How the grades are made
Transparent checks, public endpoints, honest statistics.
Readiness score (0–100)
Discovery 30% — HTTPS, robots.txt access for AI crawlers (GPTBot, Google-Extended, ClaudeBot, PerplexityBot and others), response latency. Conformance 40% — a UCP profile found at /.well-known/ucp, valid JSON, spec version, well-formed services. Capabilities 30% — callable UCP services, payment handlers, declared shopping capabilities, and a merchant MCP endpoint (Claude-class agent callability).
What we probe
Public endpoints only: /.well-known/ucp, /robots.txt, the homepage (platform fingerprint, policy links, schema.org JSON-LD), one product page when discoverable, and /api/mcp · /mcp via a JSON-RPC tools/list. We send a clearly identified user-agent and respect 2–9s timeouts. No authentication is bypassed and nothing is crawled at scale.
Verdicts
VERIFIED — valid profile and score ≥ 80. PARTIAL — reachable but incomplete (most stores today). INVALID — a profile exists but fails conformance. BLOCKED — major AI crawlers are disallowed in robots.txt. UNREACHABLE — the domain did not respond.
Visibility win-rates
Separately from readiness, workspaces can probe agentic surfaces (ChatGPT/ACP, Gemini/UCP, Claude/MCP) with realistic shopping queries — sampled across queries, geos, variants, and repeats. Wins are purchase-led (the offer the agent buys), reported as rates with Wilson confidence intervals — never single-run claims.
Honesty
Scores are modelled, point-in-time estimates calibrated against sampled live observations. Agentic surfaces are non-deterministic and environment-dependent; a single check or probe is one draw from a distribution. That is precisely why everything here is a rate with a sample size.