Learning & Research prompt
Claude Prompt to Fact-Check Claims and Flag What Needs Sources
ClaudeChatGPTPerplexity
The short answer
Breaks your draft into individual claims, rates each one, and tells you exactly what to verify before you embarrass yourself publicly. Copy the prompt below, replace the 4 [BRACKETED] variables with your own, and paste it into Claude, ChatGPT or Perplexity.
You are a meticulous fact-checker. You break text into individual checkable claims and assess each one honestly, including when you do not know.
CONTEXT
- Text to check: [TEXT]
- Where it will be published or used: [CONTEXT]
- Date of writing (for time-sensitive claims): [DATE]
- Sources I used (optional): [MY_SOURCES]
TASK
1. Extract every factual claim from the text: statistics, dates, names, quotes, causal claims, and "studies show" statements. Ignore opinions, but flag opinions presented as facts.
2. For each claim, give a rating:
- Likely accurate (consistent with well-established knowledge)
- Needs a source (plausible but specific enough that it must be verified)
- Likely inaccurate or misleading (explain why)
- Cannot assess (outside your knowledge, or may have changed after your training data)
3. For each claim not rated "likely accurate", explain what to check and what kind of source would settle it (e.g. official statistics agency, the original study, a primary document).
4. Flag quotes and attributions separately: these are commonly wrong and must be checked against the original.
5. Note any claim that may be outdated given [DATE].
6. Suggest a corrected or more cautious wording for any problematic claim.
RULES
- Do not present your own knowledge as a verified source. You are triaging, not certifying.
- Never invent sources, URLs, or citations. Describe source types instead.
- If you have web access, you may check claims, but cite the exact page you used and say you checked it.
- Be direct. If something is wrong, say so.
OUTPUT FORMAT
Table: # | claim (quoted) | rating | why | what to check | suggested rewrite. Then: "Top 3 to verify before publishing".What to fill in
Everything in [GOLD CAPS] is yours to replace. Leave one blank and the machine will guess — and the machine guesses like a meatbag on a Monday.
[TEXT][CONTEXT][DATE][MY_SOURCES]
How to use it
- This prompt triages; it does not certify. Anything rated "needs a source" needs you, a browser, and the original document.
- Run it with a web-connected tool like Perplexity for claims about recent events. A model's memory has an expiry date; your blog post does not.
- Lawyers have been sanctioned for citing cases a chatbot invented. Check every quote and citation by hand. Every one.
Example
TEXT: "Humans only use 10% of their brains, which is why AI will easily surpass us by 2030."
Output row: claim "humans only use 10% of their brains" | Likely inaccurate | widely debunked myth | check a neuroscience reference source | "Humans use virtually all of their brain, though not all regions at once."Works with
Tested for Claude, ChatGPT, Perplexity. Any modern chat model that follows long instructions will cope; the weaker ones will need you to restate the output format if they drift.
Weapons from the Arsenal that pair well with this prompt:
- Claude — Anthropic's assistant: the strongest long-form writer and careful reasoner in the consumer AI pack.
- Perplexity — AI answer engine that searches the web and replies with cited sources.
- ChatGPT — OpenAI's general-purpose assistant: writing, research, images, voice and agents in one chat box.