Fact-Check an AI Answer Claim by Claim Before You Rely on It

You get a table that splits a pasted AI answer into separate claims, rates each one, and says exactly what to verify and where, plus a cleaned-up summary.

Prompt

Prompt
Act as a careful fact-checker who reviews AI-generated text before anyone relies on it.

I will paste an answer from an AI chat tool right after this message. Check it for factual errors, weak claims and missing caveats.

Rules:
1. Split the answer into individual, checkable claims, one per row, quoting or closely paraphrasing the original wording.
2. Rate each claim Verified-likely, Uncertain or Doubtful, with a one-sentence reason based on what you know, not on how confident the original sounds.
3. For every claim not rated Verified-likely, state exactly what to check and the type of source to check it in (official agency site, peer-reviewed database, primary document). Never invent titles, URLs, authors, quotes or statistics.
4. Flag vague wording such as "studies show" or "experts agree", and any number given without a date or source.
5. If a claim depends on recent events or your knowledge may be outdated, say so instead of guessing.

Output format:
1. Claim table with columns: #, Claim, Rating, Reason, What to check and where
2. Top 3 risks: the claims that would cause the most harm if wrong
3. Corrected summary: a short rewrite keeping well-supported points and marking uncertain ones

End with one line on what you could not assess and why. If the topic is health, money or legal, add one line that this is not professional advice and when to see a professional.

Example output

Example for: a student checking a ChatGPT answer about the history of the QWERTY keyboard before using it in a class presentation.

1. Claim Table

#ClaimRatingReasonWhat to check and where
1"QWERTY was developed by Christopher Latham Sholes in the 1870s."Verified-likelySholes and collaborators are widely documented as developing the layout for their typewriter.No check needed; a museum or encyclopedia entry confirms it.
2"It was designed to slow typists down so the keys wouldn't jam."UncertainThe anti-jamming story is common, but historians dispute the "slow typists down" framing.Look for academic history of technology articles on typewriter key arrangement in a journal database.
3"Remington bought the design in 1873."Verified-likelyThe sale to E. Remington and Sons around 1873 is consistently reported.Optional: confirm the year in a typewriter museum archive.
4"The Dvorak layout is 40% faster than QWERTY."DoubtfulNo source given, and comparative studies are contested; speed gains are usually reported as small or disputed.Search economics and ergonomics papers on Dvorak vs. QWERTY; note who funded each study.
5"Studies show most typists use only the home row."DoubtfulVague "studies show" with no number or date, and the claim is implausible as worded.Ask: which study, what year, what sample? Drop it if none can be found.
6"QWERTY is used in over 100 countries today."UncertainPlausible, but the number has no source and regional variants (AZERTY, QWERTZ) complicate it.Check a keyboard layout standards reference or OS documentation for layout lists.

2. Top 3 Risks

  1. Claim 4 – A specific percentage stated as fact is the most likely to be quoted and challenged.
  2. Claim 2 – A popular myth; presenting it as settled could undermine credibility.
  3. Claim 5 – Unsourced and vague; easy for an audience to question.

3. Corrected Summary

The QWERTY layout was developed by Christopher Latham Sholes and colleagues in the 1870s, and the design was sold to Remington around 1873. One common explanation is that the arrangement reduced typebar jams, though historians disagree on the details (uncertain). Alternative layouts such as Dvorak exist, but claims that they are much faster are disputed (uncertain). QWERTY and its regional variants are used in many countries.

Could not assess: the exact country count, since it depends on current OS and standards data I cannot confirm.

Written by AI from this prompt. Your results will vary.

Tips and common edits

Why this works

  • Splitting the answer into single claims stops one true fact from making the whole paragraph look reliable.
  • Rating on evidence rather than tone counters the confident style AI answers use even when wrong.
  • Asking for source types instead of named sources prevents the checker from inventing fake citations.

Use before you quote, share or act on an AI answer. Paste the full answer, not a summary. Works in ChatGPT, Claude or Gemini; for best results, run it in a different tool than the one that wrote the answer.

Common edits

  • Add a column for how important each claim is to your decision so you check the critical ones first.
  • Ask for a version of the corrected summary written for a specific audience, like a manager or a class.

Prompt details

By
Prompts-GPT editors
Published

Did this prompt work?

Keep exploring

More in Personal & Life