How we work
Methodology
How this site researches and verifies what it publishes, what we do not do, and how to challenge anything here that you think is wrong.
Sources
What we rely on, in order of weight
1. Peer-reviewed research
Our accuracy claims come from published, peer-reviewed testing — principally Liang, Yuksekgonul, Mao, Wu and Zou, "GPT detectors are biased against non-native English writers", Patterns (Cell Press), 2023. Where we cite a figure, we name the study and the sample so you can check it.
2. Primary institutional statements
Policy claims come from the institution's own published statement, not from press coverage of it. Vanderbilt's explanation of why it disabled Turnitin's AI detector is quoted from Vanderbilt.
3. Vendor documentation, treated as a claim
We report what vendors say about their own accuracy and we label it as a vendor claim. A stated false-positive rate that has not been independently reproduced is marketing until it has been.
Limits
What we do not do
We do not publish our own detector scores
We have not run a controlled, statistically meaningful test of these detectors. Publishing a handful of ad-hoc scores as if they were evidence would be exactly the behaviour this site criticises. Where you see a number here, it comes from published research and is cited.
We do not promise detector bypass
No tool can honestly guarantee a detector result, and any product claiming to should be treated with suspicion. We list tools with their claims and our caveats; we do not endorse the claims.
We do not help you misrepresent authorship
This site exists to explain how detectors work, why they fail, and what to do when they fail on you. It is not a guide to passing off machine-written work as your own, and the guidance here would not help you do that.
We do not take payment for placement
Tool listings are editorial. MultipleChat AI is our own product and is labelled as such wherever it appears.
Corrections
Challenging something on this page
If a figure here is out of date, misattributed or wrong, we want to fix it. Detector vendors change their models and their claims frequently, and institutional policies change faster than that. Every guide carries a "last reviewed" date so you can judge how stale it might be.
Tell us what is wrong and what the correct source is, and we will either correct it or explain why we read the evidence differently. Corrections to substantive claims are noted on the page.
Last reviewed: 14 August 2026. We update these pages when detector vendors change their claims or new peer-reviewed evidence is published.