HomeGlossary › AI Hallucination

AI Hallucination

AI hallucination is when a large language model generates plausible-sounding but factually incorrect information presented with high confidence. A known limitation of current LLMs.

Definition: When a large language model (LLM) like ChatGPT, Claude, or Gemini generates information that sounds plausible and authoritative but is factually incorrect. The model presents the false information with the same confidence it presents true information, making hallucinations difficult to detect without independent verification.

How it works

LLMs work by predicting likely next tokens based on training data patterns, not by retrieving facts from a verified database. When the model lacks information, it can produce statistically plausible but invented content — fake citations, incorrect dates, fabricated quotes, non-existent products, or wrong technical details. Hallucination rates have decreased with newer models but remain a fundamental limitation.

Example

A user asks an LLM about a specific case law citation; the LLM responds with a confidently-stated case name, year, and ruling that doesn't exist. Or asks about an obscure scientific paper; the LLM invents a paper title, author, and journal that match the user's query but aren't real. Users in legal, medical, and academic contexts have been disciplined for citing AI-hallucinated sources.

Comparison + context

Mitigations: (1) Use grounded AI tools that cite real sources (Perplexity, search-grounded ChatGPT) for factual queries. (2) Always verify AI-cited sources independently. (3) Use Retrieval-Augmented Generation (RAG) systems for domain-specific accuracy. (4) For critical facts, treat AI as a draft generator, not a fact source.

See also