It was a Tuesday in October when I finally admitted I had a problem. I teach AP U.S. History, and I had just read my third essay in a row that used the phrase "it is important to note." My students do not talk like that. I knew what had happened, and I also knew I had no written policy to point to. I had said, at the start of the year, something vague like "don't let AI do your thinking for you." That is not a policy. That is a wish.
I spent the next few weeks doing what teachers do: researching, drafting, throwing it out, starting over. What I eventually built was not a list of punishments for AI use. It was a set of clear boundaries that told students exactly what AI was for on each assignment. The difference matters more than you would think.
Rules Get Broken. Boundaries Get Understood.
A rule says "no AI on essays." A boundary says "on this essay, you are doing the thinking. AI can help you check spelling after you have a complete draft, but the argument is yours." One of those statements invites a workaround. The other one explains the reason, sets the limit, and makes the student's job clear.
When I switched from rules to boundaries, the conversations with students changed completely. Instead of "did you use AI on this?" I was asking "where did your argument come from?" That is a much more interesting conversation, and it is one students can actually have honestly. The goal of figuring out how to set AI boundaries in the classroom is not to catch students. It is to give them a framework they can actually follow.
A boundary explains the reason. A rule just sets the limit. Students follow boundaries because they understand them, not because they are afraid of getting caught.
This is the foundation of any good classroom AI policy. You are not writing a list of consequences. You are writing a description of what learning looks like in your room, and what role AI plays in that, or does not play in that, depending on the task.
Not Every Assignment Needs the Same Level
Once I stopped thinking about AI as something to ban or allow entirely, I started thinking about it the way I think about other tools. A calculator is appropriate for some math. A dictionary is appropriate for some writing. Neither of those is appropriate everywhere, and no one thinks that makes the tool bad.
Here is how I actually think about the spectrum in my history classes, and I think it translates to most subjects:
- Research and background reading: I allow AI here. If a student asks an AI to explain the causes of the Civil War before we discuss it in class, that is fine. It is no different from reading a summary. The thinking comes later.
- Essay brainstorming and outlining: I allow AI as a thinking partner, not a writer. Students can ask it questions, argue with it, use it to stress-test a thesis. They cannot ask it to write the outline for them.
- Essay drafting: No AI. This is where the actual skill lives. I want to see how a student constructs an argument when it is hard, not when it is easy.
- Revision after a complete draft: Limited AI use is fine. Grammar checks, clarity suggestions, reading it aloud. The ideas must already exist on the page.
- In-class timed writing: No AI. Full stop. This one is simple.
If you teach a subject with math problem sets or lab reports, the same logic applies. Letting a student use AI to check their work after solving a problem is very different from letting AI solve the problem first. The boundary is about where in the process AI enters, not just whether it appears at all.
Put the Boundary in the Assignment, Not Just the Syllabus
This is the single most practical change I made. My syllabus has a section on AI use. Nobody rereads the syllabus in March. What actually works is writing the AI expectation directly into each assignment sheet, in plain language, before the instructions start.
For a document-based question essay, I write something like: "AI use on this assignment: You may use AI to help you understand unfamiliar vocabulary in the primary sources. You may not use AI to write any part of your response or to generate arguments. Your analysis must be your own." For a research project, it looks different: "AI use on this assignment: AI tools are permitted for background research and for helping you organize your notes. All writing must be original."
When the AI rules for students are embedded in the assignment itself, they cannot claim they forgot. More importantly, they have something specific to refer to. It removes the ambiguity that leads to bad decisions, and it removes the ambiguity that leads to unfair situations where one student asks you what is allowed and another just guesses.
Some teachers I know have started using platforms that let you set an AI access level per assignment rather than tracking this manually. Authentiya does exactly that: you configure what AI assistance is appropriate for each task, and it gives you visibility into how students are actually working. That kind of structure makes the teacher AI guidelines you have already written actually enforceable, without you having to police every submission yourself.
What to Do the First Time a Student Pushes Back
It will happen. A student will tell you that AI is the future, that banning it is pointless, that every job uses it now. Some of them will say this because they are genuinely frustrated. Some will say it because they do not want to do the work. Either way, your answer is the same.
I tell students: I am not preparing you to use AI. I am preparing you to think. AI is useful when you already know what you are trying to say and you need help saying it better. It is not useful when you do not yet know what you think, because it will give you an answer that sounds good but is not yours. That is actually worse than writing a messy, imperfect essay that came from your own head.
The students who push back the hardest are usually the ones who are most anxious about their own writing. I try to name that directly. "I think you are worried your essay won't be good enough without help. Let's talk about that." That conversation is almost always more useful than any policy enforcement.
If a student crosses a boundary you have clearly set, the first conversation should be a conversation, not a referral. Ask them to walk you through their process. Ask where their ideas came from. Most of the time, students know when they have crossed a line, and a direct conversation is more effective than a punishment. Save the formal process for repeated or severe situations.
Start Simple and Revise as You Go
I did not get my classroom AI policy right on the first try. I still adjust it. Some assignment types I thought would be fine with no AI turned out to need more structure. Some I thought needed tight restrictions turned out to be fine with more openness. The goal is not a perfect policy on day one. The goal is a clear, honest framework that your students understand and that you can actually maintain.
Write it into your assignments. Talk about it out loud with your students. Revisit it after the first major project. The teachers who struggle with AI in their classrooms are usually the ones still trying to manage it with a vague "don't cheat" statement. The ones who have figured it out are the ones who sat down and wrote something specific, put it in front of students early, and treated it as a living document rather than a one-time announcement.
You do not need to have every answer before you start. You just need to start.