Hallucination
A hallucination is output from an AI model that sounds confident but is false, unsupported or invented. Common forms include a source that does not exist, a quote nobody said, a statistic with no origin, or a wrong fact about a company, product or person. It happens because a large language model generates the most plausible next token, not a verified statement.
Why do hallucinations matter for brands?
When a buyer asks an answer engine about a company, a hallucination becomes the buyer's first impression. A wrong founding date, a fabricated lawsuit or a missing product line can spread because the answer reads as authoritative and there is no link to check. Gaps in a brand's entity profile make this more likely, since the model fills silence with a guess.
How can a communications team reduce the risk?
No team can prevent a model from hallucinating, but several practices lower the odds and speed up correction:
- Publish clear, consistent facts on owned pages so retrieval has accurate material, which supports grounding.
- Keep facts identical across the open web to avoid a fractured entity profile.
- Track what engines say with citation monitoring and run an AI visibility audit to find errors.
- Check the knowledge cutoff, because recent news may simply be absent from a model.
Deliberate manipulation is a related threat, covered under model poisoning. For the correction workflow, see how to correct outdated information across search and AI answers. Every AI-written statement should be verified by a person before it is published or cited.