AI shopping assistants are no longer a demo-day curiosity. Millions of people now ask a chatbot to find a product, compare options, and start a checkout. Shopify noticed. Over the past year the platform rolled out agent-facing tooling at a scale nobody else has matched, and orders placed from AI-powered searches grew nearly thirteen times year over year in early 2026, per Shopify’s own reporting.
That sounds like the readiness question is settled. It is not. Researcher Bill Hunt, in a recent Search Engine Journal study, audited 122 top Shopify storefronts with a harder question than most readiness checkers ask: not whether AI can find a store, but whether an agent can resolve a purchase decision on one.
Across nine checks covering the full decision, zero of the 122 stores passed everything. The best three managed seven of nine, and all three failed the same two: no stated cancellation window, and no way to cancel once the order exists.
The Platform Did the Work, Not the Merchants
In early August, Shopify switched on WebMCP tools for every Liquid storefront. The tools were installed by the platform, not built by the merchants. Hunt’s data shows the same pattern at scale: 76 of 86 collected product pages load the identical Shopify adapter script, and 83 of the 89 agents.md files in the sample are near-identical copies of one Shopify-generated template. Only five were actually written by a merchant.
If you count agents.md adoption as a sign of merchant readiness, you are really measuring Shopify’s rollout schedule. A readiness badge on the storefront says the platform shipped; it says almost nothing about the merchant behind it. What merchants do control is where the failures concentrate.
The Machinery Works. The Answers Do Not.
Give Shopify credit, because the engineering is genuinely good. Hunt’s team ran live, non-completing checkouts on every store exposing the server endpoint, and all 74 behaved correctly, with a median response time of about 1.1 seconds.
The machinery works. The problem is that machinery is only half of a purchase decision. An agent can only act on a decision after making one, and that requires the buyer’s actual questions to have answers the agent can retrieve. Hunt’s team asked each store’s own policy tool the questions real buyers ask. “What is your cancellation policy?” worked as a control, answered every time. But “What does shipping cost to Boston, and when will it arrive?” was usable at zero of 88 stores. “Can I cancel an order after placing it, and by when?” was usable at 2 of 88. “Can used or open-box items be returned?” at 5 of 88. Worse than silence were the confident wrong answers: asked about cancellation, two stores returned their shipping FAQ instead.
Callable Is Not Completable
Every one of the 74 checkouts came back with a status requiring escalation and a message flagged as requiring buyer input, matching Shopify’s default rule that checkout requires human approval. It is a reasonable safety choice, and it quietly redefines “agent-ready”: on Shopify today, the agent builds a priced cart and a person commits the money.
Delivery date is missing everywhere: a live checkout priced shipping to Boston for 36 of the 74 stores, but none returned an arrival estimate or a carrier through any agent path. Pickup is invisible: all 93 stores with a fulfillment declaration list shipping as the only method, including retailers that clearly offer in-store pickup, so the agent gets a clear, machine-readable answer that is wrong.
And nothing can be undone. Not one of the 122 stores exposes a tool to cancel, change, or return an order after it is placed. This is not the merchants’ fault: the protocol defines no operation for cancelling or amending an order. Refund terms reach the agent only as a link to the policy page, not as values. At one store, the checkout even contained a $2.98 “Free Returns Coverage” line item the shopper never chose, with nothing marking it optional.
The “Ready by Default” Trap
It is easy to look at a Shopify store today and conclude it is agent-ready: the agents.md file exists, the tools answer, and most automated readiness checkers score the store well. The audit’s verdict is sharper: declared is not the same as done.
The semantic floor is broken too, on the product page where the purchase action lives: 47% of checkable stores had at least one unlabelled input, 46% at least one empty link, 37% at least one empty button. Among stores using dropdowns, 65% had no label on the size or option selector, and 19 of 68 had no add-to-cart control in the server HTML at all. Only 31 of 68 passed the full floor.
The good news is that accessibility, UX, and agent readiness depend on the same markup. A labeled size selector, a named buy button, and an announced “added to cart” message serve a screen-reader user, a keyboard user, and a browser agent at once. Fix the markup once and all three benefit.
Why the Search Team Cannot Fix This Alone
When Hunt sorted the failures by the team that owns the missing information, a pattern appeared. Pickup availability belongs to fulfillment. Delivery estimates belong to fulfillment. Cancellation windows belong to order management. Open-box exceptions belong to legal or customer service. Return windows as data belong to legal. None of those teams owns the website.
The checks nearly every store passes are the ones the platform generates automatically: catalog, variants, price, checkout. The ones nearly every store fails need someone outside the web team to write a commercial rule and a system to enforce it. That is an organizational problem, and the search team cannot fix it alone.
What Marketers and Store Owners Should Do Now
Start with the markup, the cheapest win that helps humans too. Label every form control on the product page, name every buy button, and announce cart confirmations in a live region. Then stop deleting and re-creating products: keep variant IDs stable and inventory current across feeds and catalogs, so agents stop hitting dead IDs.
Next, write the rules down as data, not prose. A return window stated in a paragraph reaches buyers; one exposed as structured data reaches agents. Audit what your own policy tool returns for real buyer questions.
Finally, fix the traps you already set. Review every pre-added line item, fee, and coverage plan in your checkout flow, and mark each clearly optional with an explanation of how to remove it.
The agent era of commerce is arriving whether your store is ready or not. The winners will be the stores whose machinery a buyer, human or machine, can trust to tell the truth about price, delivery, pickup, returns, and what happens after the money moves.

FAQ
Is my Shopify store already agent-ready?
Probably not. Most stores carry Shopify’s default agent tooling, but zero of 122 passed all nine decision checks. Declared readiness is not working readiness.
What is the difference between GEO and agent readiness?
GEO aims to get your store cited or ranked in AI answers. Agent readiness is whether an AI agent can complete a purchase decision on it.
What is the single cheapest fix a merchant can make?
Repair the markup on product pages: label every input, name every buy button, announce confirmations in a live region. It helps screen readers, keyboard users, and agents at once.
Why do agents fail on questions like “when will it arrive”?
No store in the audit exposed delivery estimates through any agent path, and most policy tools only answer questions phrased like policy pages.
Can agents cancel or change orders on Shopify stores?
Not today. No store exposed such a tool, and the protocol defines no such operation. Agents build a priced cart; a human approves the purchase.
What should I ask my team after reading this?
Ask who owns pickup availability, delivery estimates, cancellation windows, and return data. If everyone points at the web team, you have an organizational gap, not a marketing one.