Why AI Makes Things Up (and How to Catch It)

AI tools sometimes state false things with complete confidence. This is often called hallucination, and understanding it is the best protection against it.

Language models predict plausible text. They don't look facts up in a database of truth. When they lack information, they can still produce something that sounds right: a fake citation, an invented statistic, a function that doesn't exist.

• Obscure facts, names, dates and numbers.
• Citations, case law, quotes and URLs.
• Recent events beyond the model's training.
• Questions with a false premise built in.
• Long answers where small errors accumulate.

1. Ground it: give the AI the source material and tell it to answer only from that.
2. Allow "I don't know": explicitly permit uncertainty.
3. Use search or retrieval tools for current or factual questions.
4. Ask for sources, then open them yourself.
5. Ask it to check itself: "Which of these claims are you least sure about?"
6. Keep tasks narrow. Specific questions beat sprawling ones.

• Does the source exist and say what the AI claims?
• Do numbers and dates match a second source?
• Would a mistake here cost money, health or reputation? If yes, a human expert checks it.

Treat AI like a brilliant but overconfident intern. Great for first drafts and ideas, but verify anything that matters before you rely on it or publish it.