A hallucination is an AI answer stated confidently but not supported by the model's training data or retrieved sources. Examples include 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. It also gives the brand something concrete to correct at the source.