In plain English
A hallucination is when a model generates information that sounds confident and plausible but is factually wrong or entirely fabricated, such as a made-up citation, an incorrect date, or a feature that does not exist. It happens because language models predict the most likely next words, not because they check facts. The fluency is exactly what makes hallucinations easy to miss.
Why it matters
Because hallucinations look just like correct answers, they are dangerous in anything factual, legal, or medical. This is why human review of AI output is essential.
A worked example
Asked for sources, a model may produce a perfectly formatted reference, complete with authors and a journal, for a paper that was never written.
Common confusion
Hallucination is not a bug awaiting a patch. It is a structural feature of how these models generate text, so it can be reduced but not fully removed.