Why does AI sometimes make up fake facts that sound completely real?

AI makes up fake facts — what researchers call 'hallucination' — because it's designed to predict plausible-sounding text, not to retrieve verified truth. The model doesn't have a database of facts it checks before answering. Instead, when you ask a question, it generates a response word by word based on patterns it learned during training, prioritizing what sounds like a coherent, confident answer over what's actually correct. A concrete example: if you ask for a biography of a minor historical figure, the AI might confidently invent a birth year, a quote, or a book title that simply doesn't exist. It's not lying. Lying implies intent. The model has no concept of truth or falsehood — it's just really good at the rhythm of language. I've seen this trip up smart people who assume the AI has a fact-checking layer. It doesn't. The tip here isn't to stop using these tools, but to treat them like a very eloquent, overconfident intern. Always verify anything factual they produce, especially dates, statistics, and citations. One practical habit: ask the AI to cite its sources. If it can't, or if the sources look like they were generated by the same pattern-matching machine (because they were), that's a big red flag. For a deeper dive, see our guide on what to do when AI writing tools give you unreliable outputs. **Related**: How can I spot AI hallucinations in a long document? | Do all AI models hallucinate at the same rate?