AI detectors are creating a new era of distrust
Educators and publishers are increasingly relying on AI detection tools to identify AI-generated writing, echoing the earlier rise of anti-plagiarism software. Unlike traditional plagiarism checkers, which compare text against databases of existing content, AI detectors such as GPTZero, Pangram and Turnitin's own tool use their own AI models to guess whether writing was produced by a human, analysing wording, rhythm and structure. This approach is inherently more subjective than direct text-matching and risks unfairly flagging writers, including non-native English speakers, even though the companies behind these tools claim very low false-positive rates.
A Center for Democracy and Technology survey found 43% of US teachers in grades six to 12 regularly used AI detectors between 2024 and 2025, with some universities finding Turnitin had enabled AI detection automatically since 2023. Turnitin claims a false-positive rate below 1%, Pangram cites 1 in 10,000, and GPTZero reports similarly low figures, but critics argue the underlying analysis remains murky. The growing use of these tools has fuelled online accusations of "sounding like AI," with real-world consequences: publisher Minotaur reportedly dropped a $2 million book deal last month over AI-writing suspicions.
- AI detectors are widely used by teachers and publishers despite doubts over accuracy
- 43% of US middle/high school teachers used AI detectors in 2024-2025
- Minotaur dropped a $2m book deal amid AI-writing accusations
New here? Start with this
Educators and publishers have started using AI detection tools to work out whether a piece of writing was produced by a person or by an artificial intelligence system. These tools, which include products called GPTZero, Pangram and Turnitin, work differently from older plagiarism checkers: rather than comparing text against a database of existing work, they use their own AI systems to make a judgement based on the style, wording and rhythm of the writing. That judgement is inherently more of a guess than a direct match, which raises the possibility of writers being wrongly accused, including those who are not native English speakers.
Turnitin, a company long used by schools and universities to check for plagiarism, is a key player, alongside newer specialist firms such as GPTZero and Pangram. These tools have become widespread in classrooms and are increasingly used by publishers too, even though the companies involved say wrongful flags are rare. Independent verification of those claims is difficult, since the way the tools reach their conclusions is not fully transparent.
The issue matters because being wrongly labelled as having used AI can carry real consequences, from academic penalties for students to lost publishing deals for authors. As AI-generated text becomes harder to distinguish from human writing, the reliability of these detection tools, and the trust placed in them, has become a subject of growing debate.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Educators and publishers argue that some means of detecting AI-generated work is essential to preserve academic integrity, protect the value of genuine authorship, and maintain trust in credentials and creative work; without such checks, students could gain unearned qualifications and publishers could unknowingly market machine-written material as original human work. Given the scale of AI adoption, they contend that imperfect detection tools, used as one signal among several and refined over time, are a reasonable stopgap while more reliable methods are developed, and that companies' reported low false-positive rates suggest the risk of wrongful accusation is manageable relative to the benefit of deterrence.
The case against
Critics, including many educators, writers and civil liberties advocates, argue that AI detectors rely on opaque, probabilistic judgements about style rather than verifiable evidence, making them fundamentally unsuited to decisions with serious consequences such as expulsion, failing grades or lost book deals. They point out that non-native English speakers and writers with unusual but genuine styles are disproportionately vulnerable to misclassification, that claimed false-positive rates are unverified by independent audit, and that the burden of proving one's own innocence against an algorithm's guess is unfair and corrosive to trust between institutions and the people they serve.
AI Art Business Celebrity Culture Entertainment Markets Technology