Shopify opened its billion-listing catalog to every AI agent

Shopify stopped choosing which AI agents may shop its merchants. As of October 13, any shopper can connect an agent of their choosing to their account on Shop, Shopify's order app, have it browse a catalog of more than a billion listings, and check out with one-click Shop Pay inside the conversation. The buyer approves each purchase. That last sentence is the whole compliance strategy.

The mechanics matter. The agent connects through the shopper's Shop account, searches structured catalog data (not scraped pages), and submits the order through Shopify's own checkout rails. The merchant stays the merchant of record: the legal seller of every order, same as a human-placed order. Shopify's president Harley Finkelstein told investors in August that agentic transactions "carry the exact economics as an online store transaction," and AI-driven traffic and orders to Shopify stores tripled year over year in the second quarter. The register stayed where it was; the door just widened.

The clever legal detail is in Shopify's default storefront robots.txt, the policy file agents read. It still says "Checkouts are for humans" and bars completing a checkout without "an explicit, contemporaneous human approval step." The in-chat approval on each purchase is that step. So the per-purchase sign-off is not just a UX choice; it is the hinge the whole compliance story hangs on. Remove it and the checkout policy breaks.

Behind the announcement sits the infrastructure. Since September 28, browser agents have had structured checkout over the Universal Commerce Protocol: they can read a checkout, update it, and submit it with the buyer's authorization instead of scraping pages built for people. The Catalog layer turns product listings into structured data agents can understand, and Shopify claims agents using Catalog data convert at twice the rate of agents working from scraped data. That is a relative claim with no absolute numbers attached, so treat it as direction, not evidence.

If you are a merchant, the homework is your listings. An agent can only repeat the case your product page makes. Shopify's own note is that merchants participating in Catalog are beginning to make more money, while agentic gross merchandise volume is still small and growing from a low base. The volume story is not written yet; the data-format story is.

The open question is liability, and nobody has priced it. In a PYMNTS survey of 60 US ecommerce decision-makers, 93 percent said the agent's provider should bear the loss when it buys wrong, and only 28 percent will let agents see their full product range. Visa's Trusted Agent Protocol and Mastercard's Agent Pay flag an agent at the register without shifting who eats a wrong purchase. Until someone prices the wrong purchase, the human sign-off stays. Shopify, by keeping the approval on every transaction, gets to say the wrong purchase was the buyer's call.

History is worth one paragraph. Stripe and OpenAI tried in-chat checkout inside ChatGPT in September 2025. Roughly a dozen Shopify merchants went live, sales came in near zero, and OpenAI retired the flow in March in favor of discovery and redirects. Owning the checkout surface without the merchant relationship did not hold. Shopify's version owns both, and it starts from the merchant side.

What remains unknown: whether Shopify ever reports what share of its volume actually comes from agents, whether shoppers will keep approving each purchase or push for standing budgets, and which liability regime lands when an approved-by-tap purchase turns out to be an agent hallucination.

Related: the Amex Intent Intelligence framework is the payments-side answer to the same problem, Visa, Mastercard and Ant's shared agent identity checks show who wants to own agent identity at the register, and the Muse agentic-commerce split at Meta is what the partnership-exclusive model looked like before Shopify opened the door.