The state of agent commerce
Measured across 50 storefronts with rubric v1. Every number here is reproducible: each store has a public page showing the individual measurements behind its score.
50
Stores measured
39/100
Average score
48%
Effectively closed to agents (grade F)
92% of measured stores fail stated automated-access policy.
Based on 50 stores where the signal could be measured.
Where stores fail
- Stated automated-access policy92% fail
- Agentic Commerce Protocol support90% fail
- MCP or A2A discovery document90% fail
- Machine-payable endpoint signals (x402 / AP2)90% fail
- Checkout forms are semantically labelled83% fail
- Machine-readable product feed80% fail
- llms.txt guidance file80% fail
- Category pages expose an ItemList74% fail
- Stable product identifiers72% fail
- Machine-readable contact / API surface66% fail
- Valid Product structured data61% fail
- Offer with price, currency and availability61% fail
- Cart or checkout entry point is reachable58% fail
- Content is server-rendered38% fail
- Serves content to a non-browser user agent36% fail
- Sitemap is discoverable and structured32% fail
- Time to first byte11% fail
- HTML payload size11% fail
- Product paths are crawlable8% fail
- robots.txt allows AI crawlers2% fail
- No forced login before checkout0% fail
Grade distribution
- A0 stores
- B0 stores
- C3 stores
- D23 stores
- F24 stores
Journalists: the underlying per-store measurements are available as JSON at /api/score/<slug>. Email [email protected] for the full dataset.