Hallucination
Definition
An AI hallucination is a confident, fluent, wrong answer. The model produces output that is grammatically correct and contextually plausible, but factually false. There is no internal signal that anything is wrong.
Why it happens
AI hallucinations happen because language models generate answers by predicting patterns in data, not by checking whether information is true. When the model lacks enough context or certainty, it can confidently create incorrect or completely made-up information.
Examples
- A lawyer used AI in court filings and it invented fake legal cases that never existed
- Google’s AI once said the James Webb telescope took the first image of a planet, which is false
- A chatbot in customer support made up company refund policies, leading to real complaints and legal disputes
- AI systems have fabricated scientific papers or citations, which look real but cannot be found anywhere
- Some AI tools in media generated fake reading lists with books that do not exist
- AI search systems have suggested dangerous nonsense like adding glue to food due to misreading web content.
Read: 8 AI hallucinations examples
Detection techniques
| Technique | How |
|---|---|
| Cross-reference sources | Ask the AI to cite, then verify citations independently |
| Ask for uncertainty | Ask the model to separate: what it is confident about, what it is guessing, what could be wrong or missing |
| Provide context | Paste the source document and ask the AI to answer from it only |
| Rephrase and re-ask | Ask the same question differently because inconsistency exposes hallucination |