Check an AI answer before you trust it

Updated 2 min read

Level: New to AI

Part of: Your first week with AI · Step 5 of 6

AI assistants write fluent, confident answers, even when they’re wrong. A made-up fact, quote or source is called a hallucination. The assistant isn’t lying on purpose: it predicts likely-sounding words, and sometimes those words aren’t true. Sounding sure is not the same as being right.

When to check

Always check before you act on anything about health, money, law, safety or someone’s reputation, and before you share an answer with other people. For a recipe idea or a birthday-card rhyme, a quick read is enough.

Five quick checks

  1. Ask for sources, then open them. A real link should load and actually say what the AI claims. Made-up links, book titles and quotes are common.
  2. Find it somewhere official. Prices, rules and settings change. Check the company’s or government agency’s own page, and look at the date on it.
  3. Ask the question a different way. If the answer changes a lot, be careful.
  4. Check every exact number, date and quote. These are the easiest things to get wrong, and the easiest to check.
  5. Ask a person who knows. For medical, legal or money decisions, an AI answer is a starting point, not advice.

Warning signs

  • Very specific numbers or quotes with no source.
  • A link that doesn’t load, or a page that doesn’t mention what the AI said.
  • An answer about something recent. The assistant may not know about it, or may mix up dates.
  • An answer that matches what you hoped to hear a little too neatly.

Why a high score doesn’t settle it

A model that does well on tests can still be wrong about your question. See What is an AI benchmark? for why scores only tell part of the story.

Try it

Ask an AI assistant a question you already know the answer to, about your town, your job or a hobby. Then ask it for sources and open each one. Note what it got right, what it got wrong and which links didn’t work.

The short version

Treat an AI answer like a tip from a helpful stranger: useful, but worth checking. More habits are in Using free AI models safely and Write better prompts.