Catch the invented claim
A true-or-false exercise in checking AI output: each sentence in a short AI-style summary is either accurate or wrong, and you find the wrong ones.
Read the transcript
Your AI summary is fluent, confident and nearly perfect. So why is the sentence you trust most the one that is wrong?
We assume mistakes look like mistakes: broken grammar, odd words. They don't. A wrong sentence reads exactly as smoothly as a right one.
The usual errors are quiet: a finding reversed, a number swapped, a claim stretched, a deadline the source never gave.
Fluency is not evidence. A model writes what sounds likely, so an invented detail arrives in the same calm voice as a fact.
The fix is simple and slow: hold each sentence against the source. If you cannot find it there, it does not count.
Accuracy is a habit, not a feeling. Practise it free in our lab. Which kind of error would fool you first?
Checking your account…
This lab and its certificate are free. Full certification programs, the Certified AI Practitioner and role masterclasses, are coming. Tell us which you want first.
About this lab
A fluent AI summary can be wrong in a way that reads fine: a reversed finding, a swapped number, an overclaim, advice turned inside out, a requirement the document never made. This lab trains you to catch those. You read short summaries of real documents from sources such as Anthropic, OpenAI, OWASP, NIST and Google, flag the sentences you think are wrong, and see what you caught, what you missed and why, with a link to the source.
How it works
- Practice is free with an account: 6 summaries drawn at random, with every answer explained afterwards.
- The certified round is 10 summaries, open book. You pass with 80 percent and no more than 2 false alarms, and you get a certificate with your percentage that anyone can verify.
- The wrong sentences are written by us to read the way a fluent model gets things wrong. None is a real model’s output, and every true sentence was checked against its source.
Questions
What is the Catch the invented claim lab?
It is a practice exercise in checking AI output. You read short summaries of real documents, flag the sentences you think are wrong, and then see what you caught, what you missed and what you flagged by mistake, with the reason and a link to the source for each.
Are the wrong sentences real AI mistakes?
No. We wrote the wrong sentences ourselves to read the way a fluent model gets things wrong: a reversed finding, a swapped number, an overclaim, advice turned inside out, or a requirement the document never made. Every true sentence was checked against its source page.
How is the certified round scored?
The certified round has 10 summaries. Your score is the share of the wrong sentences you caught, with half a point taken off for each true sentence you flagged. To pass you need 80 percent and no more than 2 false alarms, so flagging everything does not work.
Is practice free, and do I need an account?
Practice rounds of 6 summaries are free, with a free account. The account keeps your scores, your progress and any certificate you earn.
Can I read the sources during the certified round?
Yes. The certified round is open book, because checking a claim against its source is the skill being tested. The answers are not shown afterwards, so use practice rounds to learn.
How can someone verify my certificate?
Each certificate has its own number and a public page showing the percentage, the errors caught, the date and the version of the lab. If a certificate is revoked, that page says so.