← AI Learning Centre
Everyone · 5 min read

How to check whether AI is telling the truth

Five habits that catch most AI mistakes before they cost you.

1. Check the specific details

Mistakes hide in names, numbers, dates, article numbers and quotes. Those are exactly the things to verify at an official source: the law itself, the company register, the institution’s website.

2. Open every source

If AI gives a link or a reference, open it and find the sentence it relies on. Sources that don’t exist or don’t say what the AI claims are a classic sign of a hallucination.

3. Ask for uncertainty

Ask: “Which parts of your answer are you least sure about, and how could I check them?” Good assistants will point out their weak spots.

4. Read laterally

Don’t judge a claim only by how it is written. Open two or three independent sources in new tabs — fact-checkers do exactly this.

5. The higher the stakes, the more checking

  • Health, money, law and safety: always confirm with a professional or an official source.
  • Anything you publish or send to a customer: check every fact.
  • Private brainstorming: lighter checking is fine.
Key takeaways
  • Verify names, numbers, dates and quotes at the source.
  • A source you haven’t opened doesn’t count.
  • Ask the AI where it is unsure.
Try it

Use a ready-made prompt that makes the model check its own claims.

Open the fact-check prompt →

Check yourself

  1. AI cites a court decision. What should you do?
  2. What is “lateral reading”?
  3. When is careful checking most important?

Last reviewed: September 2026