Every test on PromptTick follows the same rules. If a page on this site says “tested”, this is what that word means here.
First try only. When we test an AI tool, the result we publish is the first result the tool returned. No re-rolls, no generating ten versions and showing the best one. If the first try is ugly, the ugly one is what you see.
Real accounts, real settings. We test in the same apps our readers use, signed in the same way, and we record what the app showed — including the model label on screen, even when that label is vague.
Scoring we show our work for. When a test produces a score, the scoring rules are written down before or during the test, the full data is published with the article (usually as a downloadable file), and the failures are shown next to the successes. When a test has limits — one scorer, a small sample, one day of testing — the article says so in plain words, next to the claims, not buried at the bottom.
Controls. Where we can, we run control cases designed to fail, so readers can see the test is capable of producing a bad result. A test that cannot fail anyone proves nothing.
Corrections. If we get something wrong, we fix the page and say what changed. Email us at contact@prompttick.com.
Our first full test under these rules: Does Gemini Keep Your Face? — 26 first-try edits, scored trait by trait, with failure controls.