In plain language
A model can produce a fluent answer that invents details because it predicts plausible output rather than checking truth by default.
Example
An AI assistant cites a policy section that does not exist in the company handbook.
Related terms
artificial intelligence · large language model · Prompt
Common misconception
Hallucinations are not eliminated merely by asking the model to be accurate or by lowering creativity settings.
Implementation guidance
Ground outputs in approved sources, require citations that can be checked, validate structured fields, measure error rates, and use human review for high-impact work.