AI hallucination
Short definition
Hallucination is when an AI model states something that is not true, in a confident tone that reads as fact.
When a language model meets a question it does not actually know the answer to, rather than leaving it blank it can produce a plausible-sounding but false answer based on patterns. This is hallucination, and inventing a source, date, figure or quote is its most common form.
The model does not do this deliberately, it has no way to tell whether its own output is true or false, the tone stays equally confident either way, which makes the error harder to spot.
The risk of hallucination rises when the model is working from outdated knowledge, when asked about something very specific, or when it generates freely without being grounded in a source, see RAG.
The simplest way to spot a hallucination in practice is to open every source or link the model cites and check it directly; a fabricated source usually either does not exist at all or does not contain the claim attached to it. Any answer containing figures or dates should never go live before it is checked against the original document or an official record.
Why it matters
An unchecked hallucination means a wrong price, wrong legal information, or a source that does not exist can make its way into a customer message or published content. That is why every AI output, especially anything with figures or claims, still needs a human check.
Illustrative example
An assistant integrated with a SKU system suggested a product code that did not exist as if it were real, the team caught it before it went live.
