Skip to content
TheBestSuperintelligence

Skill · Verify

Fact-check an AI answer

Paste any AI-generated answer and get every factual claim checked against sources, with fabricated references and unsupported numbers flagged.

The skill#

Fact-check an AI answerIndependent fact-checker
Act as an independent fact-checker.

Review the following AI-generated answer: [PASTE YOUR AI ANSWER]

For every important factual claim:

1. Identify the specific claim.
2. Find an authoritative supporting source.
3. Verify that the source actually supports the claim.
4. Identify contradictory evidence.
5. Flag unsupported statistics and fabricated references.
6. Separate verified facts from assumptions and interpretations.

Do not assume the original answer is correct.
Do not invent missing evidence.

Return a table containing:
Claim | Evidence | Source | Verification Status | Required Correction

How to use it#

  1. Copy the skill and paste it into ChatGPT, Claude or Gemini, ideally one with web search switched on.
  2. Replace [PASTE YOUR AI ANSWER] with the full text you want checked.
  3. Open the sources it cites yourself for anything you will publish or act on.

Example#

Paste a 300-word AI summary of a company's results. You get back a table of 8–12 claims, each marked Verified, Unsupported or Contradicted, with the correction for each wrong figure and a flag on any reference that does not exist.

Quick and strict versions#

Use the quick version for everyday checks and the strict version when the result matters.

Quick versionFact-checker
Act as a fact-checker. List the five claims in this text most likely to be wrong, say why, and say what source would settle each one. Do not invent sources.

Text: [PASTE YOUR TEXT]
Strict versionIndependent fact-checker working to publication standard
Act as an independent fact-checker working to publication standard.

Text to check: [PASTE YOUR TEXT]

For every factual claim, including numbers, dates, names and quotes:
1. Quote the claim exactly.
2. Name the primary source that would confirm it (official document, original study, dataset), not a secondary summary.
3. If you cannot access that source, mark the claim "Unverified", never "Verified".
4. Check every reference: does it exist, and does it say what the text says it says?
5. Mark each claim Verified, Partly supported, Unsupported, Contradicted or Unverified.

Rules: do not assume the text is correct; do not invent sources, URLs or page numbers; say "I could not confirm" plainly.

Return: Claim | Status | Source | What the source actually says | Correction

Use it as an agent skill#

Save this as SKILL.md in a folder named after the skill, and agents that support skills (Claude, Claude Code and other skill-aware agents) can load it automatically when the task fits.

markdown
---
name: fact-check-ai-answer
description: Use when the user wants an AI-generated answer, article or report fact-checked, or asks whether its claims and references are real.
---

# Fact-check an AI answer

Fill each [BRACKETED] field from the conversation. If one is missing, ask the user for it before starting.

Act as an independent fact-checker.

Review the following AI-generated answer: [PASTE YOUR AI ANSWER]

For every important factual claim:

1. Identify the specific claim.
2. Find an authoritative supporting source.
3. Verify that the source actually supports the claim.
4. Identify contradictory evidence.
5. Flag unsupported statistics and fabricated references.
6. Separate verified facts from assumptions and interpretations.

Do not assume the original answer is correct.
Do not invent missing evidence.

Return a table containing:
Claim | Evidence | Source | Verification Status | Required Correction

Download SKILL.md

Frequently asked questions

Can AI reliably fact-check AI?

It catches many errors, especially with web search on, but it can also be wrong. Treat its table as a checklist and open the sources yourself for anything important.

Which AI is best for fact-checking?

One with live web search and visible citations. The skill works in ChatGPT, Claude, Gemini and Perplexity; results depend on what each can access.

Why ask it to separate facts from interpretations?

Many AI errors are not false facts but opinions stated as facts. Separating them shows what can be checked and what is judgment.