Hallucination
Also known as: AI hallucination
A hallucination is a freely invented but confidently worded answer from an AI model that is presented as fact.
A language model produces the text that is statistically most plausible to follow an input, not the text that is provably true. When the model lacks the right information, it still produces an answer, phrased in the same confident, fluent tone as a correct one — and that freely invented but convincing-sounding statement is what's called a hallucination.
In practice this shows up as an invented source citation, a plausible-sounding but wrong legal reference, an incorrect figure in a summary, or a product feature that doesn't actually exist. Notably, the tone of the answer gives no clue whether it's correct — a hallucinated statement sounds exactly as confident as a correct one.
A common misunderstanding is assuming a newer or larger model rules out hallucinations. They become less frequent, but they don't disappear, and even a RAG setup that grounds answers in a company's own documents reduces the risk without bringing it to zero.
What it means in practice
This is exactly why AI output in business processes needs to be checked before it's used further: an AI-drafted customer reply, an invoice, or a technical spec must never go out unreviewed, because a wrong figure or a wrong citation sounds just as convincing as a correct one. Where that check is missing, the business bears the consequences, not the model.
Is this handled properly at your site?
We look at how it actually stands with you — and say honestly whether anything needs doing.
Related terms
All termsA term from your quote missing here?
Send us the passage you do not follow. We will explain it — with no obligation to order anything.