AI Tools Academy
0 / 117 (0%)

Practice · about 20 minutes

Spot the Hallucination

AI answers read just as confidently when they're wrong. Here are 12 realistic ones, each split into claims. Flag the ones you'd check or reject, then see what was really going on.

Some claims are fine: correct, or taken straight from the material the person supplied. Flagging everything isn't the answer. The challenges get harder as you go, and the later ones are the kind of subtle error that gets through at work. Every person and organisation is fictional, apart from a few harmless public facts.

Challenge 1 of 12 · 0 checked so far

Warm-upGeneral knowledge

The team quiz questions

What they asked the AI: I'm running the quiz at the Fernway summer social. Give me a few science and general knowledge facts I can turn into questions.

The AI's answer, one claim per line. Select the claims you'd flag to check or reject.

0 flagged. You can flag none, some or all.

The kinds of error in these challenges

  • Made up: a fact, number, feature or person that doesn't exist.
  • Out of date: true once, not now.
  • Real source, misread: the document exists but doesn't say that, or drops an exception.
  • Not supported: a conclusion the evidence doesn't reach, such as cause from a correlation.
  • Misleading precision: an exact-looking figure with nothing behind it, or worked out on the wrong base.
  • Invented quotation or citation: words nobody said, or a reference that doesn't exist.

The Prove It course turns spotting these into a five-check routine for real work. For the background, read how to check whether an AI answer is right.

Want a routine for this at work?

Prove It

Catching AI when it is wrong, before it costs you. Four modules. Five checks. A habit that keeps your name off the mistake.

Free short course, no sign-up.

Explore Prove It