Writing
AI Prompt for Fact-Check Review
Publish with confidence: this AI prompt reviews drafts for checkable claims — flagging what to verify before you hit publish. Works with top AI tools.
The prompt
You are a fact-checker who protects writers from embarrassing errors. Context: I need the following draft fact-check reviewed before publishing for [AUDIENCE] on [PUBLICATION]. Here is the draft: [DRAFT_TEXT] Task: Review the draft and return: (1) a numbered list of every checkable claim (statistics, dates, names, quotes, historical facts, scientific claims), each with: the claim quoted, a verdict — [LIKELY TRUE] / [NEEDS VERIFICATION] / [LIKELY FALSE] / [UNVERIFIABLE AS STATED] — and what to check or how to rephrase safely. (2) A "safe rephrases" section offering cautious wording for the shakiest claims. (3) An overall risk rating: low / medium / high, with one-line reasoning. Format: Claim list first (quote + verdict + action), then safe rephrases, then risk rating. Constraints: Be strict: when in doubt, flag it — false confidence is the enemy. Distinguish widely-known facts (no citation needed) from specific claims (verify). Never "verify" from memory alone; tell me WHAT source type to check (official stats, primary source, expert). Don't rewrite the draft — flag and suggest. Quotes must be exact or marked [PARAPHRASE].
Variables to fill in
Replace each [variable] in the prompt with your own details:
[AUDIENCE] [PUBLICATION] [DRAFT_TEXT]
Example input
Example: AUDIENCE = 'my blog readers', PUBLICATION = 'my newsletter', DRAFT_TEXT = '<paste your draft>'.
How to use this prompt
Run this BEFORE final proofreading. Work through every [NEEDS VERIFICATION] flag with real sources, apply safe rephrases where you can't verify, then publish.
Tips for the best results
- Statistics decay: even true stats need a date ('as of 2024').
- Quotes are the highest-risk element — verify wording, attribution, and context.
- A 'medium' risk rating means stop and verify; don't publish on hope.
Frequently asked questions
Can AI verify facts itself?
No — it flags WHAT to check. Verification means consulting real sources yourself.
What if I can't verify a claim?
Use the safe rephrase ('many experts suggest...') or cut it. Unverifiable claims are liabilities.