How AI-readable
is the open web?
Lyrenth crawls and audits the open web for the signals that matter to AI agents reading it: structured-data coverage, render-mode mix, heading hygiene, content density, and access friction. The counters below are live; the seven signal aggregates recompute about every 10 minutes.
What the audit measures
Each audited page is scored on seven content-derived signals. Below is the corpus-wide mean for each one: how the open web performs against AI-agent-friendly hygiene.
Whether the page carries JSON-LD blocks (Article, Product, FAQPage…). Structured data lets agents extract facts without parsing prose.
Single H1, monotonic descent through H2 / H3. Predictable structure makes a page easier to skim and section.
Share of pages served without a headless-Chromium escalation. Static-renderable pages cost less and never arrive empty to first-pass scrapers.
Ratio of meaningful markdown to raw HTML. High density means most of the page is content, not nav / chrome / ads.
Non-empty, sensible-length title and description that are not generic placeholders.
Share of indexed pages that reach readers (and agents) without paywall or login-wall markers. Paywalled pages are indexed and scored too, so this is a measured rate across the corpus, not a definition.
Whether the page has enough words to be substantive on its own. Sub-stub pages get partial credit; pages with no body fail outright.
What these numbers mean
For AI agent builders
A high site-wide structured-data percentage means agents can extract facts cheaply. Low static-renderability or density means more of your input tokens pay for headless rendering and chrome, not content. Lyrenth normalizes these into a single AIDocument shape regardless of how the source page is built.
Read the quickstart →For site owners
The signals above are exactly what Lyrenth measures per page on your verified domains. Verify a domain to see your own AI Readiness score on the dashboard, broken down page by page.
Verify a domain →Methodology: every audited page is scored on seven content signals, each 0.0 to 1.0. Per-page scores roll up to a domain average and a corpus-wide mean. Raw page content stays in our private index; only aggregate counts and means are public. Counters are live; signal aggregates recompute about every 10 minutes.