By FactInfoLive · Published 2026-11-12 · Updated 2026-11-12
The short answer Based on the sources below, AI-text detectors should not be treated as proof. OpenAI withdrew its own detector in 2023 because of a low rate of accuracy, and a peer-reviewed-journal study reported that such tools can misjudge writing by non-native English speakers. This article summarises those sources; it is not a test of any current detector.
What OpenAI's detector did On 31 January 2023, OpenAI published a post introducing an "AI classifier" meant to indicate AI-written text. In its own evaluation on a "challenge set" of English texts, OpenAI reported that the classifier correctly identified 26% of AI-written text as "likely AI-written," while labelling human-written text as AI-written 9% of the time. The post also listed limits: it was very unreliable on texts under 1,000 characters, it was recommended for English only, and OpenAI said it should not be used as a primary decision-making tool.
Why it was withdrawn A note added at the top of the same page says that as of 20 July 2023 the classifier was no longer available due to its low rate of accuracy. It adds that OpenAI was researching more effective provenance techniques for text. TechCrunch reported the shutdown on 25 July 2023 and said the author had asked OpenAI about the timing and reasoning; the article does not report a reply.
What OpenAI says now OpenAI's help-centre page for educators asks "Do AI detectors work?" and begins its answer with "In short, no, not in our experience." It says OpenAI found its own detector labelled human-written text, including the Declaration of Independence and Shakespeare, as AI-generated. It also says there were indications that detection could disproportionately affect students who were learning English as a second language, and that students can make small edits to evade detection. The page is marked as updated over a year ago, so check it for newer wording.
The bias study A paper by Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu and James Zou, first submitted to arXiv in April 2023 and later published in the journal Patterns, examined several GPT detectors. The abstract states that the detectors consistently misclassified non-native English writing samples as AI-generated, while native samples were accurately identified. It also says simple prompting strategies could mitigate this bias and effectively bypass the detectors, and it calls for caution in using detectors in evaluative or educational settings.
What the sources do not tell us The sources here are a few documents, not a full survey of the field. Other researchers or vendors may report different results, and we have not tried to review them. Readers who need a firm answer for a policy decision should look for recent independent evaluations.
These findings are from 2023, and detection tools have changed since. We did not find or test current commercial detectors for this article, so we cannot say how any specific product performs today. The OpenAI numbers describe one tool on one test set, not detectors in general. TechCrunch also recorded its own small test, in which one detector identified five of seven generated snippets correctly and OpenAI's classifier only one; seven snippets is too few to draw broad conclusions from.