6.1 What Is an AI Hallucination?

AI sometimes generates information that appears reasonable but is wrong, unsupported, or nonexistent. This is commonly called an AI hallucination.

A hallucination need not look like nonsense. It can sit among correct facts in complete sentences, professional language, and polished formatting—for example, an invented award, false quotation, or gift never promised by the source.

Why do hallucinations occur?

A large language model's core task is to generate a suitable continuation from context, not to guarantee that every sentence has been checked against a factual database. When information is missing, a question demands an answer, sources are vague, or the task exceeds its ability, it may still produce something shaped like an answer.

A reliable response should allow “I don't know,” “I cannot confirm,” and “the available material does not say.” Users can ask AI to separate known facts, guesses, and missing information.

Providing reliable material, restricting answers to specified sources, marking uncertain items, and verifying critical facts can reduce risk. No single prompt can guarantee that hallucinations disappear entirely.