The student-using-AI question, framed honestly · AI for Teachers

Before you read another line, accept the fact: your students are using AI. Not some of them. Not the ones with rich parents. Your Class 8 students with shared phones from older siblings, your +2 students copying the link from a friend’s WhatsApp, your SEE candidates desperate for one more practice essay at midnight, your undergraduate students typing prompts between classes in the canteen — all of them, in some proportion you cannot control, are typing into chatbots and pasting answers into their work. This was true last term. It will be more true next term. The only question that remains for you, as the teacher, is what you do about it.

The three positions, honestly named

Teachers in Nepal, like teachers everywhere, have landed on roughly three responses to the student-AI question. Two are common and damaging. The third is harder and is the only one that actually works.

Position one: the blanket ban. “No AI in this class. Any work done with AI is treated as cheating.” This is the position most schools have officially adopted, often in a single staff meeting after a viral news story. It feels firm. It is also, in practice, unenforceable. You cannot detect modern AI-written prose reliably — the much-advertised detectors produce false positives at rates that would fail any real evaluation, and they fail more often on Nepali-English bilingual students whose natural English already reads as a little formal. The ban does not stop the use; it only stops the conversation about the use. Students who would have asked you “is it okay to use this for the brainstorm?” learn, instead, to hide. The ban breeds dishonesty. The students who keep talking to you are the ones who never used it in the first place.

Position two: pretend it does not exist. The bigger sin. You set the same homework you would have set in 2018, you grade it as if 2018 still applied, and you tell yourself the children must have written it. This is the position adopted by the largest number of Nepali teachers right now, and it is the most damaging because it leaves students to learn the technology entirely on their own — at exactly the age when they will form bad habits that will hurt them at +2, at undergraduate level, and at work. A child who learns to outsource her thinking to a chatbot at fourteen, with no adult guiding the boundary, is in trouble. Not because she used the tool, but because no one ever sat with her and helped her understand what learning actually requires.

Position three: honest engagement. This is the harder option, and it is the only one that does not slowly damage either the children or you. It has three parts.

Part one: a written policy your students actually understand

Not a school-wide policy from on high. Your class policy, written by you, in language a Class 8 student can read, posted on the wall and revisited every term. It says, in plain Nepali and English: here is when AI is okay to use in this class, here is when it is not, here is what I expect you to tell me if you used it. The detail matters and depends on the subject. In a literature class the brainstorm may be fine and the essay must be the student’s own; in a maths class the model may be used to check a method but the working must be shown; in a coding class the model may help with syntax but the student must be able to explain every line.

A policy like this is not a substitute for trust. It is a way of making the conversation possible. A student who knows the rules can ask honestly when they are unsure. A student who does not knows only that everything is forbidden, and therefore that hiding everything is the only safe move.

Part two: a conversation about what learning actually requires

Your students know — in some part of themselves — that copying an AI’s essay does not teach them to write. They also know that finishing the homework matters tonight, and that the question of whether they learned anything is for next month, when the marks come back. The job of the teacher, the job that has not changed and will not change, is to keep reconnecting the short-term incentive with the long-term reality.

The honest conversation goes something like this. “When you ask the chatbot to write your essay, the essay gets written. Your skill at writing does not improve. When the SEE comes, the chatbot is not in the exam hall. When the +2 viva comes, you cannot paste the question. When you sit down at a job at twenty-three and someone asks you to explain your reasoning, the model is not the one being interviewed. You are. The work you skip now is paid for later, with interest.” This is not a lecture you give once. It is a conversation you return to, in different forms, across the year.

Part three: assessment that makes the learning visible

If a take-home essay can be written by a chatbot in three minutes, the take-home essay is no longer measuring what it was designed to measure. This is not the AI’s fault. The assessment was always measuring a proxy — “the student sat at home and wrote something” — and the proxy has now broken.

Productive responses, in the Nepali context where invigilated written exams are still the backbone: in-class writing, more often, even short; oral defence of the take-home — five minutes per student where they explain a paragraph of their own essay aloud; process artefacts — the brainstorm, the draft, the revision — submitted alongside the final; viva-style spot questions during marking — “you used the word paradigm here; what does it mean in your own words?” None of this is novel pedagogy. All of it is what good teachers did when they were not lulled into believing that take-home written work could be trusted to measure individual skill.

You do not need to rebuild your assessment overnight. You need to know, for each task you set, what it measures and whether that measurement is still valid in a world where the take-home draft is partially or fully AI-written. Some tasks are fine — a worksheet of computational maths, for instance, is robust to AI in ways an essay is not. Others need redesign. Walk through your tasks once, mark them honestly, and start with the ones that matter most.

The scenario: a Class 9 student admits she used ChatGPT

It is Thursday after class. Ramila, one of your better Class 9 students, hangs back. She tells you, with some embarrassment, that the social studies essay she handed in this morning was partly written by ChatGPT — she did the research, but the chatbot wrote the body paragraphs. She wants to know what to do.

What you do in the next five minutes shapes the next five years of how she relates to her own learning, and how the rest of the class will relate to you when this conversation, as it will, gets repeated to a friend over momos.

The unproductive responses: an immediate zero and a disciplinary referral (she will not tell you next time, and neither will her classmates); a lecture on dishonesty (true but not useful here, because she came to you); pretending it is fine (it is not).

The productive response is a conversation. What did the model write, and what did you write? Have her open the essay and walk through it, paragraph by paragraph. What did you learn from the parts the model wrote — did you read them carefully, did you understand the arguments, can you defend them now? What did you skip by using it? Often the answer is the part she would have learned the most from. What would the work have looked like if you had used the model for the brainstorm and written the paragraphs yourself? That is the policy you wish she had used. Now name it and make it the policy.

For this essay, an outcome. Perhaps she rewrites the body paragraphs herself this weekend and you re-mark. Perhaps she defends the essay orally on Monday. Perhaps the mark stands but with a written reflection. The detail is less important than the principle: a student who told you the truth is met with a real consequence and a real path forward, and the class sees that honesty is the productive move.

What this section is not, and what comes later

This is not the chapter on academic integrity. The full treatment — the policy templates, the assessment redesign patterns, the difficult conversations with parents and headmasters, the question of what to do when the student does not admit it and you only suspect — comes later in this course, in Chapter 6. This section is only the framing: that the question is not whether to engage with student AI use, but how, and that honest engagement starts with the three positions named honestly.

Check your understanding

Quick check

—

A Class 9 student comes to you after class and admits that part of her social studies essay was written by ChatGPT. What is the most productive first response?

Give her an immediate zero and refer her for disciplinary action, so the rest of the class learns the seriousness of the issue.
Tell her not to worry about it — every student is using AI now and the technology is unavoidable.
Have a conversation: what did the model write, what did you write, what did you learn versus what did you skip — then agree a real consequence and a real path forward (rewrite, oral defence, written reflection) that names the policy and treats her honesty as the productive move.
Run her essay through an AI-detection tool to confirm her admission before deciding what to do.

What comes next

This chapter has mapped the terrain — what AI does well, where it lands in your week, and how to think about your students using it. The next chapter moves into the first hands-on territory: using AI to plan lessons and produce teaching materials, with practical recipes you can use on Sunday evening for Monday morning’s class.