What does it mean when people say an AI is 'hallucinating'?

When people say an AI is hallucinating, they mean the tool confidently makes up information that sounds believable but is completely wrong or doesn't exist. It's not lying in the human sense — the AI doesn't know it's fabricating things. It's more like a prediction engine that guesses the next most likely word, and sometimes those guesses create a convincing but fictional story. For example, you might ask an AI to summarize a book and it'll invent a character that never appeared, or ask for a biography of a real person and it'll confidently list awards they never won. I've seen it cite fake research papers with realistic-sounding titles and author names. The danger is how polished the output looks. A hallucination doesn't come with a warning label. It reads just as smoothly as accurate information, which is why fact-checking is non-negotiable. This happens because large language models don't have a database of facts. They have statistical patterns of how words tend to go together. So when they don't know something, they don't pause and say 'I'm not sure.' They just keep generating based on those patterns. The result can be impressive fiction. One practical tip: if you're asking for factual information, tell the AI to only provide answers it can verify and to explicitly say 'I don't know' when uncertain. This simple instruction cuts down hallucinations noticeably. For a deeper dive, see our guide on troubleshooting AI prompts that produce unreliable results. **Related**: Why do AI chatbots make up fake sources? | How can I fact-check AI-generated content?