What claim-level review means
Most “AI fact-checking” gives a document a vibe score. Claim-level review is stricter: the text is broken into individual factual claims — numbers, dates, quotes, named entities, causal statements — and each claim is checked on its own.
The output is not a grade. It is a list you can act on: this sentence is fine, this one needs a source attached, this one has no support and should be rewritten or removed.
Supported, needs-source, unsupported
Every claim gets one of three verdicts. Supported means the claim is backed by the material in the run — your uploads, briefs, or cited sources. Needs-source means the claim is plausible and specific, but nothing in the run confirms it. Unsupported means the claim contradicts the available material or cannot be traced to anything at all.
Needs-source and unsupported claims are exactly the ones that end up in client corrections — or worse, in a published correction.
Two tools, one review loop
The free Claim Checker runs the same engine on any pasted text — no account needed — so you can test it on a real deliverable before adopting anything.
Inside a workspace, the Reality Checker runs automatically on every run, and verdicts land on the approval screen. The operator who approves external sends sees flagged claims in context, not in a separate report.
Honest limits
The checker does not browse the web to confirm facts yet. Verdicts are relative to the material in the run and the model's knowledge, which has a cutoff and can be wrong.
It also cannot judge intent, tone, or strategy. It catches factual exposure; a human still owns the message.
AI fact-checker FAQ
Does the fact-checker browse the web to verify claims?
Not yet. Verdicts are based on the sources inside the run plus model knowledge. Claims that need live-web confirmation come back as needs-source so a human can verify them.
Is the free Claim Checker the same engine as the in-product check?
Yes. The free Claim Checker is the Reality Checker engine on pasted text. In a workspace it runs automatically on every run and feeds the approval screen.
What kinds of claims does it catch?
Statistics, dates, quotes, named entities, comparisons, and causal claims. It is tuned for the factual statements that create liability in client deliverables.
Does it replace human review?
No — it aims human review. The operator approving a send sees flagged claims first, so review time goes to the sentences that actually carry risk.
Check a real deliverable right now.
Paste any AI-generated text and see which claims lack a source.
Run the free Claim CheckerFree · no signup required