AI sounds just as confident when it's wrong. It isn't lying — it fills gaps with a believable guess. Here's how to catch that before you send it, print it, or trust it with something that matters.
Where it slips most
- Exact facts — names, dates, numbers, prices, quotes.
- Anything recent — it may not know past its cut-off date.
- Legal, medical, tax or safety specifics.
- Sources it cites — links and studies it can invent whole.
- Your own business — details it was never actually given.
Tells it's guessing
- Very specific, but no source you can check.
- Total confidence on a narrow or obscure topic.
- Facts you never gave it, stated as certain.
- Ask again and the answer quietly changes.
- Suspiciously round or too-tidy numbers.
The 30-second check
1Ask it: "How do you know that? What might be wrong here?"
2Ask the same thing in a fresh chat — do the answers match?
3Check one hard fact yourself: a name, a price, a date.
4If being wrong would cost you, confirm it at the source.
Head it off in the prompt
- "If you're not sure, say so — don't guess."
- "Use only what I gave you. Add no new facts."
- "List anything you assumed or made up."
- Paste the real material instead of asking it to remember.
say if unsureno new factsshow assumptions
The one rule to keep
Treat AI like a fast, confident intern: great for a first draft, never the final word. Let it create — then you check anything a mistake would actually cost you.