1. Home
  2. Glossary
  3. Hallucination
AI glossary · Safety, ethics & policy

Hallucination

Hallucination: A hallucination is when an AI model produces information that is false or fabricated but presents it confidently, such as an invented statistic, a citation to a paper that does not exist, or a made-up product feature.

Language models generate text that is likely to follow the prompt, not text that is verified true. When the training data thins out, when a question is phrased as if a fact exists, or when the model is asked for specifics it does not have, it fills the gap with something plausible. The output looks exactly like a correct answer.

The most common workplace hallucinations are invented citations and case law, wrong numbers in a summary of a real document, fake quotes attributed to real people, and confident claims about a tool's features or a company's policies. Models also sometimes agree with a false premise you put in the question.

You can reduce hallucinations but not eliminate them. Give the model the source material and tell it to answer only from that text. Ask it to quote the exact passage that supports each claim. Use tools that search the web or your documents and show citations. Lower the stakes by using AI for drafts and analysis you will check, not for facts you will not.

Example at work

A marketing writer asks an assistant for statistics on small-business email open rates and gets three precise percentages with named sources. Two of the sources do not exist and the third says something different. She now asks for claims to be paired with a live link and checks every one before it goes into a client deck.

Why it matters

Hallucination is the reason AI output needs a human check, and the professional who understands this gets more value from the tools, not less. You can hand off drafting and analysis freely as long as you know which parts to verify: names, numbers, dates, citations, and anything you would be embarrassed to get wrong.

Related terms