For administrators
Before your school adopts a detector
Our detector's limits are written down on this page, with the test behind each one. Staff with a verified school email check 1,000 papers a month free.
Wrong flags in testing
At the standard setting, our detector flagged none of 1,928 human-written papers and essays it was never trained on. For a school checking 10,000 papers, that means no more than about 16 wrongly flagged papers at 95% confidence. At the sensitive setting the ceiling is about 28. This holds only for English academic and student writing like the test set, all written before 2022.
- Our detector, run v1_8_s2_seed303, standard setting (6%), measured 2026-09-15.
Tested against a commercial detector
In a 960-document test, neither our detector nor a commercial detector flagged any of 600 human-written papers and essays or missed any of 220 machine-written ones. The 220 come from models our detector trains on.
- Ours flagged 102 of 140 papers an AI had partly edited, and the commercial detector flagged 88.
- Our default line is lower than the other detector's. With both at the same line, the count is 99 to 88, and most of the gap is papers where a model added a paragraph.
- One test, 140 papers. The comparison page names the other detector and lists every caveat.
Fairness testing
Our detector flagged none of 1,195 essays by English learners that were kept out of training as a fairness test. They come from one university English language program, written between 2005 and 2012. That is a count, not proof of fairness.
- One learner collection served as the fairness test. Two more were planned and never obtained, so we have no results on them.
- Learners who now write alongside AI tools, grammar checkers or translation tools have not been tested.
Three settings
Changing the setting never turns an AI verdict into Human. It only moves the line for papers that are partly machine-written.
- Standard (6%, the default) flags a paper when more than a small part of it reads as machine-written.
- Accusation-safe (flags only above 15%) flags a paper only when a substantial part of it reads as machine-written. It removes no false flags, because the default has none, and it calls 10 more AI-edited papers human-written.
- Sensitive (2%) flags at the first small sign, and flagged 2 of 1,928 human documents. Use it to prompt a second look, not a conversation about misconduct.
Writing the policy
A flag from our detector is a reason to look closer. It is not proof that anyone cheated. Each result shows the passages our detector reads as AI-written and names the detector run. Treat the marks as a reading, not a finding. We have not measured how exactly they fall on the sentences an AI model wrote.
- No grade or penalty should rest on a flag alone
- We run and grade our own tests. No outside party has checked these results yet.
- Every human-written document we tested was written before 2022.
Where it falls short
When an AI model lightly polishes a human paper, our detector mostly misses it. Of 39 polished papers, it flagged 3. Paid humanizer tools can hide machine writing from our detector. Even at the default setting, it called 38 of 140 partly AI-edited papers human-written. It does not score text under 50 words, and readings under 150 words are weak.
See every limitStudent work
Human scores the text in memory and does not save it. Neither this site nor our detector keeps a copy of the papers you check.
Read the data privacy FAQFree for staff
Each teacher signs up with a school email, verifies it, and gets 1,000 papers a month. If your district or university needs a signed agreement before staff use it, write to hello@olive.is.
Free for your staff.
Every teacher with a verified school email gets 1,000 papers a month. Need a signed agreement first? Write to hello@olive.is.