---
Title: "What Is an AI Hallucination? Causes and Prevention"
Url: "https://computer.io/glossary/ai-hallucination"
Topic: "AI safety"
Description: "An AI hallucination is when a model produces false or made-up information that sounds plausible. Why hallucinations happen, examples and how to reduce them."
---

# What is an AI hallucination?

An AI hallucination is when a language model generates information that sounds confident and plausible but is false, unsupported or made up, such as a nonexistent policy, wrong number or fake citation. It happens because models predict likely text rather than look up facts.

## Examples in support

| Hallucination | Risk |
| --- | --- |
| Invented refund policy | Financial and legal |
| Wrong setup steps | Frustration |
| Fake feature | Lost trust |

## How to reduce them

Ground answers in retrieved sources, require citations, let the AI say "I don't know", restrict to approved knowledge, evaluate regularly and keep humans in the loop for risky answers.

## FAQ

### Why do AI models hallucinate?

They generate likely text, and gaps or ambiguity lead to invented details.

### Can hallucinations be eliminated?

Not completely, but grounding and guardrails greatly reduce them.

### How do you detect hallucinations?

Compare outputs to sources, use evals and sample reviews.

## Related terms

- [AI guardrails](https://computer.io/glossary/ai-guardrails)
- [Retrieval-augmented generation (RAG)](https://computer.io/glossary/retrieval-augmented-generation)
- [AI evals](https://computer.io/glossary/ai-evals)
- [How do you reduce AI hallucinations in customer support?](https://computer.io/answers/how-to-reduce-ai-hallucinations-in-customer-support)

## How Computer helps

- [How computer works](https://computer.io/how-computer-works)

## Further reading

- [Knowledge graphs vs rag enterprise ai](https://devrev.ai/blog/knowledge-graphs-vs-rag-enterprise-ai)