AI fact checker - reports coverage, not truth

The AI fact checker:do its sources hold up?

Upload an AI research report (or paste text). We check that its citations resolve and match, recompute its internal arithmetic, and flag what doesn’t add up. It reports coverage, not truth — it tells you where to look.

REPORTCITATIONS
PDF, .txt, or .md
Try a sample →

Source integrity

/ 100

Verify a report to see source integrity — citations resolved and matched, internal arithmetic recomputed, a shareable record.

How it works

About the AI fact checker

The fact-check sources tool takes a document — or a link to a public report — and checks the work behind it. It confirms each citation resolves and actually supports the claim it’s attached to, recomputes the document’s internal arithmetic, and flags claims that are unsupported or contradicted by their own sources. You get a summary plus a saved, shareable record of what held up and what didn’t.

It’s built for AI-generated research reports, which often cite sources that don’t support the claim — or don’t exist at all. Crucially, it reports coverage, not truth: it tells you which citations hold up and where the document contradicts itself or its sources. It does not certify that the document is “true”.

  1. 01

    Give it a document or a public link

    Paste a report with citations, upload a PDF / .txt / .md file, or pass a link to a public report and it fetches the page or file directly.

  2. 02

    It resolves each citation and checks the match

    Every reference is followed to see that it resolves, and then read to see whether it actually supports the claim it’s cited for — not just that the link works.

  3. 03

    It recomputes the internal maths

    The document’s own arithmetic — totals, percentages, figures it carries forward — is recalculated so numbers that don’t add up are caught.

  4. 04

    It flags unsupported and contradicted claims

    Claims with no supporting evidence, or that their own cited source contradicts, are surfaced. An optional deeper pass reads further across the web to test claim entailment.

  5. 05

    You get a summary and a shareable permalink

    Results are saved under a content-hash id with a shareable permalink, so you can revisit or send the findings later.

What you get back

Per-citation status

For each reference: whether it resolves, and whether it supports the claim it’s cited for.

Internal-maths findings

Where the document’s own totals, percentages, or carried-forward figures don’t reconcile.

Unsupported & contradicted claims

Claims with no supporting evidence found, or that their cited source contradicts.

A shareable permalink

A saved record under a content-hash id you can revisit or send to someone else.

Common uses

  • Vetting an AI-generated research report before you act on it
  • Checking a report’s citations before you forward it on
  • Due diligence on a vendor’s whitepaper or pitch deck
  • Editorial citation review before publishing

This tool reports coverage, not truth. “Supported” means a source backs the claim — not that the claim is true. “Unsupported” means no evidence was found — not that the claim is false. Use it to see which citations hold up and where a document contradicts itself, then judge the substance yourself.

Questions

Is this a citation checker?
Yes — a citation and reference checker in the sense that matters: it checks that each reference resolves to a real source, that the source is what the report claims it is, and that it supports the specific claim attached to it. It is not a bibliography formatter — for MLA or APA style questions you want a citation generator, not a checker.
Can it detect AI hallucinations?
The checkable kind, yes. Fabricated and misattributed references — the most consequential hallucination in professional documents — are verified mechanically: each one either resolves, matches, and supports its claim, or it doesn’t. In deep mode it also tests whether claims are entailed by the available evidence. That is coverage of the evidence, not a truth verdict: a report can pass every check and still be wrong.
Does it tell me whether the report is true?
No. It reports coverage, not truth — which citations hold up, where the maths doesn’t reconcile, and where the document contradicts itself or its sources. “Supported” means a source backs the claim, not that the claim is true.
Can I just give it a URL?
Yes. Paste a link to a public report — a page or a PDF — and it fetches the content directly. Only public http(s) links are accepted.
Why does AI deep-research need this?
AI-generated research reports frequently cite sources that don’t actually support the claim, or that don’t exist at all. This tool follows each citation and checks whether it holds up, so you don’t have to take the report’s references on trust.
Is the result shareable?
Yes. Each run is saved under a content-hash id with a shareable permalink, so you can revisit the findings or send them to someone else.
What does the deeper check add?
The optional deeper pass is slower: it reads further across the web to test whether each claim is entailed by the available evidence, beyond just the source it was cited against.
Is this a free AI fact checker?
Yes, to start — every account gets a free 7-day trial with 200 credits, no card required, and a fact check costs 2 of them. Every citation is checked that it resolves and supports its claim, with per-citation results you can inspect rather than a single score.
How is this different from fact-checker websites like Snopes?
Different jobs. Fact-checking sites verify public claims with human editors. This tool verifies the inside of a document you give it — whether its citations exist and match, and whether its own numbers reconcile. For an AI-written report, the mechanical check usually comes first.

Developer recipes & tools