AI detector comparisons

Two detectors, the same samples. Each comparison sets both readings over one essay, counts false alarms on human writing out of the same base, and ends with a verdict by type of user.

Every comparison uses the same 180-sample protocol and the same flag rule as the rest of the site, so the two tools are judged on identical writing: student essays, essays by writers working in a second language, professional prose, raw AI text and AI text a person has edited. Both readings are drawn over one essay, so you can see the paragraph where they disagree.

Each page starts with a verdict and the false alarm counts for both tools, then compares access, price with the date we checked it, report detail and published accuracy figures with their sources. It ends with a verdict for each type of user, because the better tool for a teacher with a licence is not always the better tool for a student without one.

Vendor claims are quoted and linked to the vendor's own page, and independent studies are cited by name and year, so every figure can be checked. When a new run changes a verdict, the page says which run moved it and the change is logged on the corrections page.

Looking for one pick rather than a pair? The picks by job name a single winner for teachers, students, essays, free use and images, and the alternatives pages sort replacements by the reason you are switching.

Run RUN-2026-11-A started on 9 November 2026 and is in progress: GPTZero and Originality.ai are scored on the first 120 samples, and every other tool shows a test in progress tag until its readings are in.

How we testTest runs and raw files