How to Check an AI Answer – free checklist (episode 3)

Why you need it

AI can sound confident and still be wrong. Researchers argue that AI models are often graded like a test that rewards guessing over saying "I don't know". So a good habit is to make the AI show its work, and to check what matters.

The 5 steps

0 of 5 done

Give it your document, and keep it insidePaste the text you want it to use, and tell it not to use anything else.Try: Answer using only the text below. Do not use outside knowledge.
Let it say "I don't know"Give the AI permission to admit it is unsure, so it does not feel it has to fill every gap.Try: If the text does not say, or you are unsure, reply: I don't know.
Ask for the exact quoteMake each claim traceable. A claim with no supporting quote should be removed.Try: For each claim, give the exact quote that supports it. If you can't find one, remove the claim.
Ask twice and compareRun the same question more than once. If the answers disagree, treat that as a warning sign.Try: Ask the same question in a fresh chat and compare the two answers.
Check the facts that matter yourselfNumbers, names, dates, links, and anything where a mistake would cost you. Look them up in a trusted source.General good practice. No tip removes mistakes completely.

Tap a step when you have done it. Your ticks stay on this device.

Rule of thumb

Quick rule: the more it matters, the more you check. A wrong restaurant tip is fine. A wrong dose, date, price, or legal detail is not.

Sources: Anthropic docs, "Reduce hallucinations"; Kalai, Nachum, Vempala, Zhang (2025), "Why Language Models Hallucinate", arXiv:2509.04664. Prompts may need small wording changes in your AI app.

Episode 3: How to check an AI answer

Why you need it

The 5 steps

Can an AI Say I Do Not Know?

More like this

A new free sheet every week