Can an AI Say I Do Not Know? – free checklist (episode 20)

Why it matters

An AI can give a wrong answer in a calm, sure voice. A made-up phone number looks just like a real one. These five habits help you catch it.

Your five-step check

0 of 5 done

Sure is not proofAn AI can write an answer that is wrong but sounds fine. Experts call this a hallucination. How sure it sounds tells you nothing.Try: Ask yourself: did it show any proof, or just sound sure?
Let it say I don't knowTell the AI it is OK to admit when it is not sure. One AI maker's guide says this can cut false answers a lot.Try: If you are not sure, say so. Do not guess.
Ask for the sourceAsk where the answer came from, then check the source is real. If it cannot point to anything, treat it as a guess. This is general good practice.Try: Where did that come from? Show me the exact words.
Watch the believable onesPhone numbers, names, dates and quotes are easy to make up and easy to believe. Be extra careful with these. This is our own rule of thumb.
Check what mattersThese tricks help a lot, but they do not fix everything. If it matters, check it yourself.Try: Look it up on the official page.

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

Rule of thumb

sounding sure is not proof. Allow 'I don't know', ask for the source, check what matters.

Sources: Anthropic docs, 'Reduce hallucinations' (platform.claude.com); Kalai, Nachum, Vempala and Zhang, 'Why Language Models Hallucinate' (arXiv 2509.04664). The robot test is our own staged story. Steps 3 and 4 are general good practice.

Episode 20: Spot an AI bluff

Why it matters

Your five-step check

More like this

A new free sheet every week