Hallucination is when an AI model generates information that sounds plausible and confident but is actually false or made up. It is one of the most important limitations to understand when using AI.
What it means in plain English
Because a language model generates the most likely continuation of text rather than looking up verified facts, it can produce statements that are fluent and confident yet simply wrong — inventing a statistic, a citation, or a detail that does not exist. It is not lying in any intentional sense; it has no concept of truth, only of what text is probable.
Crucially, hallucinations are delivered with the same confidence as correct answers, which is exactly what makes them dangerous. The model gives no signal that it is unsure.
A simple example
Ask an AI for sources on a niche topic and it may produce a perfectly formatted list of articles and authors that do not exist — because plausible-looking citations are a likely continuation, whether or not they are real.
Why it matters
Understanding hallucination is the single most important safeguard when using AI. It is why any factual, legal, medical, or numerical output must be verified against a reliable source, and why AI is best treated as a fast first draft to check rather than an authority to trust.
Related terms
- Large Language Model — the systems prone to hallucination.
- Retrieval-Augmented Generation — a technique that reduces hallucination.
- Bias — another important limitation of AI models.
Frequently asked questions
What is an AI hallucination?
It is when an AI model generates information that sounds plausible but is false or fabricated — such as inventing a fact, citation, or detail that is not true.
How can hallucinations be reduced?
Grounding the model in reliable sources (retrieval-augmented generation), asking it to cite sources, and verifying important output all help — but users should always fact-check consequential AI claims.
