Catch AI citations that don't say what the AI claims
AI assistants sound authoritative, and they now attach sources to back themselves up. The problem is that those citations often don't support the claim: the page doesn't contain the fact, it says something slightly different, it's about another topic, or the link simply doesn't exist. This free AI citation verifier reads the actual source for each claim and tells you whether it holds up, so you can catch a hallucinated citation before you trust it or publish it.
How it works
- Paste the AI answer with its citation markers, the [1] [2] style that ChatGPT, Perplexity, Claude and Gemini produce, or with inline links.
- Paste the source list, one URL per line, numbered to match the markers.
- We fetch each cited page, find the passage most relevant to the claim, and a language model judges whether that source supports the claim, using only the source text.
- You get a verdict per citation: supported, partly supported, not supported, not found in the source, or source unreachable, each with a short quote and reason, plus an overall trust score.
What the verdicts mean
- Supported: the source clearly states or directly backs the claim.
- Partly supported: the source is related and partly backs it, but overstates, omits a key qualifier, or only implies it.
- Not supported: the source is on-topic but does not back the claim, or contradicts it. This is the classic hallucinated-citation signature.
- Not found in source: the page doesn't mention the claim's subject at all.
- Source unreachable: the page blocks automated requests, needs a login, or renders only in JavaScript, so it couldn't be read automatically. Check these by hand.
Who it's for
Anyone who publishes or acts on AI-written content: writers and editors fact-checking a draft, students checking an AI research summary, analysts validating a report, and SEO and GEO teams making sure their own cited claims stand up. It's also a quick way to see how reliable a given assistant's citations really are.
Good to know
- It checks up to eight citations per run so it stays fast and within the free AI model's limits. Verify longer answers in batches.
- Only sentences that actually cite a source are checked. Sentences that make a claim but cite nothing are listed separately, so you know exactly what couldn't be verified.
- It reads a slice of each source and runs on free models, so treat it as a strong first pass rather than a final ruling, and confirm anything important against the source yourself.
- Your source URLs are fetched on our server to get past cross-origin limits; the answer text is sent to a free AI model to judge each claim, so avoid pasting confidential content.