How the Agent Web Score is calculated
Each scan runs deterministic public checks across six dimensions. Dimension scores are severity-weighted pass rates, so failed high-impact checks pull the score down instead of merely shaving a few points from a perfect score.
Critical, high, medium, and low checks count as 10, 7, 4, and 2 weight units.
Pass earns full credit, warning earns half credit, and fail earns no credit.
Not-applicable, bonus, experimental, and info-only checks stay visible but do not affect the main score.
1. Discoverability
Strong: robots, sitemap, canonical URLs and crawlable homepage signals.
2. Content Accessibility
Mixed: HTML readability, markdown alternatives and heading structure.
3. Agent Access Policy
Strong: robots policy, meta directives and low-risk fetch access.
4. Protocol Readiness
Weak: MCP, WebMCP, Agent Skills, OpenAPI and well-known discovery.
5. AI Search Readability
Strong: Organization, WebSite, FAQ and action-relevant structured data.
6. Task / Conversion Readiness
Mixed: semantic CTAs, forms, demo, checkout and booking signals.
No. Lighthouse measures browser rendering and performance. Agent Web Score measures whether public signals support agent discovery, comprehension, and safe action.
No. It runs a bounded public quick scan across the homepage, response headers, robots.txt, sitemap.xml, /llms.txt, and fixed discovery paths. It does not perform deep crawling or authenticated testing.
The score is weighted by impact. Passing easy discovery checks helps, but missing protocol, content-accessibility, AI-search, or task-readiness signals can still keep the overall score low.
A starts at 90, B at 80, C at 65, D at 50, and F below 50. A site also needs to avoid high or critical findings before it can be labeled Agent-friendly.
Public domain reports do not require an account or email gate, and they give teams a shareable URL for remediation work.