Sarah Chen has been teaching college English for 14 years. Last semester, she estimates that roughly 40% of the papers submitted in her freshman composition class were partially or fully written by AI. She knows this not because a detection tool flagged them, but because she's read enough student writing to recognize when a paper is suspiciously polished, oddly generic, and devoid of the specific mistakes that real learning looks like.
She can't prove any of it. And that's the problem.
Detection Is a Lost Cause
Every AI detection tool on the market — GPTZero, Turnitin's AI detector, Originality.ai — has the same fundamental flaw: they produce false positives. Students who are naturally good writers, non-native English speakers who carefully revise their work, and anyone who uses Grammarly gets flagged. Accusations based on these tools have led to wrongful academic misconduct charges, and several universities have quietly stopped relying on them.
OpenAI itself tried to build a classifier to detect AI-generated text. They shut it down after six months because it was wrong too often. If the company that built ChatGPT can't reliably detect its output, what chance does a $10/month SaaS tool have?
The Arms Race Nobody Can Win
Students aren't naive. The ones using AI to cheat have adapted rapidly:
- They prompt models to "write like a B+ student, not an A+ student"
- They ask for intentional minor grammar errors
- They run output through paraphrasing tools
- They write the first draft themselves and use AI to improve it — a gray area that defies simple categorization
For every detection method, there's an evasion technique. It's an arms race where the defense is always one step behind.
What Smart Institutions Are Doing Instead
The universities that seem to be handling this best have stopped trying to detect AI and started redesigning their approach to assessment entirely:
In-class writing has made a comeback. If the essay is written during a proctored session, authorship isn't in question. It's old-fashioned, but it works.
Process-based grading evaluates outlines, drafts, revision notes, and the final product together. AI can produce a finished essay, but fabricating a convincing revision process is much harder.
Oral examinations are returning to disciplines that abandoned them decades ago. If a student can discuss and defend their paper in depth, it doesn't much matter who wrote the first draft.
AI-integrated assignments take the radical approach of making AI use part of the assignment. "Use ChatGPT to generate a first draft, then critically evaluate and improve it." This teaches the skill students will actually need in the workplace.
The Bigger Question
Underneath the cheating debate is an uncomfortable question: if AI can write a passable college essay in 30 seconds, what exactly are we testing when we assign one? The answer might force a rethinking of education that goes far deeper than plagiarism policies.