You've decided your course needs a written AI policy — maybe because students keep asking what's allowed, maybe because your department is asking every instructor to have one, maybe because you're tired of deciding case by case. That's the right instinct. A course AI policy is table stakes now, the same way an academic-integrity statement or a late-work policy is. The hard part isn't deciding you need one; it's that "no AI allowed" is neither enforceable in any way you could actually check, nor usually what instructors mean once they think it through — you probably don't intend to ban a grammar checker or a quick definition lookup. Here's a practical way to write a policy specific enough to use, short enough that students will actually read it, and honest about where detection fits into enforcing it.
A policy that says nothing more than "use AI responsibly" or "don't over-rely on AI tools" isn't really a policy — it's a mood. It leaves people worse off in three different directions, not just the students you were trying to catch.
Students without clear boundaries have to guess, and they guess differently. Some read "responsibly" generously and use a tool for tasks you'd consider off-limits; others read it narrowly and avoid AI entirely, including for uses you'd have been fine with — a grammar pass, a brainstorm, a definition lookup. That second group is quietly over-restricting itself, and you'll never know it happened, because the whole point of an over-cautious student is that the AI use they skipped never shows up anywhere.
And if a question does come up later — a flag from a detector, a suspicion, a conversation you have to have — a vague policy gives you nothing to point to. "Use AI responsibly" doesn't tell either of you whether what actually happened crossed a line, so the conversation starts from scratch every time instead of from ground you both already agreed to.
A policy that holds up isn't one blanket rule for the whole semester — it's a short, specific answer to three questions, set per assignment type rather than once for the whole course: what's permitted, what requires disclosure, and what's prohibited. A take-home reflection, a timed in-class essay, and a semester-long research paper test different skills, so the same AI boundary rarely makes sense for all three.
You don't need a committee or a semester to write this. Work through it assignment by assignment, in this order:
This is a starting point to adapt, not a template to paste verbatim — change the specifics to match your assignment and, more importantly, your institution's actual rules before you use it:
"For this assignment, you may use AI tools to brainstorm ideas and to check grammar in your final draft; you may not use AI to generate any sentences or paragraphs you submit as your own writing. If you used an AI tool at any stage, add a two-sentence note at the end of your submission naming the tool and what you used it for. Undisclosed AI-generated writing will be reviewed as an academic-integrity matter following the process described in the course syllabus, including a conversation with you before any decision is made."
Change the permitted/prohibited line to match what the specific assignment is testing, and change the last sentence to match your institution's actual integrity process rather than inventing one — more on that below.
A detector is one input into a human process. It is never the policy itself, and it is never the decision. "We use an AI detector" is not a policy on its own — it doesn't tell a student what's allowed, and it doesn't tell you what to do with a result once you have one. The policy is the set of permitted/disclosure/prohibited rules above; a detector is a tool you might use to check whether that policy was followed.
AI writing leaves measurable statistical patterns — uniform sentence rhythm, predictable word choice, formulaic structure — and ScribeLens is built to detect and surface exactly those patterns, sentence by sentence, with reason keywords naming what triggered a flag. That's real, usable evidence if a question comes up. What it isn't is proof: a result is a probability for you to weigh, not a verdict that settles a case by itself.
This is where sentence-level evidence actually earns its place. "Your score was 74%" is an accusation with nothing to discuss. "These four sentences carry this specific pattern — walk me through how you wrote them" is a conversation a student can engage with, and a position you can defend if the case goes further. The full workflow for using a result fairly once you have one — reading the evidence before the score, weighing drafts and context, treating a flag as the start of a conversation rather than the end of one — is covered in our companion piece on using a detector fairly with students.
Your course policy sits underneath your institution's policy, not beside it. If your department, school, or university already has an AI or academic-integrity policy, it sets the floor — you can be more specific at the course level, but you shouldn't contradict it. Before you finalize anything, check your faculty handbook or academic-integrity office for an existing policy, and check whether your institution has a required disclosure format or approval process for AI use in coursework.
If nothing exists yet at the institutional level, say so plainly rather than inventing authority you don't have — a course policy that references "the process described in the student handbook" when no such process actually exists won't hold up if it's ever tested. What's permitted at one institution may be restricted at another, so checking what other instructors in your department have already adopted is often the fastest way to a policy that's actually consistent with what your program expects.
If you take one thing from this:
You can, but be precise about what that actually covers if you do — a genuine ban still needs to say whether spell-check, grammar tools, and basic lookups count, because students will assume some baseline tools are fine unless you say otherwise. A full ban with no permitted-use list tends to produce the same guessing problem a vague policy does, just in the opposite direction.
Write your course-level policy anyway, but don't claim institutional authority you don't have — describe your own review process explicitly rather than referencing a handbook process that doesn't exist. Check with your department or academic-integrity office; if a policy is in development, ask whether your course-level version needs to align with an expected direction.
Ask for two things: which tool was used, and which step it touched — brainstorming, outlining, editing, drafting. "I used AI" on its own isn't useful to you as an instructor; "I used ChatGPT to outline my three main points" is. Keep the required note short — two or three sentences is enough to be useful without becoming a separate assignment.
Treat a detector result as the start of a conversation, not the end of one. Ask the student to walk you through the specific flagged sentences, and weigh whatever drafts or version history they can provide alongside the result. A denial isn't proof either way, and neither is a score by itself — the review process you defined in your policy is what actually resolves it.
A stricter rule mostly changes what students hide if it isn't paired with a clear disclosure path and a review process students trust. A policy that's specific about what's allowed, and honest about how disclosure protects a student, tends to get more honest compliance than a strict rule with no safe way to admit AI use.
Both, not one or the other. State the full policy once in your syllabus so it's the authoritative version, then restate the assignment-specific permitted/disclosure/prohibited lines directly on each assignment brief — a student checking the night before a deadline is reading the assignment sheet, not scrolling back to week one of the syllabus.
If a submission raises a question under your policy, ScribeLens shows the sentence-level evidence and reason keywords behind a result — something specific to discuss, not just a score. Free, no signup required for a quick check.