Evidence first.
Every score explained.

Our free audit checks whether a public web page exposes useful technical signals to search crawlers. It does not measure the probability of an AI citation.

What we collect

We request up to three HTTPS pages from one website and their robots.txt. Each page is assessed separately; the headline score belongs to the first page. We inspect the initial HTML, not a rendered browser session. The report stores the page URL, business information you enter, timestamp, extracted evidence and suggested prompts. We do not store the full page body.

How the score works

The technical check pass rate is passed checks ÷ known checks × 100, rounded. Ten equally weighted checks cover title, description, H1 structure, noindex directives, canonical presence, initial text, parseable JSON-LD, and robots.txt access for OAI-SearchBot, Googlebot and PerplexityBot. Unknown crawler checks are excluded. This is not an AI citation, recommendation, traffic or revenue score.

Some checks are heuristics: one H1 and 100 words are diagnostic conventions, not search-engine requirements. A present canonical is not necessarily correct. Parseable structured data is not necessarily valid or useful. A high score does not guarantee indexing, citations or revenue.

What crawler access means

We evaluate the submitted URL against robots.txt rules. A pass means no applicable disallow was found; it does not verify firewall permissions or prove the real bot can crawl the site. Cross-domain redirects may have different rules and require a separate scan of the destination.

What we do not claim

Website audits do not measure AI citations. Manually supplied answers are analyzed as observations with their original scope. An API response is not a consumer ChatGPT observation. Research prompts are editorial templates, not real user query volumes. Review the coverage matrix for current automation and collaboration availability.

Evidence maturity, not confidence theater

Every executive report and Agent visibility snapshot states the valid observed-answer denominator, source basis, comparable remeasurements and excluded automated attempts. Failed, empty, truncated or provenance-incomplete runs remain visible as collection issues but never become negative observations.

• Not collected: no valid observed answer exists.
• Early signal: one or two scoped observations exist, or the source basis is still too narrow.
• Directional evidence: at least three valid observations exist and at least one has a citation-capable source basis.
• Repeatable evidence: at least ten valid observations, five citation-capable observations and three exact scopes repeated on later dates, with at least 90% valid automated completion when automated runs exist.

These labels describe operational evidence maturity. They are not a statistical confidence interval, market-share estimate, search-volume estimate or promise that an AI system will repeat an answer. A comparable remeasurement must match the buyer question, provider, market, location and language on a later date.

Sources behind the approach

Google’s AI features guidance ↗ explains that established SEO fundamentals apply and no special optimization is required. OpenAI’s publisher guidance ↗ describes OAI-SearchBot access.

Workspace and retention

Guest reports are private to a secure browser cookie valid for 90 days. Create a username/password account to keep access across devices. Save your one-time recovery code securely: there is no email reset service. Export JSON or print to PDF for a portable copy. You can delete a report from its report page. Reports remain on the server until deleted. Deleted data may remain in operational backups for up to 14 days. Free workspaces can save up to 50 reports.

Reproducible measurement

Review the metric dictionary and our research protocol before interpreting mention, recommendation or citation rates. Public sample reports include timestamped raw findings from our own pages.

Inspect the frozen panel, analysis tool, citation metadata and version history on GitHub →

Can a GEO change be verified?

A fix task retains its baseline observation and target page. After the change ships, collect a new valid answer for the same buyer question, provider, market, location and language. The proof view compares the dated answers, brand mention and first-party source links; a task is not marked proven merely because content was published. If the follow-up answer is missing, failed or outside the original scope, the outcome remains unverified. A changed answer is an observed result, not proof that the page edit caused it. See the measurement-to-fix workflow →