Know what an AI flag is worth.

Yes. With a verified school or university email, you check up to 1,000 papers a month free. Anyone else gets 2,000 words a day free.

The detector is ours. We run it ourselves on a cloud GPU provider, and no other company's detector sees the text. The free school plan is how we get it in front of teachers. The paid plans for heavier use pay for it, once billing opens.

No. A flag from our detector is a reason to look closer. It is not proof that anyone cheated. Read the marked passages, then talk with the student with their drafts and earlier work in front of you.

Not in our tests. Our detector flagged none of 1,195 essays by English learners kept out of training as a fairness test. They come from one university English language program, written 2005 to 2012. It also flagged none of 600 more it had never scored, from collections whose other essays were partly used in training. That is a count, not proof of fairness.

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.

Our detector flagged none of 1,928 human-written papers and essays it was never trained on. The 95% ceiling on its false-flag rate is 0.155%. The ceiling is the highest error rate still believable given what we saw. Our detector called all 481 fully machine-written test documents AI, at every setting. The 95% ceiling on its miss rate is 0.62%. They were written by AI models used in training, or rewritten by our own tools. We don't give a single accuracy percentage. Each kind of error is reported on its own test instead.

In one 960-document test we ran, the two tied on writing by people and on fully AI-written text: neither flagged any of 600 human-written documents or missed any of 220 AI-written ones. On 140 papers an AI had partly edited, ours flagged 102 and Pangram 4 flagged 88. Our default line (6%) sits lower than Pangram 4's own line (about 10%). At the same 10% line the count is 99 to 88. Most of the gap is papers where a model added a paragraph. One test, 140 papers. The 220 AI-written documents come from models our detector was trained on.

Turnitin says its own document-level false positive rate is under 1% for documents it scores at 20% AI or higher, and shows no percentage at all below that line (Turnitin, 2023-06-14). A 2026 study, "Who wrote this? Evaluating the reliability of AI detection tools in higher education" (International Journal for Educational Integrity, 2026-06-23), found Turnitin scored all 40 fully AI-written papers in its set under that 20% line, missing every one. The 40 were master's-level papers one AI tool wrote for that study, not real student submissions. We have not tested Turnitin ourselves. See the studies on the comparison page

Partly. It flagged 102 of 140 papers an AI had partly edited. When an AI model lightly polishes a paper, our detector mostly misses it: of 39 polished papers, it flagged 3.

Yes. Paid humanizer tools can hide machine writing from our detector. The test is on the model card.

Our detector does not score text under 50 words, and readings under 150 words are weak. It tells you that instead of guessing.

Start with standard, the default. It flags a paper when more than 6% reads as machine-written. Accusation-safe (flags only above 15%) misses more AI-edited papers. Sensitive (above 2%) flagged 2 of 1,928 human-written documents. Check whether your school already has a rule.

No. Our detector reads English academic and student writing. It is not built or tested for other languages or other kinds of writing.

Not yet. We run and grade our own tests.