An AI model predicts plausible text. Usually the plausible answer is also the true one. But when the model does not know something, it does not naturally say "I don't know". It produces something that sounds right: a study that was never published, a statistic that does not exist, a confident summary of a document it misread. That is a hallucination.
The dangerous part is the delivery. Hallucinated answers read exactly like correct ones. There is no stammer, no hedging, no tell.
Why you keep hearing it
Because it is the most practical limitation of AI at work, and the reason "the AI said so" is not a source. Newer models hallucinate less, and grounding techniques like RAG help by making the AI answer from real documents. But the risk does not go to zero, and every serious AI workflow is designed around that fact.
What it means for you
Simple working rules. Anything with a name, number, quote or claim that will be published gets verified by a human. Ask the AI for its sources and check that they exist. Be most suspicious when the answer is exactly what you hoped to hear. AI is a brilliant drafting partner and a terrible witness.