Free tool · Evidence comparison

AI citation forensics

Paste an observed AI-search prompt and cited URL. OpenForBots compares the public evidence on that source with your site, shows verifiable differences, and separates useful tests from things no external audit can prove about the AI provider.

No AI engine is queried Same rules for both sites Nothing is stored

Methodology reviewed

  • Mention + citation

    The answer named your brand and linked to one of your pages as a source.

  • Mention only

    The answer named your brand but did not link to your site as a source.

  • Citation without explicit brand mention

    The answer linked to one of your pages as a source but did not name your brand in the answer text.

  • Not discovered

    The answer neither named your brand nor linked to your site.

Record what you observed

Compare your evidence with the cited source

Hostname only, for example example.com.
Without it, only site-wide evidence is compared for your side.
The exact text you asked, up to 500 characters. Do not paste the answer.
The answer named your brand and linked to one of your pages as a source.
One public HTTPS page the answer linked to.

No account needed. The observation is not saved, and analytics never include your domain, URLs or prompt.

How to use it

Record one observation, then compare the evidence

The comparison is only as good as the observation. Record exactly what you saw, once, with the date.

  1. 1

    Ask the answer engine your real question and note the date, the product you used and whether web search was on.

  2. 2

    Record what happened for your brand: named and linked, named only, linked without your name, or not there at all.

  3. 3

    Copy the full URL of one source the answer cited. Add your own best page for the same question if you have one.

  4. 4

    Run the comparison, then fix only the gaps that have an existing, evidence-backed fix and re-check later.

Understand the result

What the comparison confirms—and what it cannot tell you

What it can confirm

  • Whether each public page returned readable HTML to an OpenForBots request at the time of the check.
  • Differences in site-wide evidence under the same OpenForBots rules: crawler policy, structured data, identity signals and llms.txt.
  • Exact-page differences in server-delivered content, foundational HTML, structured data, declared authorship and declared dates.
  • Which of your gaps already has an existing OpenForBots fix that is useful regardless of this one answer.

What remains unknown

  • Why the answer engine chose, ranked or cited any source. Its retrieval and ranking systems are private.
  • Whether the same prompt returns the same sources tomorrow, in another country or for another user.
  • Whether the answer text was accurate. This release records your accuracy outcome but does not check it.
  • How completely each page answers your prompt. That is reported as not determined rather than guessed.

Learn and take action

Related guides and methodology

Common questions

Questions about citation forensics

Does this tool ask ChatGPT or another AI engine anything?

No. You supply the observation. OpenForBots only retrieves the public pages you name and runs its standard rule-based audit on both websites. No answer engine is queried, scraped or logged into.

Can it tell me why a competitor was cited instead of me?

No external tool can see that. The result shows which public evidence differs, which differences are worth fixing and testing, and which criteria do not separate the two sources at all. None of it is presented as the reason for the citation.

What does “Citation without explicit brand mention” mean?

The answer linked to one of your pages as a source but did not name your brand in the answer text. Record your own cited page as the cited source URL to compare it with your best page, or record a competitor source that the answer named instead.

Is my prompt or my competitor’s URL stored?

No. The observation is used to run the comparison and returned to your browser. It is not saved to a database, and OpenForBots analytics record only coarse workflow steps such as “started” or “completed”, never domains, URLs or prompt text.

Why are some results marked not comparable?

A difference is shown only when both sides were checked under the same methodology. If a page could not be retrieved, a site audit failed, or the two audits used different rule versions, the criterion is marked not comparable instead of being guessed.