Why AI sometimes makes things up, and the 10 second fix

You ask an AI assistant a simple question. The answer comes back clear, well written and completely wrong. It happens to everyone, and once you see why, it becomes easy to prevent.

AI predicts, it doesn’t look up

Tools like ChatGPT and Copilot don’t search a database of facts when they reply. They predict the most likely next word, then the next, and so on. Think of a very well read friend finishing your sentence.

If you start with “The capital of France is…”, one word wins by a mile. Paris scores far higher than anything else, so the answer is right.

When it doesn’t know, it still has to pick

Now try “My manager’s birthday is…”. The AI has never seen that fact. Every date scores about the same, but the model still has to choose one. The result is a guess delivered in the same confident tone as a real fact. That’s what people call a hallucination.

The fix: give it the facts

1. Paste the source into the chat: your notes, a policy, a contract clause or an email thread.
2. Add one line: “Only answer from my notes. If it’s not there, say so.”

You’ve turned a guesser into a reader. The answer now comes from your information, and when the information is missing, the AI tells you instead of inventing something.

This same idea sits behind most business AI assistants. Copilot answers from your organisation’s documents for exactly this reason.

One practical AI idea a day from AI Powered 365.