AI Detectors Don't Work. Here's What Schools Are Doing Instead.

The fantasy was simple: students would use AI, schools would run the text through a detector, and the truth would come out with a percentage.

That fantasy did not survive the semester.

AI detectors promised certainty in a situation built out of uncertainty. They were asked to answer a question that writing has never made easy: who, exactly, produced this sentence?

Why detectors fail

Detectors look for patterns. But student writing is full of patterns. So is second-language writing. So is highly edited writing. So is boring writing written by a tired human at 1 a.m.

A detector score is not a confession. It is a signal with error bars most people never see.

False positives are the nightmare case. A student who wrote their own work can suddenly be forced to prove a negative against software they cannot inspect.

The arms race problem

Even when detectors catch some AI text, the game changes fast. Models change. Students paraphrase. Other tools rewrite. Detection becomes less like a lock and more like a weather forecast delivered after the storm.

Schools are realizing that punishment built on uncertain evidence is fragile.

What schools are doing instead

The better shift is toward process-based assessment:

• in-class writing
• oral defenses
• draft checkpoints
• source annotations
• revision histories
• personalized prompts
• reflections explaining choices

These methods are not perfect, but they make the student's thinking more visible.

A professor who asks you to explain why you chose a source learns more than a detector score ever could.

What this means for students

Keep your process. Save outlines. Keep drafts. Use version history. Write notes about why you made decisions. If you use AI, document how.

That may sound paranoid, but it is really just good academic hygiene.

The bigger lesson

AI detectors failed because they tried to turn trust into a number. Education needs something sturdier than a percentage with a scary color beside it.

The future is not better surveillance. It is better assignment design.

If schools want authentic work, they have to ask for work that reveals the person behind it. A detector cannot do that. A real conversation sometimes can.