Why does ChatGPT sometimes make up facts that sound totally believable?

ChatGPT invents facts because it's designed to predict the next most likely word in a sequence, not to verify truth—a behavior researchers call 'hallucination.' It doesn't have a fact-checking brain. It has a pattern-matching engine trained on a massive amount of text from the internet. When you ask it something, it builds a response that statistically looks like a good answer, even if the underlying facts are completely wrong. I've seen it invent book titles, scientific studies, and even fake historical events with dates and names that sound incredibly convincing. For example, you might ask for a biography of a minor 19th-century inventor, and ChatGPT will confidently produce a detailed life story—birthplace, patents, family—all of which are pure fiction. It's not lying in a human sense. It's just doing its job of assembling plausible-sounding text. The real danger is how confident it sounds. A 2023 study from Stanford's Human-Centered AI group found that hallucination rates vary wildly by topic, sometimes exceeding 20% for niche legal or medical questions. That's why you should always treat its output as a first draft, not a final source. A practical tip: add 'If you don't know, just say so' to your prompt. It doesn't fix the problem entirely, but it can reduce the AI's overconfidence. For a deeper dive on getting more reliable results, see our guide on troubleshooting prompts. **Related**: How can I spot AI hallucinations in a text? | Do all AI chatbots hallucinate at the same rate?