Language models & chatbots
An AI hallucination is when a language model states something false or invented as if it were a fact, such as a fake quote, citation, or event.
A language model always produces the most plausible-sounding next words. It has no built-in sense of 'I do not actually know this.' So when you ask about something outside its training, or something too rare to have learned properly, it fills the gap with text that looks right. That is a hallucination.
Hallucinations are fluent and confident, which is what makes them dangerous. They are common with precise details: dates, statistics, legal cases, web addresses, and references. The model has learned the shape of a citation better than any particular citation, so it generates a convincing one that does not exist.
The cure is not to scold the model. It is to give it sources, through RAG or web search, to ask it to cite its evidence, and to verify anything important yourself. Newer models hallucinate less, but as of 2026 none are immune.
Asked for books about a niche topic, a model recommends a title, author, and year that all sound real but were never published.