Nine AI agents can now talk to your Shopify store on a shopper’s behalf. ChatGPT, Claude, Gemini, Copilot, Perplexity and four others. Sooner or later somebody is going to ask one of them what your return window is.
Shopify has already written the answer. It read your storefront, made its best guess, and filed the result in a new page at Sales channels then Agentic, where you can go and see what it decided on your behalf.
That page is the Knowledge Base, and it is the only thing in this release with a deadline attached. It holds the facts agents receive about your store: return windows, shipping destinations, gift services. Shopify drafts them and asks you to review. It is the first place a merchant can state, in a form a machine will read, what an agent may tell a shopper about their business.
Leave it empty and the agent will not go quiet. It will infer.
What shipped
The nine agents Shopify’s connectors now work with are ChatGPT, Claude, Claude Duet, Microsoft Copilot, Google Gemini, Grok, Grok Bot, Muse and Perplexity.
Separately, Shop Pay checkout is now live inside personal AI agents, starting with Muse from Meta, Claude Duet and Instinct. Those two lists are not the same thing and the difference matters: a connector means an agent can work with a store, while Shop Pay inside an agent means the purchase can complete without the shopper arriving at a storefront at all.
Shop.md is the mechanism underneath, described by Shopify as a skill that turns any AI agent into a personal shopper, searching more than a billion products.
The rest of the release is a list. Universal Checkout puts one checkout in the Shop app for carts built from several merchants. Sidekick has been rebuilt and can now run calculations, write code, search the internet and remember what a merchant teaches it. Canvas, which lets a merchant see a store on one canvas and edit it with Sidekick, is in early access. Seven new local payment methods arrive: Alipay, WeChat Pay, MB WAY, OXXO, Satispay, Scalapay and GrabPay. Shop Promise merchant eligibility has grown twentyfold.
Tool comparison · FastMoss vs Kalodata
Deciding which of your products are worth putting in front of an agent starts with knowing which of them actually move in their category. FastMoss and Kalodata do that job for TikTok Shop, reporting category-level demand and product movement rather than a platform’s own summary.
Comparison coming soon
FTC disclosure: our tool comparisons carry affiliate links. If you sign up through one we may earn a commission, at no extra cost to you. We rank on merit, never on commission, and the verdict is written by a human.
From the publisher
Everything we know about running Amazon, in one course.
The commercial strategy, taught by operators who run our own brands.
The figures, with their qualifiers attached
Several numbers in this release are Shopify’s own, about Shopify, and they need their limits stated before they travel.
More than 400 million people are said to trust Shop Pay. That is a count of people with Shop Pay, published by the company that operates it, and it is not a measure of how many will buy through an agent.
Shopify Payments is said to prevent more than 12,000 high-risk payments every minute. Twelve thousand out of what? Without the total processed in that same minute the figure could describe a razor-sharp filter or a very blunt one, and it says nothing at all about how many legitimate payments were caught in the same net.
Shop Promise merchant eligibility has grown twentyfold, with no base disclosed. Twentyfold from a small base and twentyfold from a large one are different claims, and the release supports only the multiplier.
The products may work perfectly well. These are still launch figures from an interested party, and they belong in any planning conversation with those words attached.
Why the Knowledge Base is the actionable one
Every other item in this release is something Shopify has built. The Knowledge Base is something the merchant has to fill in, which is why it is the only one with a deadline.
An agent answering a shopper’s question about returns, delivery or gift options will produce an answer either way. Given published facts it uses them. Given nothing, it derives something from the storefront, and a derived answer is correct only by luck.
The risk is not an embarrassing sentence but a transaction completed on terms the merchant never set. A shopper told by an agent that returns run to sixty days, when the policy is thirty, has been told that by something acting on the merchant’s behalf at the moment of purchase. Whatever the legal position turns out to be, the practical position is a dispute the merchant will absorb, at the time of year when dispute volume is highest.
That is the case for treating this as clerical work with a date on it rather than a feature to evaluate. It takes an afternoon. It is not strategic. And the window for doing it closes when peak starts.
A second consequence is already forming, even though nothing can be done about it yet. When a purchase completes through Shop Pay inside an agent, the shopper never arrives at the storefront, which means the merchant’s own analytics never see the session that produced the sale. The order appears; the behaviour that led to it does not. Merchants who optimise from on-site data have spent years building a picture of how shoppers decide, and this is the first channel that removes that picture while keeping the revenue. It is not a reason to avoid the channel. It is a reason to expect attribution to get worse before it gets better, and to be sceptical of any conversion-rate comparison that puts agent orders beside storefront orders as though they were measured the same way.
Tool comparison · FastMoss vs Kalodata
Agent orders arrive without the session that produced them, which makes independent category data more useful, not less. Our head-to-head of FastMoss and Kalodata looks at how each tool reports category demand and product performance across European markets.
Comparison coming soon
FTC disclosure: our tool comparisons carry affiliate links. If you sign up through one we may earn a commission, at no extra cost to you. We rank on merit, never on commission, and the verdict is written by a human.
What this does not tell you
Shopify has not published where these connectors are available, so the agent list should not be read as a global switch. Nor has it said how an agent chooses between competing merchants once several have published their facts, which is the question that decides whether this becomes a ranking surface with its own economics.
There is no pricing attached to any of it in this release, and no stated position on who carries liability when an agent misstates a term it was given. Those are the two commercial questions a merchant would want answered before building anything on top of this, and neither is answered.
Canvas is in early access, so it is not something to plan a store rebuild around this quarter.
What a seller does differently on Monday morning
There is one task here and it is concrete.
- Open Sales channels then Agentic, and read what Shopify has drafted about your store. It has written these facts from your storefront, which means it has guessed, and the guesses are now the answer agents give.
- Correct the return window, the shipping destinations and anything about gift services, and make them match what your policy page actually says. Where the two disagree, fix both.
- Check the edge cases a storefront usually leaves implicit, because those are where an inferred answer does most damage: who pays return carriage, what happens to a sale item, and which destinations you do not ship to at all.
- Do it before the end of November rather than in January. The value of a correct answer is highest in the weeks when volume and dispute rates are both at their peak.
- Leave the rest of the release alone for now. Universal Checkout, Canvas and the new payment methods are worth a look in the planning cycle, not this week.
Where the honest uncertainty sits
The figures in this release are Shopify’s own and carry no independent verification: the Shop Pay user count, the product count behind Shop.md, the payments-prevented rate and the twentyfold eligibility growth are all company-published, and three of the four lack a denominator or a base.
Geographic availability is not stated for the connectors or for the Knowledge Base, so a merchant outside the markets Shopify has prioritised may find the page present and the agent traffic absent.
And the commercial mechanics are entirely open. Nothing in this release says how agents will rank merchants, whether that ranking will be purchasable, or where responsibility sits when an agent states a term incorrectly. Those answers will matter more than any feature announced this week, and none of them has been given.
The three-line version
- If you run the store: open Sales channels then Agentic this week, correct the drafted facts about returns and shipping, and finish before the end of November.
- If you have just taken on the channel: the gap to name is that an empty Knowledge Base does not mean silence, it means inference, and an inferred return window is a dispute the business absorbs.
- If you sign the budget: nothing to approve. No pricing, no ranking mechanics and no liability position has been published, so any investment case built on agent traffic this quarter rests on economics nobody has seen.