GPTZero — AI-Text Detection and Its Academic-Integrity Risks

GPTZero was created by Princeton undergraduate Edward Tian in January 2023 as one of the first widely used tools for detecting AI-generated text in student submissions, using "perplexity" and "burstiness" text statistics.

The tool grew explosively: 30,000 uses in its first week, over 1 million users by mid-2023, and roughly 4 million users by July 2024, with a partnership announced with the American Federation of Teachers in October 2023. The company raised $3.5 million in seed funding (2023) and $10 million in a Series A (2024), and was later acquired by Superhuman.

A peer-reviewed benchmark of 14 detection tools (Weber-Wulff et al., International Journal for Educational Integrity, 2023) found that none exceeded 80% accuracy and only five exceeded 70%, concluding that "the available detection tools are neither accurate nor reliable," with performance degrading further once text is lightly paraphrased. Press investigations (Futurism, Washington Post, Ars Technica) reported teachers relying on GPTZero to falsely accuse students — in one case nearly 20% of a class — including flagging the U.S. Constitution itself as AI-written. In February 2025, a Yale School of Management EMBA student sued the university alleging he was wrongly suspended for a year based on a GPTZero flag; a federal judge denied his injunction request in May 2025.

GPTZero is included as a documented cautionary case: it shows both how quickly an AI-detection tool can scale into near-universal classroom use, and how thin the underlying evidence base for that use can be, with real, litigated consequences for students. Institutions using such tools without independent verification of accuracy risk exactly this outcome.