Feature · Layer 2, Citation

AI Visibility Audit

Crawl your site the way an AI crawler does, and find out which pages return almost nothing when a bot asks for them, plus the exact reason each one fails.

25 pages, no card, about two minutes.

The audit summary showing three headline numbers, each mapping to a different failure mode.Product screenshot
The audit summary. Three numbers that each map to a different failure mode.

A browser and a bot see different websites

Your pages probably look fine. You checked. But AI crawlers frequently do not execute JavaScript, and a page assembled client-side can hand a bot an empty shell while showing a human a complete page. Nothing about this is visible from your browser, your analytics or your CMS.

We fetch every page twice, once as a plain request with no JavaScript, once fully rendered, and compare what survives.

Content Visibility Ratio (CVR) is the visible words in a raw, no-JavaScript fetch divided by the visible words in the rendered page. A commercial page should be at or above 0.85. Below that, a bot is reading a fraction of what you wrote, and on some stacks it reads nothing at all.

What the audit checks

Seven scoring rubrics, around fifty individual checks. Every one produces a located finding rather than a grade.

Retrieval readiness

Section independence, entity naming, temporal anchors, answer position, vocabulary match, markup integrity, table framing and specificity, the eight properties that decide whether a passage can be lifted into an answer.

Bot access

Whether AI crawlers are permitted, blocked or silently failing, including the ones your robots.txt never mentions and your CDN may be challenging.

Page citability

Question headings, front-loaded claims, statistic density, quotes and attribution, section length, scannable structure.

llms.txt validation

If you publish one, we check it properly: single H1, name-shaped heading, blockquote summary, sectioning, link syntax, whether descriptions actually carry facts.

Claim anchoring

Answer capsules, section autonomy, proof pairing and brand association, whether a claim can survive being quoted away from its page.

GEO readiness

Server rendering, load speed, schema markup, consistent entity naming, hard statistics, named expert quotes, first-party data and earned media.

Findings, not a number

A score out of 100 tells you nothing you can act on. Every check that fails produces a finding: which page, which problem, which fix, worth how many points, with the evidence attached, the actual markup or text that failed.

Findings are ranked by remediation priority, so the list starts with what will move the number most for the least work.

A finding only changes state on evidence. If a page was skipped, failed to fetch, or wasn't sampled on a given crawl, its findings carry forward unchanged. We never mark something resolved just because we stopped looking at it.

A list of located, point-weighted findings ordered by what moves the score most.Product screenshot
Findings are located and point-weighted, so the list can be ordered by what actually moves the number.

Crawl volume by plan

PlanPagesCadence
Free audit25One-off
Diagnostic500Weekly
Visibility2,000Weekly
Program10,000Weekly

Full pricing →

Related

Findings & Fixes

What we generate once a fault is found.

Read more →

Share of Model

Measuring whether the fix worked.

Read more →

Competitor Tracking

Who is being recommended instead.

Read more →

The evidence behind this feature: do AI crawlers render JavaScript? and how AI crawlers index your site.

See what a bot actually reads

Run the free audit on your own domain. 25 pages, real findings, no card.