---
title: "Hallucination"
slug: "hallucination"
category: "glossary"
tags: ["glossary", "ai-agents", "hallucination", "evaluation", "grounding"]
status: "stable"
last_updated: 2026-05-14
summary: "A hallucination is a confident model output that is not grounded in the input or in verifiable facts: a fabricated citation, a wrong date, a non-existent API."
related:
  [
    "[[ai-agents/evaluation]]",
    "[[glossary/retrieval-augmented-generation]]",
    "[[ai-agents/rag]]",
    "[[glossary/llm-as-judge]]",
    "[[ai-agents/rag-citations]]",
    "[[ai-agents/structured-output]]",
  ]
---

## Overview

This page is the atomic definition. The mitigation tooling lives at [[ai-agents/evaluation]] and [[ai-agents/rag]].

## Definition

A hallucination is a confident model output that is not grounded in the input or in verifiable facts. Examples: a fabricated citation, a wrong historical date, a non-existent API method, an invented function signature, or a confident contradiction of the source document the model was told to summarize. Hallucinations come from the model's autoregressive nature; it samples the most probable next token, not the most truthful one. Mitigations include grounding with [[glossary/retrieval-augmented-generation]], requiring citations, constraining output with [[glossary/structured-output]], and evaluation with a [[glossary/golden-set]].

## When it applies

Plan for hallucinations in any model output that an end user might trust. Stakes scale with domain: medical, legal, and financial outputs need stricter grounding and verification than chitchat.

## Example

A model asked to summarize a paper cites "Smith et al. (2023)" with a plausible title and journal, but the paper does not exist. The user copies the citation into a literature review; the error is caught only on submission.

## Related concepts

- [[glossary/retrieval-augmented-generation]] - the primary mitigation for factual hallucinations.
- [[ai-agents/evaluation]] - the discipline that catches hallucinations before they ship.
- [[glossary/llm-as-judge]] - one technique for scoring groundedness at scale.
- [[ai-agents/rag-citations]] - citation requirements that make hallucinations visible.
- [[glossary/structured-output]] - schemas that constrain the space of fabricable outputs.

## Citing this term

> See [[glossary/hallucination|Hallucination]] (llmbestpractices.com/glossary/hallucination).

## Related

- [[ai-agents/evaluation]]
- [[ai-agents/rag]]
- [[ai-agents/rag-citations]]
- [[glossary/retrieval-augmented-generation]]
- [[glossary/golden-set]]