Try it now
Can you tell when an AI is wrong?
Three real questions from the Reviewing AI Answers test. No sign-up, nothing to install. Takes about two minutes.
How this works
- You read a question someone asked, and the answer an AI gave.
- You decide how good that answer is.
- Then you see how our experts scored it, and why.
Question 1 of 3
Someone asked: Who won the Nobel Prize in Literature in 1953?
The AI answered
The 1953 Nobel Prize in Literature was awarded to Ernest Hemingway, recognised for his mastery of the art of narrative and the influence he has had on contemporary style.
Pick one of the three options first.
The correct score is: Wrong.
Winston Churchill won the 1953 Nobel Prize in Literature. Hemingway won it the following year, in 1954. The wording the AI used is even taken from Hemingway's real citation, which makes it sound more convincing.
Why this matters: this is the single most common way people fail the test. The answer is fluent, confident and well written, so it feels right. Fluency is not accuracy. Check the claim, not the tone.
Question 2 of 3
Someone asked: My bread dough keeps sticking to everything. How do I stop it?
The AI answered
Put a little flour on your hands and on the surface before you start. Only use a small amount — too much extra flour will make the bread dry and heavy.
Pick one of the three options first.
The correct score is: Good.
It is accurate, it directly answers what was asked, and it adds the one warning that actually matters. Nothing is missing.
Why this matters: lots of people mark short answers down for being short. That is a habit, not a judgement. A brief correct answer beats a long one that pads. We are scoring whether it helps, not how much of it there is.
Question 3 of 3
Someone asked: Summarise this report for me in exactly three bullet points.
The AI answered
• Sales rose 12% over the year
• Growth came mostly from the north region
• Costs stayed flat
• Two new products launched in autumn
• Staff numbers were unchanged
Pick one of the three options first.
The correct score is: Has a problem.
The person asked for exactly three bullet points and got five. Every fact is right, but the instruction was ignored.
Why this matters: being correct is not the same as being right. If someone asked for three and you hand back five, they now have to do the work themselves. Following the instruction is scored separately from accuracy, and it is worth 20%.