Students already use AI: bans fail, do this instead

Why a ban does not hold

A ban works when violations are visible and the cost of being caught exceeds the benefit. Neither condition holds here. The tool is free, it runs on a phone, it leaves no trace, and AI-text detectors get it wrong in both directions — they flag the honest writing of non-native speakers and they flag carefully edited drafts. An accusation built on a detector percentage falls apart in the first conversation and costs you the trust of the whole group.

What actually happens after a ban

Usage does not drop, it goes underground. Nobody asks you how to verify a claim, because asking would expose them.
Inequality grows. Cautious students abstain and lose time; bold ones use it quietly and win.
You lose your data. You no longer know who can actually do what, because submissions have stopped representing people.

The workable alternative: three modes

Instead of "not allowed", assign a mode to each task and write it into the wording. Three modes are enough, and they are legible to a school pupil and to an adult in corporate training alike.

MODE A — no AI. Work is done in front of me: in class, orally, on paper, within a set time.
MODE B — AI as assistant. Allowed for: finding ideas, checking spelling, explaining something you did not follow.
   Required: attach the prompts you used and what you rejected from the answers, and why.
MODE C — AI as the object of study. The task is to get a model answer, find the error in it, and correct it with sources.
Every assignment header carries: MODE ___ . No label means Mode B.

Insight. The mode system removes the core conflict: a student no longer chooses between honesty and results. They follow a rule, and you set that rule to match the goal of that particular task.

Common mistake. Putting Mode A on everything. That is a ban in new clothing, and it turns a course into a series of proctored tests. Mode A makes sense for foundational skills that must work without support.

Pro tip. Decide in advance what happens in Mode B when someone attaches no prompts. The cleanest response is to treat the work as a draft and request the process record. That is a sanction for a procedural failure, not an accusation of cheating, and it cannot be argued with.

Cheat sheet

AI-text detectors are unreliable; a percentage is not evidence of anything.
A ban does not reduce use, it makes use invisible to you.
Three modes — no AI, AI as assistant, AI as object of critique — go in the assignment header.
Sanction the missing process record, never the suspicion.

1. Why can an accusation not rest on an AI-detector score?

2. What does Mode C mean in this scheme?

3. What is the right response when a Mode B submission has no prompt record?

Task — checked by AI

Take three real assignments from your next two weeks and assign each one Mode A, B or C. For each, write the exact line you will paste into the assignment wording, plus one sentence on what you will do if the mode is broken. Then explain why one of the three got Mode A rather than Mode B.

← Back

🔒 Answer the question correctly to move on to the next lesson.