Don't trust it blindly: how to check what AI tells you

What you'll see in this lesson

1. Why AI makes things up so confidently

   Models learn to always give a plausible answer, and on their tests guessing scores better than saying "I don't know".

2. Where to look closely: the risk zones

   You don't need to distrust everything: there are types of data where AI fails much more and that you should always check.

3. Made-up citations: the mistake that's already costly

   In Spain, the High Court of the Canary Islands has already fined two lawyers for filing briefs citing rulings that didn't exist.

4. Ask for sources in a way you can check them

   "Give me sources" isn't enough: ask for the link and the exact sentence backing each fact, and let it say "I don't know".

5. Quick answer or deep research

   Deep research searches dozens of sources for minutes and delivers a report with citations; a quick answer takes seconds.

6. The four deep research tools of 2026

   ChatGPT, Gemini, Claude and Perplexity each have a deep research mode, with different limits depending on the plan.

7. How to review a report with citations without losing the afternoon

   You don't need to check everything: review the claims your decision depends on.

8. A four-step method for any answer

   Stop, go to the original source, cross-check and decide: it works for a quick chat and for a long report alike.

9. Summary: trust, but check what matters

   AI speeds up finding information a lot; verifying it is still your job.