Writing an AI Use Policy for Your Syllabus: Templates for Allowed, Limited, and Banned Levels

A blank “AI policy” line in your syllabus is a trap. Leave it vague and students will fill the gap with their own definitions, which rarely match yours. The fix isn’t a longer lecture on integrity. It’s three clear sentences that tell students exactly where the line sits for each assignment.

Below are copy-paste templates for the three policy levels most instructors actually use, plus the wording and enforcement habits that make them hold up.

Key takeaways

• A vague policy is worse than no policy. “Use AI responsibly” tells students nothing and is unenforceable.
• Pick one of three clear levels per assignment: Allowed, Limited, or Banned. Mixing them across a course is fine, as long as each task says which applies.
• Write the policy in plain language, give concrete examples of what crosses the line, and tell students how to cite or disclose AI use.
• Don’t anchor your policy to detector scores. Detectors output probabilities, not proof, and they produce false positives, so design around process and disclosure instead.
• Treat AI like any other tool: define the boundary, explain the why, and make consequences for crossing it predictable.

Why “use AI responsibly” doesn’t work

Here’s the problem with the soft version. “Responsibly” means one thing to the student who used ChatGPT to outline an argument and write it themselves, and something completely different to the student who pasted the prompt and turned in the output. Both think they followed your rule. When you grade them differently, the one who guessed wrong feels ambushed, and they have a point.

A good policy removes the guessing. It states the level, gives an example of what’s fine and what isn’t, and explains what happens next. You’re not writing a legal document. You’re closing a loophole.

The three-level model maps to how most courses already think about tools. A calculator is banned on a mental-math quiz, limited on a homework set where you show your steps, and fully allowed on a take-home stats project. Generative AI fits the same frame.

The three levels (with copy-paste templates)

Level 1: Allowed

Use this when AI is just another resource and the learning goal isn’t threatened by it: brainstorming, study guides, explaining a confusing concept, drafting low-stakes discussion posts.

> AI Policy. Allowed. You may use generative AI tools (ChatGPT, Claude, Gemini, and the like) freely on this assignment, including for brainstorming, drafting, editing, and feedback. You’re still responsible for the accuracy, originality, and quality of everything you submit. If a tool hands you a wrong fact or a fake citation, that’s on your final work, not the tool. Add a one-line note at the end stating which tools you used and how.

The disclosure line matters even here. It normalizes talking about AI openly and gives you a read on how your class is actually working.

Level 2: Limited

This is where most major assignments should land. AI helps with specific stages, but the student’s own thinking has to stay visible and dominant. It’s also the hardest level to write well: “limited” needs a defined boundary or it collapses back into vagueness.

> AI Policy. Limited. You may use generative AI for the following: brainstorming topics, generating counterarguments to test your thesis, and checking grammar or clarity. You may not use AI to write paragraphs you submit as your own, generate your thesis or main analysis, or produce citations. Submit your outline and at least one earlier draft with the final version. End with a short note describing exactly how you used AI and which prompts mattered.

Notice the structure: an explicit “may” list, an explicit “may not” list, and a process requirement. The drafts aren’t busywork. They’re the evidence that the analysis is the student’s, which is what lets you grade fairly without playing detective.

Level 3: Banned

For in-class work, exams, foundational skill-building, or anything where the act of producing the text *is* the learning. Diagnostic essays. Language acquisition. First-year composition fundamentals.

> AI Policy. Banned. Generative AI tools may not be used in any part of this assignment, including brainstorming, drafting, editing, or generating citations. This work must be entirely your own. If you’re unsure whether a tool counts (the AI features built into your writing software, say), ask me before you use it. Submitting AI-generated or AI-edited work here is an academic integrity violation under [course/institution policy].

The “ask me before you use it” line does a lot of quiet work. Grammarly, autocomplete, and translation tools all blur into “AI” now, and a student acting in good faith deserves a way to check rather than a trap.

A mini-scenario: one course, three levels

Picture a hypothetical intro political theory course whose syllabus carries a small table:

• Weekly discussion posts. Allowed. Low stakes, the goal is engagement.
• Two analytical papers. Limited. AI for brainstorming and grammar; thesis and analysis must be the student’s; outline plus a draft submitted.
• Final in-class essay exam. Banned. The point is to show what you can reason through unaided.

A student uses ChatGPT to brainstorm objections for paper one, notes it in her disclosure, and submits her outline. Fine, that’s exactly what Limited allows. On the exam she can’t touch it, and she knows that because the table told her in week one. That’s the whole goal: a policy that answers the question before the student has to ask it.

Write the “why,” not just the rule

Students tend to follow rules they understand better than rules they resent. In my experience, a sentence or two of reasoning per policy cuts down on the “but why” emails.

For a banned assignment: “I’m asking you to write this without AI because the skill we’re building this week is forming an argument under your own steam, and outsourcing that skips the part that makes you better at it.” That lands differently than a flat prohibition. You don’t need a paragraph; you just need to show the rule isn’t arbitrary.

Don’t build your policy on detector scores

This is the part instructors get wrong most often, so it’s worth being blunt. Detectors feel like an objective backstop. They aren’t one.

These tools only ever return a probability, not a verdict, and they’re wrong often enough to matter. A 2023 Stanford study (Liang et al.) found that GPT detectors were biased against non-native English writers, misclassifying a large share of their genuine essays as AI-generated. OpenAI pulled its own AI text classifier in July 2023, citing low accuracy. If the companies building these tools can’t make them reliable, a “95% AI” reading shouldn’t decide a student’s grade.

So write your policy around things you *can* defend:

• Process artifacts. Outlines, drafts, version history, source notes. These show the work happened.
• In-class checkpoints. A short in-person writing sample early in the term gives you a baseline for each student’s voice.
• Conversation. Ask a student to explain a passage. Genuine authors can; that’s hard to fake.

A detector can be one input among several, the way a plagiarism flag prompts a closer look rather than an automatic zero. But it can’t be the policy. If you’re weighing whether to bring a checker into the loop at all, our pricing page lays out what one costs and what it does and doesn’t tell you.

A quick checklist before you publish

• Does every assignment state its AI level explicitly? No assignment should be silent.
• For Limited assignments, is there both a “may” list and a “may not” list?
• Did you tell students how to disclose or cite AI use, and in which format (APA, MLA, Chicago all have one now)?
• Did you explain at least one “why”?
• Are the consequences and review procedure stated up front, so a suspected violation starts with a conversation, not an accusation?

For more on teaching alongside these tools, browse more guides for educators. The better you understand detection’s limits, the more honest and defensible your policy becomes.

Frequently asked questions

Can I just ban AI entirely and run everything through a detector?

You can ban it, and for some assignments that’s the right call. But leaning on a detector to enforce the ban is shaky ground. Detectors return a probability, not proof, and they generate false positives, including a documented bias against non-native English writers (Liang et al., Stanford, 2023). OpenAI even retired its own AI text classifier in July 2023 for low accuracy. If you ban AI, enforce it through process evidence such as drafts, version history, in-class writing, and oral defenses, rather than one score from a tool.

How do students cite or disclose AI use under a Limited policy?

Keep it lightweight so they’ll actually do it. A one or two sentence note at the end works: which tool, what it was used for, and which prompts mattered. For example, “I used an AI assistant to brainstorm counterarguments in section 2 and to check grammar in the final draft.” APA, MLA, and Chicago all publish AI citation formats now, so point students to whichever one your course already uses.

What if a student uses AI to fix their grammar? Is that cheating?

That depends entirely on what your policy says, which is the whole point of writing one. Grammar and spelling support is usually fine under Allowed and Limited policies, and it’s a normal use of word processors anyway. The line most instructors draw is between editing the student’s own ideas and generating the ideas or sentences for them. Spell that out so a grammar check doesn’t get treated like a violation.

Should my AI policy be the same for every assignment?

Not necessarily, and a one-size policy is often the weakest choice. A common pattern: Banned for in-class exams and foundational skill-building, Limited for major papers, and Allowed for low-stakes brainstorming or study aids. Just label each assignment clearly so students never have to guess.

How do I handle a suspected violation fairly?

Start with a conversation, not an accusation. Ask the student to walk you through their process, show drafts, or explain a passage in their own words. Build this in from day one by requiring process artifacts (outlines, version history, source notes) so you have evidence beyond a hunch or a detector reading. A clear, pre-stated procedure protects honest students and makes genuine violations easier to address.

Closing

A syllabus AI policy isn’t about predicting every way a student might misuse a tool. It’s about removing ambiguity, so the honest majority knows what you expect and the rare violation is easy to handle fairly. Pick a level per assignment, write it in plain words, explain the why, and build enforcement on process rather than a probability score.

If you want to see how your own writing reads to these tools before you bring them up in class, try PaperBleach’s checker and watch how much the numbers wobble.