What is an AI hallucination?
An AI hallucination is when a model produces an answer that is fluent, confident, and specific, but factually wrong or entirely invented. It happens because language models are built to produce plausible text, not to check whether that text is true. The model has no internal alarm that fires when it crosses from fact into fiction, which is why a hallucination reads exactly like a correct answer. That similarity, not the error itself, is what makes hallucinations dangerous.
Why they happen
A large language model predicts the next word based on patterns in its training data. When it has seen enough real information about a topic, that prediction lands on the truth. When it has not, it still produces a fluent, confident answer, because producing fluent, confident text is the one thing it always does. It fills the gap with something plausible rather than saying "I do not know." The result is a hallucination.
Why they are hard to catch
The tell people expect, a hesitant or garbled answer, is not there. Hallucinations are often the most confident, most detailed, best-formatted answers a model gives. In one run on this site, a model asked about a line of code invented a full medical alert company, complete with pricing and citation markers. Nothing in its tone flagged the problem. Fluency is not evidence. Sometimes fluency is the warning.
What actually works
You do not detect a hallucination by reading harder. You detect it structurally: ask several models independently, make every claim name a source you can open, and keep only what survives. A hallucination has no real source, so it fails that test. This is the Council Method, and there is a page on exactly how to stop AI from hallucinating using it. The full treatment is in Let the AI Be Smart.
Go deeper: this site's hub page on the Council Method is the full definition. Related questions: How do I stop AI from hallucinating?, Why do AI models give wrong answers so confidently?, How do I know if an AI answer is correct?.