When an AI “hallucinates,” it generates information that sounds confident and plausible but is actually false, fabricated, or unsupported by any real data. The term captures how the AI presents made‑up content as if it were factual—even when it isn’t grounded in truth.

This happens because large language models don’t retrieve verified facts; they predict the next word based on statistical patterns they’ve learned. When they don’t know the answer, instead of saying “I don’t know,” they often produce a fluent but incorrect response—effectively confabulating rather than conveying truth.