A hallucination is an AI answer stated confidently but not supported by the model's training data or retrieved sources — including invented products, wrong prices, and misattributed features.
In brand contexts, hallucinations cut both ways: a model can invent a feature you don't have, or confidently recommend a product that was discontinued years ago. Grounded answers with citations hallucinate less, which is one reason retrieval-backed engines dominate buying queries.
For visibility measurement, hallucination is a reason to keep receipts. When a reported answer looks wrong, the stored response settles what the model actually said — and gives the brand something concrete to correct at the source.