I've read a lot of school AI policies this year. Most of them share a structure: two paragraphs about the transformative potential of technology, a long list of prohibited behaviors written in passive voice, and a signature line that turns the whole thing into a liability shield.
Ask a student what the policy actually says and you get a shrug. Ask a teacher and you get a slightly more apologetic shrug. A policy nobody can recite is a policy nobody follows, and it definitely isn't a policy anything can enforce.
The schools doing this well write their policies the way good teachers write classroom rules. Short, concrete, in the voice of a person.
The index card test
Here's the bar: a student should be able to read your AI policy in thirty seconds and predict, with confidence, whether a specific prompt would be fine or not. If they can't, the policy isn't done.
Compare these two.
The lawyer version: "Students shall not utilize generative artificial intelligence systems in a manner inconsistent with the academic integrity expectations of the institution."
The teacher version: "AI can explain things, quiz you, and give feedback on your drafts. It cannot write your assignments, answer your test questions, or do your homework for you."
Same rule. The second one is enforceable because it's understandable. It names the allowed uses, names the banned uses, and leaves room for a kid to ask about the gray areas instead of guessing.
Steal these guidelines
Good plain-language rules I've seen actual schools use, lightly edited:
- "Use AI like you'd use a tutor, not like you'd use a ghostwriter."
- "If you'd be embarrassed to show the prompt to your teacher, don't send it."
- "AI feedback on your work: always fine. AI doing your work: never fine."
- "Cite it like a source when it shaped your thinking."
- "During assessments, no AI at all unless the assignment says otherwise."
Notice none of these mention specific tools. Tools change weekly. Behavior categories don't. A policy that says "no ChatGPT" was outdated the day it was printed; a policy that says "no answer-writing" still works five tools from now.
Plain language is now executable
For most of the history of school policy, there was a gap between what you could write and what you could enforce. You could write "no essay writing," but the enforcement was a teacher's eyeball and luck.
That gap is closing. The current generation of governance tooling takes guidelines written in ordinary English and applies them to every AI prompt before it leaves a school device. The teacher writes "no full essay writing, tutoring and feedback are fine," and that sentence becomes the actual rule the system enforces. No regex, no domain lists, no IT ticket.
Which means the index card version of your policy isn't just friendlier. It's literally the version a machine can enforce, because intent-level rules are what intent-level detection runs on.
Three mistakes to skip
Don't write for the worst case. Policies built entirely around the cheater read as hostile to the 80% of students who just want to study. Lead with what's allowed.
Don't freeze it. Put a review date on the thing. The tools your students use in March didn't exist in September. A policy with a revision history earns more trust than a stone tablet.
Don't skip the why. One sentence does it: "These rules exist so your diploma means you can actually do these things." Students follow rules they understand the point of, at least at meaningfully higher rates, and you're a teacher, so you already knew that.
Write it short. Write it human. Then back it with enforcement that actually reads intent, and the policy stops being a document and starts being how the classroom works.