Teachers

You Think a Student Used AI. You Can't Prove It. Here's What to Do.

May 2026

By Mark Reeves — author of all 47 AI Field Guides. Written and tested in a working business.

AI-generated portrait. Pen name. Real work.

The honest guide to AI suspicion, detection tools, and what actually holds up.

The Paper Lands on Your Desk

The writing is too clean. The vocabulary doesn't match what you've heard in class. The argument is structured in a way this student has never structured an argument before. You've been doing this long enough to notice.

But you can't prove it.

That's not a failure of your instincts. That's the reality of where AI detection technology actually is right now.

AI Detectors Are Not Reliable Enough to Act On

This is the thing schools need to hear more clearly: AI detection tools have false positive rates that can run as high as 50% depending on the tool, the text, and the student. That means up to half of the flagged papers could be original work.

A 2023 study from Stanford found that AI detectors disproportionately flag writing by non-native English speakers — often because their syntax is more formal and less idiomatic, which the detectors read as machine-generated. Students with dyslexia or other language-processing differences are also flagged more often.

If you act on a detector result alone and you're wrong, you've accused a student of academic dishonesty they didn't commit. That has consequences — for the student, for your relationship with them, and potentially for the school.

The tools aren't there yet. Using them as your primary evidence is a problem.

What Actually Helps: Process-Based Assessment

The most effective shift you can make is moving from product-based to process-based assessment — and it doesn't require detecting AI at all.

If a student has to turn in a rough draft, then a revised draft with a change log, then a final version, the AI shortcut becomes harder to hide. Not impossible, but harder. You can see the thinking move across drafts. You can see where they struggled and where they pushed through.

In-class writing still matters here too. A timed in-class paragraph on the same topic as the at-home essay gives you a baseline. If the register is completely different, you have something to talk about — not as evidence of cheating, but as a discrepancy worth understanding together.

This isn't about surveillance. It's about building assessment structures that show student thinking, not just student output.

Ask the Student to Explain Their Work

This is underused and it works.

If you have genuine concern about a submitted piece, ask the student to walk you through it. Pick a specific argument or passage and say: "Tell me more about your thinking here. How did you get to this conclusion?"

A student who wrote the work can do this. A student who submitted AI output without engaging with it usually cannot — or will give you a generic explanation that doesn't track with the specific reasoning in the text.

This isn't an interrogation. Frame it as a learning conversation. "I'd love to hear more about your process on this one" lands differently than "I think you cheated." One opens a door. The other closes it.

If the student can't explain their own paper to you, that's real information. And if they can — even if the paper is unusually polished — that's real information too.

Redesigning Assignments AI Can't Easily Complete

Some assignments are more resistant to AI than others, and the difference is specificity.

"Write about a theme in this novel" is easy for AI. "Write about the moment in chapter 12 where the character's choice contradicted what they said in chapter 3, and connect it to something you've seen in your own life" is not. The more personal, the more locally specific, the more the assignment requires something only the student has — a memory, a classroom observation, a conversation from last week — the harder it is to outsource.

Oral defenses of written work. Annotations in the student's hand. A "what surprised you" reflection attached to every submitted piece. These aren't perfect, but they raise the floor significantly.

The goal isn't to make AI unusable. It's to make the student's irreplaceable perspective necessary.

What Not to Do

Don't file academic integrity charges based on a detector output alone. The tools are not admissible as standalone evidence in any educational context I'm aware of, and using them that way exposes your school to real complaints — especially from families of non-native English speakers or students with documented learning differences.

Don't assume. The student who submits an unusually polished paper might have a parent who's an editor, a tutor who gave significant feedback, or a peer who helped them. That may still violate your policy, but it's a different conversation than "you used AI."

And don't skip the conversation. The worst outcome is a student who used AI because they were overwhelmed, scared, or didn't understand the policy, and you respond with punishment before you understand what happened.

The Policy Gap Is Real

Most schools still don't have a clear AI policy, and teachers are being left to make judgment calls without institutional backing. That means you're often making this up in the moment — which is exhausting, inconsistent, and unfair to students who don't know what the rules are either.

The most defensible position is a documented, communicated policy, applied consistently, with room for conversation when cases are ambiguous. Teachers without that institutional backing are personally absorbing every judgment call. When a complaint escalates, the teacher who made an inconsistent call is exposed — the institution that didn't give them a policy is rarely the one who faces the parent. If your school doesn't have that yet, you're not alone — and you're not wrong to be frustrated about it.

Try This With Your Next Suspect Assignment

Before you raise a concern with a student, run this test. Write out the three things that made you notice: the vocabulary difference, the structural shift, the argument pattern. Now write one specific question you could ask the student about the passage that would require them to have engaged with the material themselves — not "did you write this?" but "in paragraph three you argue X — walk me through how you got there."

If you can write that question, you have a conversation. If you can't, you need more specific evidence before you proceed. The discipline question always comes second to the understanding question.

This article covers what to do when you suspect AI use but can't prove it. It doesn't cover three things teachers need alongside it:

How to write your school's AI policy before the next disputed case arrives. The conversation with a student is much harder without institutional backing. The policy problem is where most of the vulnerability lives for individual teachers.

How to use AI yourself for the grind work. Lesson plans, rubrics, IEP language, parent emails — the same tools your students are using can cut your Sunday night down significantly. That context matters for how you talk to students about AI.

How to redesign assessments so that AI can't complete them. The conversation-based approaches above help in the moment. Assignment redesign prevents the problem at the source.

Related reading

• Will AI replace teachers?
• How to know if an AI answer is actually true

Mark Reeves is a pen name. AI Field Guide publishes role-specific, practical guides for using AI tools in real work.