A hallucination is an output that reads well but is wrong: a made-up statistic, a function that does not exist, a citation to a paper nobody wrote. It happens because a large language model generates likely-sounding text rather than looking facts up, so a gap in its knowledge can be filled with a plausible guess.
You cannot eliminate hallucinations, but you can reduce and catch them. Give the model the facts it needs (AI agent grounding and Retrieval-Augmented generation), allow it to answer that it does not know, keep prompts specific (prompt engineering), validate structured output, and have people review anything high-stakes.
For publishers, hallucination is also a content risk: AI-written claims need checking against a primary source before they go live.
Related: Generative AI, AI Agent Context, Knowledge Base.