# What is Hallucination?

> When an AI model states something confident and wrong.

Canonical: https://www.decimal.app/glossary/hallucination · Last reviewed: 2026-09-23

A hallucination is output from a language model that sounds plausible but isn't supported by any source, such as a setting that doesn't exist or an API parameter the product never had.

## Why it matters

In support, a hallucinated answer is worse than no answer: the customer acts on it, it fails, and trust drops for every answer after it.

## Best practices

1. Ground and cite. Require a source for every factual claim.
2. Check against the live system. Docs go stale; code and config are current.
3. Review before sending at first. Run in draft mode until accuracy is proven.

## How Decimal helps

Decimal grounds answers in your real systems and shows its evidence, so reviewers can verify a response before it reaches a customer.

## Related terms

- [Grounding](https://www.decimal.app/glossary/grounding): Tying an AI's answer to specific source material it can cite.
- [Human-in-the-Loop](https://www.decimal.app/glossary/human-in-the-loop): A design where people review or approve an AI's work at key points.
- [Retrieval-Augmented Generation (RAG)](https://www.decimal.app/glossary/retrieval-augmented-generation): Having an AI model look up relevant documents before generating an answer.

---

Decimal · Book a demo: https://cal.com/decimal-demo/30m · Markdown index: https://www.decimal.app/llms.txt