Editorial policy
Every pick on this site is made the same way: from a dated run, with fixed weights, false alarms on human writing counted out loud, and the pick's main limitation stated next to it.
Page updated: .
How a pick is made
A pick comes from a run, not from an opinion. Every detector reads the same 180 samples in the same run, every reading is scored against the same flag rule, and the scores use fixed weights: false alarms on human writing 40%, detection on raw AI text 30%, detection on human-edited AI text 15%, access and price 15%.
Job pages change which samples count, never the weights. The pick for teachers is scored on student work; the pick for free tools is scored on free plans only.
What every pick must show
The run ID and date its numbers come from. False alarms on human writing as a count with its base, written as a count out of 100 human samples. The same fields for every tool: verdict, score and sub-scores, best for, main limitation, and price with the date we checked it. What the pick gets wrong, in plain words, next to what it gets right.
How we write about detectors
We call a flag a flag, not a finding. We never describe a detector score as proof that someone used AI, and every ranking page carries the same line: no detector score should be the only evidence in a misconduct case.
When a well known tool flags real human writing, we say so and show the paragraph it flagged. Vendor claims are quoted and attributed to the vendor, with a link to the page where the vendor makes them.
When a pick changes
A pick changes only when a new run changes the order. The page's dates line and chart header then cite the new run, and the change is logged on the corrections page with the old pick, the new pick and the run that moved it.
Product names and logos
We use detector names and logos only to identify the tools we test. Logos are shown as supplied by their owners and are not altered.
Commercial relationships
Tools are scored on the same samples and fields regardless of who makes them.
Corrections and disputes
Anyone, including a detector company, can report an error through the contact page. We check the claim against the run file and log the outcome on the corrections log, whether we change the page or not.
Related reading
About Best AI Detectors and Its EditorsWho edits the site and why it exists.
How AI Detectors Are Tested: Our Samples, Scoring, DatesThe protocol behind every pick.
Corrections and Disputes: Every Change We MadeEvery change, with its date and evidence.
Contact Best AI Detectors: Corrections and TipsCorrections, disputes and questions.