Hallucination
A hallucination is an answer from an AI model that sounds confident and is wrong: an invented fact, a source that doesn't exist, a number with nothing behind it. It happens because language models generate likely text rather than look facts up.
What it means
Hallucinations can't be removed entirely, only made rarer and less harmful. Grounding the model in real documents, testing it against an evaluation set, letting it say it doesn't know, and putting a person in the loop where mistakes are costly all reduce the damage.
The design question is what happens when the model is wrong. A product that shows its sources, flags uncertainty and hands over to a person gracefully can survive a wrong answer. One that presents every answer with the same confidence can't.
How we use it
Our method When we design an AI product, we design for the moment the model is wrong or unsure: where its answers appear, how it signals doubt, and who it hands over to. It's part of the Design step in Decide · Design · Ship.
Related terms
Published 28 September 2026. All terms