also searched as: is zerogpt accurate · how accurate is zerogpt · is zerogpt reliable · is zerogpt a good ai detector · zerogpt accuracy · how accurate is zerogpt ai detector · is zerogpt good
No: ZeroGPT catches most unedited chatbot text, but it flags real human writing too often for its score to decide who wrote anything. Edited AI text mostly slips past it.
Trust it for a free first look at text you suspect was pasted straight from a chatbot, the case it gets right most.
Double-check it when the writing is formal or formulaic, like an academic paper or an essay in a second language, because that is where it flags humans most.
Don't rely on it if the score would be used to fail, discipline or fire someone: its own terms say no detector can determine authorship with certainty.
sources checked 2026-10-13 · by dan scutaru
| who tested it | what they found |
|---|---|
| Acta Neurochirurgica, 2025 | 16% of human papers flagged 40 of 250 published neurosurgery articles from before ChatGPT scored above 50% AI, while no GPT-3.5 or GPT-4o text scored below 50%. Tested June 2024. source |
| RAID benchmark, ACL 2024 | could not get under 5% false flags The only one of 12 detectors that could not be calibrated to a 5% false-positive rate; at its strictest it flagged 48% of human Wikipedia articles. source |
| Weber-Wulff and 7 co-authors, 2023 | 59% overall, 6th of 14 tools 54 documents, March 2023: all 9 human texts cleared, 15 of 18 raw ChatGPT texts caught, but only 3 of 18 edited or paraphrased. source |
| Stanford data, 2023 (our count) | about half of 91 ESL essays In the study's published scores, 44 of 91 TOEFL essays by non-native writers scored above 50% AI, against 0 of 88 US eighth-grade essays. source |
| Ars Technica test, July 2023 | Genesis scored 88.2% AI Benj Edwards ran a Genesis passage and part of the US Constitution through ZeroGPT; both came back AI-generated. source |
| ZeroGPT's own claim | "98.4%" accuracy, "<1%" false flags Both on the zerogpt.com homepage today, with no test set or method published. source |
It depends more on the text than on the tool. ZeroGPT handles a chatbot answer pasted without edits, and fails on the two cases that matter: AI text someone reworked, and human text that happens to be plain. Ars Technica's explanation is that detectors like it score how predictable the wording is, and academic papers, encyclopedia entries and second-language essays are predictable by design. ZeroGPT's own FAQ concedes that formulaic writing can resemble AI.
No, and the company does not claim it is. Its terms of use, updated on 7 October 2026, say no AI detector, ZeroGPT included, can determine authorship with absolute certainty. In the most recent peer-reviewed test it rated one in six published medical papers as more likely AI than not. A score alone is not evidence: ask which sentences were flagged and what else points to AI use.
On false flags it is one of the weaker ones when tested side by side. In the 2025 neurosurgery study GPTZero put none of 250 human papers above 50%, ZeroGPT put 40 and Corrector 76. RAID could not hold it to a 5% false-flag rate. For detectors as a group, see our page on whether AI detectors are accurate.
No. ZeroGPT is run by Olive Works LLC, a Wyoming company; GPTZero is a separate company founded by Edward Tian and Alex Cui, and we cover its record on its own page.
On unedited chatbot text, mostly yes. On edited or paraphrased AI text, mostly no.
Yes, it is a real product from a registered company, with paid plans from $14.99 a month. Legit is not accurate, and the same site sells an AI humanizer.
Mostly. Its FAQ says checked text is not saved, published or used for training. The privacy policy sets no fixed deletion date, so leave out anything confidential.
Yes: 16% of published medical papers in one peer-reviewed test. Its own FAQ says polished, formulaic writing can resemble AI.