Lesson 2 of 6
Communicating confidence & error
6 min read
Your AI states a wrong fact in exactly the same confident tone as a right one. If your interface passes that tone straight to the user, you've shipped the bug.
Confidence is not correctness
An AI's fluent, self-assured tone has little to do with its actual [calibration](glossary://calibration) — how well its confidence matches its accuracy. It sounds equally certain whether it's right or making things up. Tap through these answers: the surest-sounding one is wrong, and a hesitant one is right.
The confidence a user hears is tone, not truth. Your UI has to add the signal the model's voice leaves out.
Show uncertainty, don't hide it
So design the honesty in. Surface a source or citation the user can open, flag low-confidence output visibly, and make verification one tap away instead of an act of faith. A quiet 'double-check this' beats a slick answer that hides how shaky it is.
Good AI UX communicates how sure alongside what — sources, confidence cues, and an easy path to verify.
Resist fake precision. A made-up '94% confident' badge is worse than none — show cues you can stand behind (a real source, a 'this may be wrong' flag), not a number the model can't actually justify.
The shape of it
- —AI sounds equally confident when it's right and when it's wrong.
- —Surface sources, flag low confidence, and make verifying one tap away.
- —Avoid fake precision — show cues you can actually stand behind.
Your summarizer sometimes invents a detail. What's the most honest UX response?
Continue in the app
Take the whole Designing AI UX (for uncertainty) course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.