AI glossary · Beginner · Language models · Safety & alignment
Hallucination
Hallucination is when an AI confidently produces content that is untrue or unsupported as if it were fact.
About this term
- Level
- Beginner · Core ideas for first-time readers
- Category
- Basics
- Added
- 2026-10-07
How it works
Language models produce text based on 'what is likely to come next' rather than 'what is true.' So when the information they learned is insufficient or the question is ambiguous, they can produce answers that look plausible in form but are wrong in content. Typical examples are nonexistent paper titles, incorrect numbers and made-up quotations.
Why it matters
Hallucinations look natural on the surface, so users find them hard to notice. In fields where accuracy matters, such as medicine, law and finance, they can lead to serious harm. That is why reducing hallucination is considered a key challenge when bringing AI into real work.
Things to know
Hallucination can be reduced by methods such as RAG, which finds reliable material and presents it alongside the answer, or by having the model show its sources. However, the general view is that it is hard to eliminate completely. It is a good idea to always double-check important information against the original source.
Example
When asked for references on a particular topic, a chatbot may point to papers whose authors and titles sound plausible but that do not actually exist. This kind of hallucination is also why many AI services display notices such as 'AI can make mistakes' on screen.