Find the claims in an AI answer that need checking.
Paste an AI answer and its sources. Ask reviewers to identify unsupported claims, contradictions, and missing context, then verify the important findings against primary evidence.
Two free Refinement reviews each month · No credit card required
Plausible wording is not evidence.
A response can mix sourced facts, inferences, and speculation without making the difference clear. Ask reviewers to separate those parts and identify what supports each claim.
This review is a starting point for fact-checking. Several models can repeat the same mistake, and a citation still needs to be opened and checked.
What to look for in the review.
Separate the claims
Break a broad conclusion into statements that can be checked.
Question the support
Ask whether the supplied source actually supports each statement.
Keep uncertainty visible
Mark what remains unverified instead of treating agreement as confirmation.
Different claims require different checks
A fact checker becomes more reliable when it classifies the claim before deciding what evidence and language are appropriate.
| Claim type | Common failure | Required review |
|---|---|---|
| Date or event | Plausible but wrong chronology | Authoritative record and publication versus event date |
| Number | Invented precision or denominator | Original dataset, unit, period, population, and calculation |
| Quotation | Paraphrase presented as exact words | Primary transcript and compliant exact wording |
| Causal claim | Correlation promoted to mechanism | Study design, alternatives, and strength of inference |
| Current status | Outdated law, price, role, or feature | Recent primary source and jurisdiction or version |
| Opinion | Treated as objectively verifiable | Assumptions, criteria, counter-case, and attribution |
| Recommendation | Conclusion outruns evidence | Decision context, trade-offs, uncertainty, and accountable owner |
Prepare a useful review.
Give reviewers a clear question and enough context to examine it.
Preserve the original wording
Review the exact answer and original question.
Rank claims by consequence
Verify the claims that control the recommendation, create legal or safety exposure, contain precise numbers, or are likely to be repeated.
Model verification is a concern signal until evidence is attached and checked.
A verifier model can hallucinate citations, misread dates, or confidently endorse information from its training data. Cross-model agreement may reflect a shared source rather than independent confirmation. For consequential claims, open the primary source, confirm the exact support, and record who performed the check.
Some facts are time-sensitive, jurisdiction-specific, private, or inaccessible to the panel. State the cutoff and scope. Protect confidential material, respect copyright and access restrictions, and escalate medical, legal, financial, safety, and reputational claims to qualified reviewers.
What to know before you run the panel
Can an AI fact checker browse the web and verify current information?
Models can use available search tools when configured, but retrieval is only the beginning. A human should confirm the primary source, publication date, applicable version or jurisdiction, and whether the source supports the exact claim rather than a nearby statement.
What should I paste with the AI answer?
Include the original prompt, exact answer, any cited sources, the intended audience, the decision or publication context, and the date. Identify claims that would cause the most harm if wrong so the panel can prioritize them.
Does agreement mean the answer is correct?
No. Models can repeat the same mistake or rely on the same weak source. Read the objections and check important claims against evidence.
Why use other models instead of asking one to critique itself?
Self-critique can help. Other models and specific review roles provide additional responses to compare. The format determines whether reviewers begin independently or see earlier answers.
Does this replace human review?
No. Use the findings to improve a draft or prepare questions for a responsible reviewer. The models do not approve a decision or verify every claim.
Bring your next draft.
Create a free account to draft or import your work. Choose the reviewers and settings before starting a review.