Lesson 6 of 6
Check its work
6 min read
You don't have to choose between trusting AI blindly and not using it at all. There's a middle path the pros live by: let it draft, then check the parts that matter — in seconds, not hours.
A confident answer is a draft
Everything so far points to one habit: don't take a checkable claim on faith. The good news is that catching a mistake is usually much faster than making the answer was. A single well-aimed move — asking for the source, checking it a second way, or re-asking — is often enough. That habit has a name: verification.
Notice what didn't work: trusting the confident tone. The claim only cracks when you check it against something outside the model.
A small toolkit for everyday checks
You don't need all of these every time — pick the one that fits the claim. For a fact, ask for the source and actually open it. For a number, redo it yourself or in a calculator. For anything important, re-ask in a fresh chat and see if the answer holds. Consistency and a real source are your two best tells.
Match the check to the risk. A throwaway brainstorm needs none; a medical dose, a legal clause, or a number you'll act on deserves a hard second look.
Your verification toolkit
- —Ask for the source — then open it and confirm it's real and says that.
- —Redo numbers yourself, or in a calculator or spreadsheet.
- —Re-ask in a fresh chat; if the answer changes, trust none of them yet.
- —Scale the effort to the stakes — more checking where a mistake would hurt.
An AI gives you a striking statistic for a work report, with a named source. The fastest worthwhile check is:
Continue in the app
Take the whole When AI Gets It Wrong course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.