Half the Stores an AI Agent Can Read Aren’t Ready to Sell to One.

What an agent needs, and how far we got

An AI shopping agent has to clear four gates on your store: crawl it (robots.txt allows its bot), read the catalog (a machine-readable product source), parse each product (server-side Product and Offer schema with price and availability, not JavaScript-only), and trust it (agent-readable reviews). We measured all four.

Of 1,600 sampled Shopify storefronts, 888 (55.5%) were reachable and 729 exposed a machine-readable catalog we could read a product from. Everything below is measured on those 729 — the stores an agent can already read — because that is the fair denominator. The stores it can’t even reach are a worse problem we don’t count against this bar.

A third of readable stores show the agent a blank page

Here is what the 729 readable storefronts actually expose in their served HTML, before any JavaScript runs, which is all a non-rendering agent gets.

35.0% (95% CI 31.6–38.5) serve no server-side Product schema at all — their product data is injected by JavaScript, so an agent that doesn’t render the page sees no product, no price, no stock. Even among those that do, only about half expose structured price and availability.

Almost nobody gives the agent a reason to trust them

Reviews are the single strongest signal an AI shopping assistant leans on when it decides what to recommend. 94.5% of readable storefronts (95% CI 92.6–95.9) expose no AggregateRating or Review data an agent can read. The star ratings are on the page for humans, rendered in a widget, invisible to the machine doing the recommending.

1 in 11 blocks the agent at the door, probably by accident

Before any of that matters, the agent’s crawler has to be allowed in. 8.7% of reachable storefronts (95% CI 7.0–10.7) disallow at least one core AI crawler in robots.txt. The tell that it’s not a considered choice: the block rate is nearly identical across every major bot.

A blanket Disallow copied from a template takes you out of ChatGPT, Claude, Perplexity, Google’s AI, Amazon’s Rufus, and Apple’s intelligence in one line, for stores that almost certainly didn’t mean to opt out of the channel their shoppers are moving to.

Put together: only half are transaction-ready

Combine the gates — allows the core crawlers and serves Product schema and price and availability — and 52.4% of readable storefronts clear the bar (95% CI 48.8–56.0). The other 47.6% are visible to a human and functionally invisible, or unreadable, to an agent trying to act.

Home & kitchen (n=129) edged out apparel & accessories (n=375), but both carry the same two holes: roughly a third serve JavaScript-only product data, and over 90% expose no agent-readable reviews.

Method, in short

Seeded random sample (seed 613) of 1,600 custom-domain Shopify storefronts drawn from the frame we built for our Shopify llms.txt study, re-crawled 15 September 2026. For each store we fetched robots.txt and parsed the agent-bot policy; fetched /products.json to confirm a machine-readable catalog, classify the vertical from product types and tags, and pull one product handle; fetched that product page and extracted every JSON-LD block, looking for Product, Offer (price, availability), and AggregateRating/Review; and fetched /llms.txt. Readiness metrics are computed on the 729 storefronts that exposed a readable catalog, so schema is measured only where we actually fetched a product page. All proportions carry Wilson score intervals.

Known limits, stated plainly: a static fetch cannot see JavaScript-rendered schema — which is the point, because non-rendering agents can’t either, though some agents do render and would see more than we credit here; we read one product page per store, not the whole catalog; stores that disable /products.json are excluded from the schema metrics rather than counted as failing; the frame is Shopify custom-domain stores from a list built five weeks earlier, so 44.5% were unreachable on re-crawl and the readable set skews to live, trafficked stores; and llms.txt presence is near-universal only because Shopify auto-serves it. Raw per-store data and the crawler ship with the study.