the shopper finds the product in ChatGPT and completes the purchase on the store's own checkout, in an in-app browser or a new tab. For Google AI Mode and Gemini, Microsoft Copilot and Meta's surfaces, including its Muse assistant, "if you have direct checkout activated, then customers can complete their purchases in the channel's Shopify-powered direct checkout" without leaving the assistant.
The controls are under Sales channels, then Agentic, in the admin, where each channel can be switched on or off and direct checkout managed. Using the feature means agreeing to Shopify's Agentic Storefronts Supplemental Terms of Service. Stores on the Agentic plan, which exists for running agentic storefronts beside an existing online store, have direct checkout deactivated by default.
By default, Shopify sends "title, description, options, images, price, availability, and other key attributes" for every eligible product. Eligible means direct-to-consumer: Shopify's products page says agentic storefronts "support only direct-to-consumer (D2C) sales", and products are automatically excluded when they are B2B-only, detected through B2B catalogues, customer authentication requirements or a password-protected storefront.
Custom data does not travel on its own. Metafields, metaobjects and custom categorisation reach the channels through Shopify Catalog Mapping, which tells Catalog where to source each attribute. A store whose sizing, materials or compatibility live in metafields has to map them or the assistant will not know them.
Two things that surprise people. To keep a product out of the AI channels entirely, Shopify's route is to set it to Unlisted, which also "hides the product from sitemaps, search engines such as Google, and your online store search"; there is no AI-only hide. And blocking AI crawlers in robots.txt does nothing here, because, in Shopify's words, Catalog sends product data to activated channels regardless; the two are separate mechanisms.
Direct checkout is a commercial choice rather than a technical one. On, the sale closes inside the assistant on a Shopify-powered checkout: fewer steps for the buyer, no visit to the store, and whatever the store does at checkout (the upsell, the newsletter opt-in, the remembered card) happens inside the channel's frame or not at all. Off, the assistant refers the shopper to the store, where the brand controls the page and the checkout but adds a step the assistant was built to remove.
The honest test is the same one applied to any marketplace: how much of the store's margin comes from what happens around the purchase rather than from the purchase itself. A store that lives on repeat orders and account sign-ins has reason to keep the shopper on its own pages, where
the 2026 checkout changes for signed-in customers do their work. A store selling a considered one-off purchase has less to lose by closing it where the shopper already is.
A sale made through an assistant is still the store's sale. Shopify's help page says orders from AI channels "display in your Shopify admin with channel or referrer attribution", and that the store retains "full ownership of the customer relationship and post-purchase experience".
Fulfilment and returns run as usual. The order arrives in the same list as every other order and goes through the same fulfilment, shipping and returns. A shopper who bought inside an assistant will usually come to the store, not the assistant, when a parcel is late.
Attribution comes in two kinds. An order closed on a channel's direct checkout is attributed to the channel. A ChatGPT sale, which completes on the store's own checkout, arrives as a referral. Orders placed by a shopper's personal AI agent are attributed to the agent, in Shopify's words. Reading the AI channels together means reading all of these.
Check the store's own tracking. Measurement set up for the online store may not see a purchase that closes inside another company's interface. Place a test order through each channel with direct checkout on, and see what the store's own reports record, before relying on them.
Compare like with like. Margin, discount use, return rate and repeat purchase by channel, against the online store. A channel that sells well and returns twice as often is a different channel from the one its sales figure describes.
That comparison is what the direct checkout decision above should be revisited with after the first month, channel by channel.
The most direct check on what an AI channel does with a catalogue is to shop in it. Once a month, take the questions a buyer of the store's best sellers would ask, and ask them in each channel the store sells through.
Ask as a buyer, not as the brand. "A waterproof jacket that fits over a fleece, under $150" tests whether the product is recommended. Asking for the brand by name only tests whether the assistant can find it.
Write down what comes back. Whether the product appears, what the assistant says about it, the price and stock it shows, and which competitors appear beside it. Note the date and the exact question, because answers vary between sessions and over time.
Compare it with the catalogue. A wrong price or a missing size usually traces back to the fields Catalog sends. A description of the product that misses its main advantage usually traces back to a description that never states it.
Treat it as a spot check. A handful of questions is not a ranking report, and an assistant's answers can be personalised and change from day to day. What it does show is whether the facts the store wants repeated are being repeated.
Follow one purchase through. On a channel with direct checkout on, complete a test order and a return, so the store has seen the experience its customers are having before one of them describes it.
An assistant recommending a product is reading the fields Catalog sends and deciding, from them, whether this product answers the shopper's question better than the next one. That makes the quality of those fields the whole of the opportunity, and Shopify's own advice is to keep "complete and well-structured product information in Shopify Catalog" and follow SEO best practice.
The description. If it was copied from the supplier, every other retailer sent the same paragraph and the assistant has no reason to prefer yours.
Product descriptions and supplier feeds is the longer argument; the short version is that the description has to state who the product is for and what question it answers.
The options and attributes. Size, colour, material, compatibility, what is in the box: as structured fields, mapped through Catalog, not buried in a paragraph. This is the same work
structured data for online stores does for search engines, now with a second reader.
Price and availability. Sent live, which means an assistant will recommend a product that is in stock and not one that is not. A store that lets stock and price drift on long-tail products has handed the recommendation to a competitor whose data is current.
The collections. Catalog reads how the store is organised, and
Shopify's new collections describe how that organisation is now built at variant level and nested, which is how an assistant understands a range.
The answers. The reasons a shopper hesitates, stated on the page.
Answer engine optimisation for Shopify stores covers how a store gets cited by an assistant; agentic storefronts are the step after citation, where the assistant sells the product directly, and the content requirements are the same.
Product fields answer what a product is. Shoppers also ask about the store: how long delivery takes, how returns work, whether there is a warranty. Shopify has a separate place for that, the knowledge base, under Sales channels, then Agentic, then Sources, in the admin.
What it holds. In Shopify's words, it is where a store can "publish facts about your store for AI retrieval": the return process, shipping timelines and brand values, grouped into categories including company policies, returns and warranty, fulfilment, brand and identity, contact and payment, with a category for additional facts.
Where it comes from. Shopify generates it from what is publicly available on the storefront and from admin data, and keeps what it finds unpublished until the store publishes it. Shopify says it never changes knowledge base content on the store's behalf, so a fact goes live only when somebody approves it.
Who reads it. Once published, it is available through Shopify's Storefront MCP to AI channels, including Shop and third party platforms. It holds up to 1,000 facts, and stores on the Agentic plan can add only additional facts.
What to check. Read every suggested fact against the policy pages before publishing it. A generated summary of a returns policy that drops the exceptions is worse than none, because an assistant will repeat it to a shopper as the rule.
Keep it current. Delivery promises change at peak and policies change through the year. The knowledge base is one more place each change has to reach, on the same day it reaches the policy page.
Open Sales channels, then Agentic, and read what is on. Decide direct checkout per channel on the margin test above. Export the catalogue and read the default fields for the twenty best sellers as a stranger would. Map the metafields that carry the answers. Set to Unlisted anything that should not be sold through an assistant, knowing what else that hides. Then read the performance view for each channel monthly, because this is a sales channel with its own numbers and it should be run as one.
That review, and the product data work behind it, is what our
commerce services are for: the listing content, the structured attributes, the search and the margin, treated as one system because an assistant reads them as one. If you would like to know what an AI channel currently sees when it reads your catalogue, that is a short export and we are glad to look.