What is Hallucination (AI)

Definition

An AI hallucination is when a generative model produces output that is fluent and confident but factually incorrect or fabricated, presenting invented details, sources, or claims as if they were true because the model predicts plausible text rather than verifying facts.

« Back to Glossary Index

Understanding hallucination helps teams design safeguards against false outputs
Recognising it guides use of grounding techniques like RAG and citations
Awareness drives human-review processes for high-stakes AI outputs
Measuring it enables monitoring and reduction over time

Real World Example

A chatbot confidently cites a non-existent court case when asked for legal precedent; recognising this hallucination, the team adds retrieval grounding and a citation check so answers must trace to real source documents.

FAQs

Why do AI models hallucinate?

Because they generate the most plausible-sounding text from learned patterns rather than checking facts, they can produce convincing but false output.

How can hallucinations be reduced?

Grounding the model with retrieved sources, requiring citations, constraining outputs, and adding human review all help reduce hallucination.

Are hallucinations always obvious?

No, they are often fluent and confident, which is what makes them risky and why verification matters for important uses.