Last spring, a student turned in an essay that was technically flawless. Perfect transitions, no grammar issues, a clean thesis. It also had absolutely nothing to say. The ideas were vague in that specific way AI gets vague when it has no real stake in the topic. I asked her to talk me through her argument, and she couldn't. She hadn't written the essay. She hadn't even really read it.
That was the moment I stopped pretending I could just ban my way out of this.
I'd already tried the hard-line approach. Plagiarism warnings on every assignment, AI detector screenshots in my slides, the whole performance. It didn't stop anything. It just made students sneakier and made me paranoid. I was spending more time trying to catch cheating than actually teaching writing. That's not a trade-off I'm willing to make.
But I also wasn't ready to throw up my hands and let students treat ChatGPT like a ghostwriter for every assignment. There had to be a middle path, and figuring out what that looked like in a real classroom took me the better part of a year.
Why Banning AI Hasn't Worked
The schools and teachers who went full prohibition in 2023 are mostly quietly walking it back now. The reason is simple: the ban was never enforceable in the first place. AI detectors flag innocent students and miss guilty ones. Students use phones, home computers, accounts their teachers can't monitor. And frankly, the tools are everywhere. Telling a teenager not to use AI is like telling them not to use Google. You can say it, but you're not going to win.
More importantly, the ban approach teaches the wrong lesson. The students who will be entering the workforce in four to eight years are going to be expected to know how to work with AI tools. Companies are already building AI into their workflows. If we spend the next few years treating every AI interaction as cheating, we're not protecting academic integrity. We're just leaving students unprepared.
That said, "AI is everywhere so let anything go" is equally bad advice. The student who can't articulate her own argument because she never wrote one is not prepared for anything. She just has a clean essay. Academic integrity in the AI era isn't about catching cheaters. It's about making sure students are actually doing the cognitive work that school is supposed to produce.
What Responsible AI Use Actually Looks Like
Responsible AI use for students isn't about following a rule. It's about understanding what the assignment is actually trying to build. When I assign a personal narrative, I'm trying to build voice, memory, and the ability to reflect. If a student uses AI to generate that narrative, none of those things get built. The assignment fails regardless of whether the final product looks polished.
But when I assign a research summary, the skill I'm after is synthesis and source evaluation. If a student uses AI to help organize notes they've already taken, that's different. They're still doing the intellectual work. The AI is handling formatting, not thinking.
Here's a concrete example. One of my juniors was working on a literary analysis of The Great Gatsby. She was stuck on her thesis. I told her she could use AI to brainstorm three possible thesis angles, but she had to pick one and write the full essay herself without further AI assistance. She came back with a thesis that was genuinely hers, one she'd had to argue for in our conference, and an essay that reflected actual engagement with the book. That's responsible AI use. She used it as a thinking partner, not a replacement for thinking.
Another student in the same class used AI to help him check his grammar on a final draft after he'd written the whole thing himself. Also fine. The writing was his. The AI caught some comma errors. Nobody lost anything.
The question worth asking isn't "did the student use AI?" It's "did the student do the thinking this assignment was designed to develop?"
A Simple Framework: Three Levels of AI Access
After a lot of trial and error, I landed on a framework I now use for every assignment. I borrowed pieces of it from colleagues, adapted some from conversations with other English teachers, and stress-tested it with two full semesters of actual students. It has three levels, and I put the level right on the assignment sheet so there's no ambiguity.
Level 1: No AI. This means exactly that. No AI tools at any stage. I use this for in-class writes, personal essays, exams, and anything where the whole point is to see what the student can do on their own. I'm transparent about why. I'll say something like, "This one needs to be entirely yours because I need to see where you actually are as a writer."
Level 2: AI for process, not product. Students can use AI to brainstorm, outline, ask questions, or check understanding. They cannot use it to draft or revise their actual writing. I sometimes describe this as "AI as a tutor, not a ghostwriter." Students can ask it to explain a concept, push back on their thesis, or suggest research directions. The words on the page have to be theirs.
Level 3: AI for everything, with documentation. On some assignments, I genuinely don't mind if AI does heavy lifting, as long as students can show me their process and explain their choices. They submit a brief process note alongside the final product: what they prompted, what the AI gave them, what they kept and what they changed and why. This teaches critical evaluation, which is a real skill.
Three levels. Clear labels. Every assignment. It's not a perfect system, but it gives students actual guidance instead of vague warnings, and it gives me something concrete to point to when there's a problem.
How to Talk to Students About This Without Making It Feel Like Surveillance
The way you introduce this framework matters more than you might expect. If you lead with "I'll be monitoring your AI use," students immediately get defensive and start thinking about how to hide things. That's the wrong culture to build.
What has worked better for me is framing it around what they're trying to get out of the class. I'll ask, "If you let AI write your college essays, what happens in the interview when they ask you to expand on something you supposedly wrote?" That question lands differently than "cheating has consequences." It connects responsible AI use to their actual interests.
I also tell them I'm not trying to trick them or catch them. The levels are there because different assignments have different goals. Using AI on a Level 1 assignment isn't just an integrity violation. It's also just bad strategy, because they're cheating themselves out of the practice they'll need later.
For teachers who want a more structured way to enforce these levels, tools like Authentiya are worth looking at. Authentiya lets you set specific AI permissions per assignment and tracks student writing in a way that shows process, not just a final product. It doesn't feel punitive to students because it's built around transparency rather than gotcha moments. I find it especially useful for Level 2 assignments, where the boundary between "AI for process" and "AI for product" can get blurry without some structure behind it.
The goal here isn't to police students. It's to be clear enough that they don't have to guess, and honest enough that they understand why the lines exist. Most students, when they actually understand the reason behind a rule, will follow it. They're not the enemy in this. They're just trying to figure out what's expected, the same as we are.
We're all navigating this in real time. There's no version of this where you get it exactly right from the start. But having a framework, communicating it clearly, and being willing to adjust it when something doesn't work: that's about as good as any of us can do right now.