Method the rules every review follows

How we test AI tools

A BrokenCtrl review is a record of what happened when we used the tool: the plan we were on, the date, the input, the screens, and what the tool got wrong. This page is the whole method.

1. We sign up ourselves, at the vendor, not through our own link. Self-referral can void an affiliate account, and it would make the review a sales page.
2. We use the free plan or a no-card trial. Nothing is bought for a test. If a card is asked, the test stops until a person decides.
3. The input is ours and public. A published paragraph, a clip we made, a file with no personal data. Never a demo the vendor chose. The Trap Kit lists the inputs.
4. Every review shows the date and the plan. AI tools change monthly; a review without a date is not evidence.
5. Screenshots are cropped. No email address, name, avatar, billing detail or key appears on a page.
6. Numbers are re-read before they are printed: word counts against the script, citations on Crossref or arXiv, prices on the vendor’s own page on the test day.
7. We say what we did not check. «We did not read the papers in full» is a sentence you will find in our reviews.
8. Every fault goes in the Wrong Ledger with its date, input and a link to the evidence.
9. Affiliate links are disclosed in the same sentence as the link. A link never decides a verdict, and a review can end with «a chatbot you already have does this».

A person signs up, uses the tool, speaks the verdict and decides what is published. An assistant runs the tool alongside, logs the screens and checks the numbers. Where an assistant ran part of a test, the page says so, and where a verdict is still owed by the person, the page says that too.

One input, one day, one plan. A tool that failed our trap may pass yours, and a tool that passed may fail next month. That is why every review carries a date and why we re-test.

See every fault we found →   The Trap Kit →   Comparisons →