I have been teaching college English for sixteen years. I have read thousands of student essays. I have seen every kind of writing - brilliant and terrible, inspired and plagiarized, sincere and cynical. I know what student writing sounds like. And I am now required by my department to submit every essay my students write to an AI detection tool that I know, with certainty, sometimes gets it wrong.
This is the essay I cannot assign to my students, because it is about them.
The Policy
Last fall, our department adopted a mandatory AI detection policy. Every written assignment in every section of every course must be screened before grading. The tool we use - I am not permitted to name it, per my employment agreement - assigns a percentage score indicating the probability that the text was AI-generated. Submissions scoring above 40% are automatically flagged for review.
I voted against this policy. I was in the minority. My colleagues are not unreasonable people. They have seen the same articles I have about students submitting AI-generated work. They are worried about academic integrity, about the value of the degrees they are signing, about what it means to teach writing in a world where writing can be automated. These are legitimate concerns. I share them.
What I do not share is the confidence that a statistical tool can address them.
The First False Positive
Her name was Amara. She was a sophomore, pre-med, taking my composition course as a requirement. She sat in the third row and rarely spoke in class, but her essays were precise, thoughtful, and quietly beautiful. Her first assignment - a personal narrative about growing up bilingual - scored 52% on the AI detector.
I knew it was a false positive. I had watched Amara work through three drafts in my office hours. I had read her outline, her freewriting exercises, her peer review comments. I knew the essay was hers because I had watched her write it. But the policy required me to flag it, which meant a meeting with Amara, a conversation about academic integrity, and the implicit accusation that her work might not be her own.
She cried. Not dramatically - quietly, the way you cry when you have been unjustly accused and lack the institutional power to fight back. "I wrote every word," she said. I told her I believed her. I told her the flag was a formality and that I would clear it. She nodded. She stopped coming to office hours after that.
The Damage
Amara's case was resolved in a week. The formal notation was removed from her record. No penalty was imposed. By every institutional metric, the system worked. But something had broken that cannot be measured by institutional metrics: the trust between a teacher and a student, the fragile willingness to be vulnerable that makes learning possible.
Amara still submits her essays. They are competent. They are no longer beautiful. She writes safely now - shorter sentences, simpler vocabulary, nothing that a machine might mistake for machine-generated prose. She learned the lesson the detector taught her: don't write too well. Don't take risks. Don't sound like yourself if yourself sounds too polished.
She is not the only one. Across my three sections, I have noticed a flattening. Students who wrote with personality now write with caution. The messy, experimental, sometimes-failing-but-always-trying quality that characterizes the best student writing has been replaced by something careful and generic. They are not writing for me anymore. They are writing for the algorithm.
What I Cannot Say
In department meetings, I cannot say what I believe: that we have substituted a tool for judgment, that we have outsourced our expertise to a probability model, that the very act of requiring detection communicates to students that we do not trust them - and that the most devastated by that message are the students who deserve trust the most.
I cannot say that the real crisis is not students using AI. The real crisis is that we have decided to address that problem by surveilling all students equally, as if the guilty few justify the suspicion of the innocent many. I cannot say that education requires trust, and that trust cannot survive institutional suspicion.
I say these things here, instead, in the hope that someone who makes policies will read them and understand what is happening in the classrooms where those policies land. My students think I don't trust them. They are wrong - but the institution I work for has made it impossible for me to prove it.