We scanned 26 live storefronts. None of them could be read by an agent.
Between 21 and 22 October 2026 we ran the same 14 checks against 26 live UK, EU and US direct-to-consumer storefronts (fashion, beauty, home, food). The sample is small, and it is a convenience sample — but the pattern is consistent enough to be worth publishing.
Method: public pages only. No accounts, no checkout, no paid tools. Each store scored once. Same 14 checks, same weights, one run per store.
Pass rate per check
| Check | Passed | Rate |
|---|---|---|
| Contact details | 26 / 26 | |
| AI crawler policy | 26 / 26 | |
| Returns policy | 24 / 26 | |
| Sitemap | 24 / 26 | |
| Checkout signal | 24 / 26 | |
| Shipping policy | 22 / 26 | |
| llms.txt | 22 / 26 | |
| Product feed (/products.json) | 21 / 26 | |
| Terms & privacy | 20 / 26 | |
| Stock availability in data | 10 / 26 | |
| Product JSON-LD | 8 / 26 | |
| Offer / price JSON-LD | 8 / 26 | |
| Product image in data | 8 / 26 | |
| Agent endpoint (MCP / agents.txt) | 0 / 26 |
What this means in practice
1. Catalogue data is the gap, not policy pages. Contact, returns and sitemap are mostly fine. Price, stock and image — the fields an AI shopping agent actually needs — fail in roughly seven out of ten stores.
2. Nobody has an agent door. Zero stores exposed /.well-known/mcp.json, /agents.txt or any comparable endpoint. Agents can only scrape HTML, which is exactly what the industry says it wants to stop doing.
3. The fixes are cheap. Most of these are template-level changes: a JSON-LD block on the product template, a footer link, a static text file. The expensive part is knowing what is missing and in what order.
Get the $99 report for your store →
Or read the practical guide: what GPSR actually requires from a Shopify store. Run the same 14 checks yourself, free: open the checker.
Limits
This is a convenience sample of 26 stores, not a statistically representative survey. Scores depend on what the homepage exposes and can change with a theme update. It is not legal advice and says nothing about whether a store complies with product-safety law — only about whether its public surface is machine-readable.