Check AI sources with the CRAP test

How to run the check

Paste the AI answer into the chat and give it a precise review task: "Check every factual claim and source in this answer with the CRAP test. Create a table with claim, source, Currency, Reliability, Authority, Purpose/Point of View, warning signs, and verdict. Mark missing or inaccessible evidence clearly and prefer primary sources."

Do not stop at a general link review. Ask the AI to connect each concrete claim to a source. Anything without evidence gets its own row with the verdict "not supported".

The four questions

C means Currency: Is the source recent enough for this claim? Dates matter a lot for law, medicine, prices, software, and AI models.

R means Reliability: Can you trace where the information came from? Strong signals include methods, original data, clear evidence, and verifiable citations.

A means Authority: Who is behind the source? Check authorship, publisher, expertise, and possible conflicts of interest.

P means Purpose or Point of View: Why was the source published? Advertising, opinion, lobbying, or a sales funnel are not automatic deal-breakers, but they are warning signs.

A better prompt

Weak: "Check your sources."

Better: "List each claim separately, name the best available primary source, and score it with CRAP. End with three short notes: what is well supported, what remains uncertain, and what I must verify manually."

That turns the answer into a verification checklist instead of a comforting paragraph.

Good to know

CRAP and CRAAP are related, but not identical. CRAAP adds Relevance and uses Accuracy instead of Reliability. For quick AI checks, CRAP is often easier to remember and apply.

Still, an AI verdict is not proof. Open the most important links yourself, check dates, authorship, and context, and do not rely on a score for critical decisions.