An AI hallucination is a confident-sounding statement that a language model generates without any factual basis. The model does not know it is wrong. It produces text that looks authoritative and specific, including invented citations, false statistics, fabricated quotes, and non-existent laws or court cases.
Hallucinations happen because large language models predict the next most plausible word based on patterns in training data. They are not looking facts up or retrieving verified information in real time. When a model fills a gap in its knowledge, it generates plausible-sounding content rather than saying it does not know.
Hallucinations are more likely when the prompt involves specific numbers, names, dates, legal citations, medical details, or recent events. They are also common in long answers where the model must sustain detail across many sentences.
To spot a hallucination, look for claims that are very specific but unverifiable, citations to sources that cannot be found, statements that seem surprisingly confident about something obscure, and facts that contradict what you know from trusted sources.
To verify AI output, check each specific claim against a primary source independently. Do not use the same AI to verify its own output. For legal, medical, or financial claims, consult a licensed professional. For news claims, verify against multiple independent news outlets or official agency statements.
Hallucinations become most dangerous when used to inform medical decisions, legal arguments, financial investments, or content published as factual. Scammers exploit hallucinations by feeding false information into AI tools and presenting the AI output as independent verification of their claims.