AI hallucination: why chatbots make things up
2026-11-10
Definition
An AI hallucination is an answer from an AI model that sounds confident and fluent but is false, unsupported or invented, such as a fake source, a wrong figure or a policy that does not exist.
Why it happens
A large language model produces the most plausible next words, not a checked fact. When it lacks the information, it can still fill the gap with something that reads well. The risk is highest for specific details: legal citations, prices, dates and names.
You cannot remove hallucination completely, but you can shrink it. Give the model your real documents using retrieval-augmented generation, tell it to say "I don't know" when the source does not cover a question, and keep a human in the loop for anything that reaches a customer. The answer page on AI hallucination goes further, and human review for AI output shows how to set up checks.
A hypothetical example
Picture a recruitment agency using AI to draft job ads. The model adds a benefit the client never offered, such as a car allowance. A recruiter who skims the draft before it is published catches the line. Without that review, the agency advertises something it cannot honour.
Deciding where AI output needs checking is a core part of AI strategy.