Skip to content

How Check Your Website measures AI visibility readiness

Methodology last updated 2026-08-31

What the check measures

One check audits one public page across nine readiness areas. The areas always appear in the same order:

  1. AI AccessCan AI open your site?. Checks whether AI tools and search crawlers can open the page: fetch success, robots.txt rules, meta directives such as noindex or nosnippet, and whether main content only appears after JavaScript runs.
  2. Page IdentityWhat your page is about. Reads the page title, meta description, and visible text to measure how clearly the page states what it is about.
  3. Answer ExtractabilityAnswers AI can quote. Measures whether the page contains self-contained answers an AI tool can lift and quote directly, instead of useful statements buried in prose.
  4. Proof ReadinessProof people can trust. Looks for proof signals that support what the page states: specific facts, numbers, dates, and structured data near the claims.
  5. Schema MarkupHidden page labels. Evaluates the structured data on the page: whether JSON-LD blocks exist and whether they carry meaningful types.
  6. Machine ReadabilityA version AI can read. Tests whether the page offers a machine-readable version: a negotiated Markdown response, a page-specific Markdown twin, a valid root /llms.txt file, and a machine-readable discovery Link relation.
  7. Internal LinksLink text. Scores the link text on the page and checks that internal links resolve. Broken links lower the area.
  8. Page SpeedLoading speed. Scores severe mobile loading problems from PageSpeed Insights evidence: large LCP, slow TTFB, long tasks, layout shifts, and load failures. It is a severe-consumption signal, not a generic Lighthouse score.
  9. Content StructureEasy-to-scan layout. Judges whether the page is easy to scan: heading structure, extractable content blocks, and image context that help readers and AI parse the layout.

An area that cannot be measured reports unknown, and an area skipped because AI access is blocked reports locked. Neither state is ever shown as a score.

Input and output

You submit one public URL. The check audits that page only: workers do not crawl the site, discover links, or sample other pages. The check is free and needs no signup or install.

The anonymous preview returns the overall score and the AI access status. Creating an account claims the run and unlocks the full result: per-area scores, evidence, and the prioritized list of changes.

How scoring works

Every score is a percentage from 0 to 100. The overall score is the equal-weight average of the nine areas, shown to one decimal.

A score of 100 is clean, 90 to 99 is good, 70 to 89 needs work at risk, 50 to 69 needs work at alert, and below 50 fails. A hard blocker caps the overall score at 69%.

A schema markup score is capped at 90%; every other area can reach 100%. The live scoring formula is versioned as readiness-v2.

Final scores, blockers, fix order, and verification identity are deterministic projection outputs. A bounded AI layer only assists summaries and labels inside the evidence; it does not set scores.

Evidence model

Each area stores a bounded evidence packet: worker status, score, summary, recommendation, up to 8 findings, up to 5 errors, and an optional AI assessment.

The runner also stores bounded page evidence packets. Raw HTML, cookies, authorization headers, and secrets are never stored.

Competitor benchmarks

Competitors are optional and chosen at the first run, up to 3. Each competitor is benchmarked per area with the same projection as your page, and a public result waits until every requested benchmark has finished.

Verify after publish

Verification rechecks one issue at a time by re-running the single worker that backs that area. There is no global verify-all.

Each recheck ends in one of four outcomes: fixed, improved, still failing, or run failed. Before and after scores are persisted, so you can see what the change was worth.

If AI access blocked a check and is fixed later, a gated redo recovers the areas that stayed locked.

What this methodology does not cover

  • It is not a mention tracker: it does not watch live AI answers or count citations.
  • It is not a Google ranking report and it does not forecast visibility.
  • It gives no guarantee of rankings, traffic, or AI citations.
  • AI never edits your site. You publish the changes, then verify them.
  • One check covers one page at a time.

Read the topic, then run the check

The AI visibility page explains what the product is measuring toward. The methodology above explains how the measurement works.