AI Detector Checker is a free AI-writing detector for English text. You paste at least 100 words and get a 0-100 AI-writing signal score, a stability range for that score, one of three bands, and the measured rate at which human writing of that length was wrongly flagged in our sealed test. There is no account, no paywall and no upsell.
Who runs this site
The site is published and independently operated by Thinkly for Digital Business, a small digital business. No detector vendor, university, platform or publisher funds this site or reviews what it says. Pages are written and reviewed by the AI Detector Checker editorial team at the publisher; where an individual author is credited, their name and role appear in the byline. We do not invent bylines: if no person is named, the organisation is the author.
What the tool is, and what it is not
It is a review signal: one measurement of how strongly a passage matches the writing patterns a frozen classifier learned from AI text, reported with the error rate we measured for that length of text.
It is not proof of authorship, and it is never enough on its own for a decision about a person. It does not identify which model wrote a text. It does not tell you that a person wrote something: in our sealed test many AI texts drew no strong signal at all. It is validated on English of 100 words or more; other languages are accepted but nothing about those results has been checked.
How it is tested
Thresholds are frozen before testing. The test set is sealed: it is built without being looked at, scored exactly once, and then retired. The numbers from that single run are published as they came out, including the two targets the run did not meet — the short-text false-flag rate and the hardest human family. The publisher decided to release with those two limitations disclosed rather than re-tune the system against a test it had already seen. The full figures, with raw counts and Wilson intervals for every source of human writing, are on the evidence page.
How to report a mistake
If your own writing was wrongly flagged, or a page here is wrong, write to us through the contact page. Submitted text is not retained, so we cannot look up your check afterwards: please include the length of the text, the band you were shown and what you expected. Reports about wrongly flagged human writing are the ones we most want, because they tell us where the published rates understate the risk for a kind of writing we did not measure.
Editorial policy
- Every accuracy figure on this site comes from a measurement we ran and published, with its denominator. We do not quote marketing numbers, ours or anyone else’s.
- We do not describe a result as proof, and we do not describe the score as a probability or a level of certainty.
- When a test target is missed, it is published, not removed.
- Pages are dated. A “Reviewed” date appears only when a person actually re-checked the page; editing a typo does not create one.
- Corrections are made on the page itself rather than quietly replaced, and the evidence page keeps the history of engine versions.
Contact
General questions, corrections, privacy and security reports: use the contact page. Legal and privacy detail is in the privacy policy, terms, cookie policy and security pages.