What does it actually mean when an AI 'hallucinates' and how can I spot it?

An AI hallucination is when the tool confidently generates information that sounds correct but is completely made up, like a student who didn't study inventing a historical event on a test. It's not lying in the human sense — the AI doesn't have intent. It's just a pattern-prediction machine that sometimes fills gaps with plausible-sounding nonsense because its primary job is to produce coherent text, not accurate text. I've seen this happen with everything from fake book citations to invented scientific studies. For example, if you ask an AI for a quote from a famous author, it might generate a sentence that sounds exactly like something they'd write, but you won't find it in any of their published works. The model isn't checking a database of real quotes. It's stringing together words based on statistical likelihood. You can spot hallucinations by watching for overly specific details that don't quite add up — a CEO's name that doesn't match any LinkedIn profile, a statistic that feels too neat (like exactly 47%), or a case study about a company you can't verify exists. A good habit is treating AI output like a well-meaning intern's first draft: it's a starting point, not a finished fact sheet. For a deeper dive, see our guide on troubleshooting when AI prompts go wrong. **Related**: Why does AI sometimes give different answers to the same question? | How do I fact-check AI-generated content efficiently?