Zero of 122 leading Shopify storefronts passed all nine checks for whether an AI shopping agent can fully resolve a purchase decision, according to an audit conducted between August and September 2026 — even though 89 of them now serve an /agents.md file and most expose a working, agent-callable checkout.

The audit, run against storefronts selected from published rankings of top Shopify merchants, set out to answer one question: not whether AI can find or cite these stores, but whether an agent can gather the information, apply the rules, execute the purchase and handle what follows. On the execution side, Shopify performs remarkably well. On everything around it, the defaults fall apart.

Who actually built the 'agent-readiness'?

In September 2026, Slobodan Manic argued that text-only versions of websites served to AI strip out the wrong layer, leaving "a brochure … served to the user most eager to act." He reported that Shopify switched on WebMCP tools (a protocol that lets an AI agent call site functions directly) for every Liquid storefront on August 5, and that "the merchants did not build it." The audit data confirms this at scale:

• 76 of 86 collected product pages load the same Shopify WebMCP adapter script.
• 89 of 122 stores serve an /agents.md file; 83 are near-identical Shopify templates. Only five in the entire sample were written by the merchant.

Count agents.md adoption as merchant readiness and you are really measuring Shopify's rollout schedule.

What does the checkout test show?

The strongest finding in the audit is that Shopify's tool layer works. Across 74 live, non-completing checkouts (create, read back, cancel, cancel again — never completing a purchase or sending payment details), every single store returned typed statuses, coded error messages, readable checkouts, confirmed cancellations, refused repeat cancels with machine-readable errors, and provided a handoff URL to a human. Median response time was about 1.1 seconds.

This is confirmed tool behavior, not speculation. It also directly addresses Manic's observation that agents re-submit forms because confirmations are "rendered for eyes." On September 28, Shopify announced WebMCP support in checkout, and a September 30 browser test at two sites registered the four announced tools — get_checkout, update_checkout, complete_checkout, navigate_to_storefront — with a read-only call returning full line items, tax-inclusive totals and coded messages about missing fields.

Where the defaults break down

The failures cluster where the platform's automatic data ends and merchant rules begin. Asked six buyer-style questions through each store's own policy tool:

• "What is the return window for this item?" got a usable answer at 70 of 88 stores — the one clear success.
• "What does shipping cost to Boston, and when will it arrive?" returned anything at just 1 of 88 stores, with 0 usable answers.
• "Can I cancel an order after placing it, and by when?" returned anything at 8 of 88; usable answers at 2.
• "Can used or open-box items be returned?" produced usable answers at just 5 of 88.

Two stores, asked about cancellation, returned their shipping FAQ instead. A confident wrong answer is more dangerous than none, because the agent has no reliable way to detect it. None of the 74 live checkouts returned an arrival estimate or a carrier, and all 93 stores with a UCP (Universal Commerce Protocol) fulfillment declaration list shipping as the only method — including retailers that visibly offer Click & Collect, whose endpoints answer "No pickup locations available" with a clear, machine-readable, wrong answer.

Every checkout also returned the status requires_escalation with requires_buyer_input — a deliberate safety rule that redefines "agent-ready": on Shopify today, the agent builds a priced cart, and a person commits the money.

An organizational problem, not a web problem

When the audit sorted failures by owning team, the pattern was stark. Pickup belongs to fulfillment, delivery estimates to fulfilment, cancellation windows to order management, return rules to legal. The checks nearly every store passes — catalog, variants, price, checkout — are the ones the platform generates automatically. The ones nearly every store fails need someone outside the web team to write a commercial rule and a system to enforce it. Not one of the 122 stores exposes a tool to cancel, change or return an order after placement; the UCP specification itself defines no such operation.

Product-page accessibility, which shares the same markup agents rely on, shows the same gap: 65% of stores using dropdowns have unlabeled size or option selectors, and 19 of 68 tested pages have no add-to-cart control in the server HTML at all. Only 31 of 68 pass the full semantic floor.

What to monitor next

Shopify already holds much of the missing data — return windows, cancellation windows, delivery dates, pickup availability — in structured settings, and could close the largest gaps in a single release. Until then, the audit's closing lesson cuts both ways: having the information isn't enough. Watch whether Shopify generates agent surfaces from those settings, and whether the UCP spec gains post-purchase operations. Most stores that look ready by default are only ready until the buyer asks a real question.