It happens more and more. The teacher reads an assignment, something does not fit (the style, the vocabulary, the perfect structure) and knows what he is seeing. Or the student leaves it visible: the uncleaned text, the strange quotation marks, the paragraph that has nothing to do with what was requested…
And then comes the question that no one prepared: What am I doing now?
This guide does not go on how to detect if an assignment is done with AI. It’s about what to do when you’ve already detected it, or when you suspect yes. And above all: what should your center have so that you don’t have to improvise every time it happens.
The question is not whether students use AI, the question is what decision your center has made before that happens. |
First of all: distinguishing between substitution and support
Not all uses of AI in an assignment are the same. Before acting, it is advisable to make this distinction:
Substitution of thought: The student delivers something that the AI has generated and that he has not processed or built. The work does not reflect your learning.
Support for the process: The student used AI in some phase (to look for ideas, review structure, correct writing) but the work is his.
That difference matters. Not because the second is always acceptable, but because the consequences and the conversation that must be had are different in each case. The problem is that many centers do not have defined what each thing is. And without that definition, the teacher cannot act in a coherent or defensible way.
What not to do first
Do not sanction without a previous institutional position
If the center does not have any documents that establish which uses of AI are acceptable and which are not, the teacher does not have the basis to sanction. It can generate a conflict with the family, a formal claim or a situation without institutional support. Before acting disciplinary, you have to have something written.
Do not blindly rely on automatic detectors
Tools like Turnitin AI Detection or GPTZero have a relevant false positive rate. They have marked as generated by the texts written by people with certain styles of writing, people who write in their second language or very structured texts. They are indicators, not tests.
Do not generalize the answer
Applying the same consequence to the entire class or to very different types of use is not coherence, it is a reaction. The reactions generate more problems than those they solve.
What to do: Specific steps for the teacher
1. Talk to the student before doing anything
The first answer is the conversation. Ask the student to explain his work orally: to develop an idea, to justify a decision, to explain a source. If your homework is yours, you can do it. If not, it will be evident without the need for accusation. This conversation also opens something more important: the possibility of working the criteria.
2. Document what you have observed
Before scaling, document: what you have observed, in what assignment, with what evidence. Not to start a file, but to have a record if the case progresses or if the family asks.
3. Check if the center has protocol
If your center has an AI usage policy, apply it. If you don’t have it, point it out to the Head of Studies. Not as a complaint, but as a real need: this will continue to happen and each teacher cannot manage it differently.
4. Decides the pedagogical consequence, not only the sanction
The most effective consequence is usually pedagogical: ask the student to redo the work by demonstrating real understanding, presenting his conclusions orally, or reflecting on what he delegates when he uses AI without criteria. That teaches more than a zero.
The student who uses AI without criteria does not have a bad intention in most cases. He has an absence of criteria and that is worked on. |
The underlying problem: the center has no position
If every time something like this happens the teacher has to improvise, the problem is not the student. It is that the center has not made an institutional decision on how AI is used.
That decision includes: what uses are acceptable in what types of tasks, how the students are informed, what consequences exist if the conditions are not respected, and who responds if there is a conflict. Since February 2025, Article 4 of the European AI Regulation requires centers to ensure that their personnel have the necessary literacy to operate and supervise the AI systems they use. Most centers do not have it documented.
What does your center need to have?
— An institutional position on the use of AI in academic tasks: what is allowed, under what conditions and with what level of declaration by the student.
— A protocol for when a use contrary to that position is detected: who decides, what consequences there are and how it is documented.
— Real formation of the cloister: not a ChatGPT workshop, but a shared framework so that teachers can make coherent and defensible decisions.
Frequently Asked Questions
What do I do if I find out that a student has used ChatGPT to do a job?
Before acting, it distinguishes between use as a substitution of thought and use as support for the process. The first answer should not be the sanction: it should be the conversation. Without a center protocol, the teacher cannot act in a coherent or defensible way.
How do I know if a job has been done with AI?
Automatic detectors have high error rate and are not reliable test. What works: Know the real level of the student, ask oral questions about the work and observe inconsistencies of style. No detector replaces the teaching criteria.
Should you prohibit the use of AI in jobs?
The prohibition without criteria does not work: the students will continue to use it and will learn to hide it better. What works is to define which use is acceptable in each type of task, under what conditions, and what level of declaration. That requires leadership, institutional position, not teaching improvisation.
