# Hallucination

> When AI states something false as if it were true. It predicts fluent text, so a wrong answer can still sound confident.

**Category:** Core concept, Safety

**A hallucination is when AI says something untrue with a straight face.** Because a [large language model](/glossary#large-language-model) works by predicting fluent text, not by looking facts up, it can produce a clean, confident answer that is simply wrong: a made-up quote, a fake statistic, a source that does not exist.

The fix is a habit, not a setting: **treat AI as a fast drafter, not an oracle.** Check anything factual, and switch on [web search](/glossary#web-search) for current facts so Claude can cite real sources.

## At work

Claude hands you a confident stat for a client deck. Before you paste it in, check the source. A fluent number can still be invented.

## Often confused with

A simple mistake or a lie. The model is not lying; it has no idea it is wrong, because it predicts plausible text rather than checking truth.

## Related terms

- Large language model
- Generative AI
- Web search

## See it in a lesson

[How generative AI works](/en/learn/foundations/how-generative-ai-works)

## Further reading

- [What are AI hallucinations? (IBM)](https://www.ibm.com/think/topics/ai-hallucinations): What hallucinations are and why they happen.
- [Hallucination (artificial intelligence) (Wikipedia)](https://en.wikipedia.org/wiki/Hallucination_%28artificial_intelligence%29): A well-sourced overview with real examples.
- [The hidden incentives driving AI hallucinations (IBM)](https://www.ibm.com/think/news/hidden-incentives-driving-ai-hallucinations): A deeper look at why models guess.

---

Last updated: 24/07/2026  
Source: Rollo Academy glossary