How to check an answer that sounds right.
AI is fluent whether it is right or wrong. Five checks catch most of the problems, and take minutes.
Why it sounds right
A language model predicts plausible text. It is fluent when it is correct and just as fluent when it is not. Confidence in the wording tells you nothing about whether the answer is true.
That is not a reason to avoid it. It is a reason to check, in a way that is cheap enough to do every time.
Five checks
- 01
Ask for the source
Ask where it comes from, then open the source. If it cannot point to one, treat the answer as a guess.
- 02
Check one fact yourself
Pick the claim that matters most and confirm it somewhere you trust. If it is wrong, assume more are.
- 03
Test an edge
Try a case where the answer should change: a different country, a small number, an empty input.
- 04
Ask for the other side
Ask it to argue the opposite. Gaps often show up when it has to defend the other view.
- 05
Name who is accountable
Decide who signs the work. It is a person, and the person has to understand what they are signing.
A habit worth keeping
Write down what you checked and what you found, even in two lines. It takes seconds, it makes you actually do the checks, and it gives anyone who relies on your work a way to trust it.
Practise it on a real task in a Labs project.
Start with a free project in Labs, or build your first AI solutions in the free Masterclass.
- Learn by building, from the first session
- Free to start
Practise it on a real task in a Labs project.
AI is fluent whether it is right or wrong. Five checks catch most of the problems, and take minutes.