Sign in Start free

← Docs  /  SEO Health & AI Visibility

AI Visibility

Updated 2026-09-11

28 AI-readability checks plus a probe of what ChatGPT, Claude and Perplexity actually say about the business — what it checks, the score, and how to run it.

Browse documentation

What it does

AI Visibility has two halves. The first crawls the site against 28 checks for how readable it is to an AI assistant and its crawlers: whether GPTBot, ClaudeBot, PerplexityBot, Google-Extended and Bingbot are blocked in robots.txt, whether llms.txt exists, Organization/WebSite/Article/FAQ/Breadcrumb schema, question-style headings and direct answers, visible dates and author bylines, thin or JS-only content, canonical tags, Open Graph, the lang attribute, HTTPS, and response speed. Each check rolls into the same 0–100 score and bands as SEO Health.

The second half is the Ask AI probe: it asks real AI assistants a set of questions about the business with no context fed to them — on purpose, so the answers reflect what the model already knows, not what today's homepage says — and classifies each answer against the site's own facts as correct, partially correct, incorrect, or "doesn't know". This is the only part of AI Visibility that spends AI credits.

The score, explained

Uses the same four-band scale as SEO Health — Excellent (90–100), Good (70–89), Needs attention (40–69), Critical (0–39) — computed from the 28 readability checks. The Ask AI probe's answer classifications are shown separately, as evidence, and are not folded into the numeric score.

How to run it

  1. Open a site's AI Visibility tab to see the score, the 28-check breakdown by category, and the probe's latest answers.
  2. Click Rescan for an on-demand crawl of the 28 checks at any time.
  3. Run the Ask AI probe separately when you want a fresh read on what assistants say — each run spends AI credits from the site's organization.
  4. Scans run on the same schedule as SEO Health: weekly on Free, daily from Studio up.

Limits & defaults

SettingDefault
Readability checks28, crawler-based, same schedule as SEO Health
Ask AI probeOn-demand, spends AI credits per run
White-label PDFStudio and up
Score bandsSame four bands as SEO Health

Troubleshooting

An assistant's answer is out of date or wrong
That is exactly what the probe is designed to surface: the model is answering from its own training and public sources, not the current page, so a stale or wrong answer is a real visibility gap worth fixing (updating schema, adding a direct-answer section, publishing the fact somewhere crawlable) rather than a bug in the check.
The probe says a run failed
Ask AI probe runs consume AI credits like any other AI action; check Settings → AI usage if runs are failing for insufficient credits.