How Do I Write a Course AI Policy and Communicate It to Students?

Three weeks before term, the syllabus template has a line for "AI policy" and it currently reads: see university academic integrity rules. Or a colleague's course came back with a run of submissions nobody thinks were written in the room, and Thursday's program meeting ends with a request that every course state its position in writing. Or the policy is already written and the marking period is showing you which of its sentences you can actually apply.

The bottom line: name the tasks you grade, say what AI may do at each one, put the declaration in a single place, and enforce with evidence you can see. Five specific permissions hold better than a page of prohibitions. A policy that explains itself does assessment design and communication work at the same time, which is why it costs an afternoon and saves a semester of arguments.

Everything below is the course-level document for one module, course or program strand. The assumption is that you have decided what you are assessing; the work here is saying so and making it stick.

The short version

1. Inventory the graded tasks and ask of each one what, specifically, you are assessing. The boundary follows that answer.

2. Write the rule at the level of the step: brainstorming permitted, composing objectives not. Image curation permitted, slide-level message design not.

3. Put the summary on the syllabus, the operative wording on the assignment sheet, and say the rule out loud once at the moment it first applies.

4. Name one standard place and form for disclosure, and keep the cost of declaring lower than the cost of hiding.

5. Enforce with work you can observe: a prompt log, an in-class exercise, an oral account. Detector scores stay outside the enforcement path.

Start from the tasks you grade

The clearest worked method is McCorkle (2025), who was replacing a blanket prohibition that students did not believe applied to them. Read that failure correctly: it was misalignment rather than defiance, and the students did not see themselves as behaving dishonestly. A rule they have already decided is not about them cannot be argued with, only ignored.

McCorkle's method runs in four passes. Inventory every task in the semester project, ask of each "what, specifically, am I assessing?", pair each task with a plausible professional use of generative AI, then decide task by task. Each decision weighs the need to assess a capability against the value of building a competency the student will use at work.

The output is deliberately uneven. Brainstorming is permitted; composing specific and measurable learning objectives is not. Image curation is permitted; slide-level message design is not. Scripts and narration are permitted, with required evaluation of the AI output. Written out, that list looks arbitrary until you read the reason beside each line, which is why the reasons travel with the rules.

Decide how strict to be, knowing what everyone else does

Two data points tell you what a normal policy looks like this year, and both are worth carrying into the meeting where someone proposes a ban.

A content analysis of 116 institutional GenAI policies from 131 US R1 universities and 98 computer-science syllabi from 54 of them found institutions broadly pro-use: 63% encourage GenAI use, 41% offer detailed classroom guidance, 27% discourage it. Half the syllabi (50%) outright prohibit it, 41% permit partial use for specified activities, and 7% communicate encouragement. Ganguly et al. (2026) call this the top-down versus bottom-up gap. 92% of syllabi give explicit guidelines, yet local rules drift from institutional guidance, and only 47 institutions had both levels detectable. Citation is the one requirement both levels agree on, at 83% of syllabi.

Chirikov (2026) tracked over 31,000 syllabi at a large Texas public research university from 2021 to 2025. Regulation rose from near zero before ChatGPT to 55% of courses by Fall 2025, while fully restrictive policies fell about 5 percentage points a year. By Fall 2025 instructors most commonly restricted drafting and revising (79%) and reasoning and Problem Solving (65%). They most commonly permitted editing and proofreading (83%), study support and synthesis (80%) and coding (75%). Ideation was the most contested use, at 46% permit and 54% restrict.

The wording moved with the numbers. Academic-integrity mentions fell from 63% to 49%, references to AI's effect on learning rose from 1% to 29%, and attribution requirements rose from 16% to 43%. The field is shifting from "do not cheat" toward "here is why this boundary protects your learning," and the second version is the one a student can act on.

Two consequences for your document. Being stricter than your institution is normal, and it is a commitment you take on alone, because the enforcement cost lands on you. A citation requirement is close to free, since four fifths of the field already asks for it and no student has to guess what you mean.

Put each rule where the student will meet it

Three documents, three jobs.

The syllabus carries the short summary: the tasks, the permitted use at each, how disclosure works, and the consequence of a crossed boundary. One screen, no philosophy.

The assignment sheet carries the call-out. A single summary is too coarse for a complex project, and this is where McCorkle's wording did its work, in a rationale addressed to students in the second person that named the assessment behind each restriction.

The room carries the first oral statement. Say the rule once, at the moment it first applies, tied to the work in front of them. A rule a student meets alone is a rule they interpret alone.

Disclosure mechanics need their own paragraph, and it should be the most concrete one on the page. Students who co-designed a course policy through guided inquiry prioritized training for students and instructors, standardized disclosure procedures, stronger institutional support and greater involvement in decisions about AI (Hingle and Johri (2026)). Say how, where and in what form a student declares use, and say it once.

Pisan (2026) found a stronger version of the same move: make the prompt log the graded artifact. When a model does the production, the prompt and the interaction record are the thing worth versioning, so the declaration stops being paperwork and becomes evidence the student is building.

Write only rules you can apply in one marking cycle

A rule that needs per-task monitoring, or a judgment about when an assessment "begins," will be applied inconsistently across a class and across two markers. Inconsistency is an equity problem before it is an administrative one. A regime built on detectors adds a surveillance cost that lands unevenly, along with proctoring and data-retention questions you may not be able to answer at course level.

Nash and Burriss (2026) analyzed 27 classroom AI policies written by preservice English language arts teachers. 26 of 27 permitted some AI use, almost always on teacher-specified terms, times and places; one prohibited it entirely. Ideation was the most broadly permitted use (22 of 27) and the most ambiguous. 22 disallowed or left unclear the AI composition of sentences, paragraphs or papers. 16 permitted grammar checking and proofreading while 16 prohibited producing large AI-generated text. And 22 of 27 did not address reading at all.

The recurring failure was operationalization. One policy allowed AI "to get your thinking started" and declared "this is where the line should be drawn" without saying where the line was. That ambiguity leaves students unable to comply and you unable to apply a consistent standard. It is a defect, not a compromise.

Chirikov's finding of substantial disciplinary and task variation argues for frameworks that grant instructors autonomy within their own domain rather than one mandate for everyone, which is also the honest description of what a course policy is: a local judgment inside a wider rule.

There is an equity dimension here. Opaque policies assume shared background knowledge about authorship, attribution and the norms of academic work, so they fall hardest on students who arrive without it. McCorkle's case is that transparency dismantles part of that hidden curriculum and lowers the chance a policy failure becomes a disciplinary matter.

The cheapest version of all this is few and specific: the tasks you assess, the uses permitted at each, the disclosure mechanics, and the consequence of a crossed boundary, each with its reason attached.

Say it out loud, and watch what you model

Disclosure rules do not work as compliance mechanisms on their own, and treating them as if they do is the most common design error. In a mixed-methods study of 409 undergraduates, Qu and Wang (2026) found non-disclosure was strategic adaptation to perceived peer norms and low interpretive Trust in instructors rather than moral negligence. Perceived peer disclosure and comfort with the instructor were the strongest predictors of students' own disclosure; moral disengagement had weaker effects.

That leaves you two levers you actually control: what students believe their peers are doing, and whether they trust you to read a declaration fairly. Both respond to how you talk about the rule in the first week. Neither responds to the severity of the wording.

Kim et al. (2026) found five distinct sites where a course's real policy lives: faculty intention, formal policy, student interpretation, student self-policy and student practice. The gaps between them are systemic. They also catalogued 23 rationalizations in six classes that students used to justify AI use in academic writing, from "no human victim" to "instructor indifference." Those rationalizations were ad hoc and post hoc: students explained behavior that had already happened rather than reasoning beforehand, which is why clearer wording and harsher detection do not close the gap by themselves.

One participant who wanted to obey a strict no-AI rule kept using AI and said the prohibition "is creating conflict for me, because I'm breaking the rules." The ban intensified moral conflict without preventing use. A rule that produces guilt without compliance is eroding the relationship you need for disclosure.

Watch what you model, too. Most assignments in Pisan's course disclosed the prompt used to generate them, and one student read that transparency as permission: "the dependency on ai from the teacher for grading and creating assignments also made it difficult to not use ai for assignments in the same way." Modeling a norm is part of the policy as students experience it, whatever the syllabus says.

A statement you can adapt:

AI use in this course. I want you to leave this course able to do X yourself, so I assess X directly. You may use AI for [permitted uses] on [task list]; you may not use it for [restricted uses], because those are what I am grading. If you use AI anywhere in your work, tell me in the [single named place, such as a disclosure line on each submission] what you used it for and paste the prompts. Declaring it will never lower your grade. Failing to declare a use is the thing that carries a consequence. If you are unsure whether a use is allowed, ask me before you submit, not after.

That last sentence moves the decision point to before the work, which is the only point at which the question is cheap to answer.

Plan for the four conversations before they happen

The student who used AI and did not say so. Handle this as a process question, not a character question. Mohamed and Temimi (2026) model assessment as a problem of imperfect information: the student knows how the work was produced, and the institution sees the artifact plus partial traces. Their response-region model compares three responses, no AI use, disclosed use and hidden use, and asks which one each mechanism makes most attractive. Prohibition sets the formal boundary and leaves hidden use attractive when students see detection as weak. Monitoring raises the expected cost of hidden use without raising disclosure, so students become cautious without becoming transparent. Permission with disclosure makes honest reporting viable only while the cost of honesty stays low.

In the room, that means asking what the work was meant to assess, asking the student to account for their process, and reading a late declaration as information rather than a confession.

The student who cites AI for everything. This is usually a skills gap in a compliance costume. Attribution requirements rose from 16% to 43% of syllabi while academic-integrity mentions fell from 63% to 49%, so the field is asking for citation rather than confession. Specify the format once, on the assignment sheet rather than in a syllabus footnote. State what adequate attribution looks like in your discipline and give one worked example. Where the real problem is that the student cannot yet do the underlying work without the model, the answer is the task design.

Group work. Five students can hold five interpretations of one sentence, and one submission gets graded. For each group deliverable, decide and publish whether AI use is permitted, who declares it, and what happens when one member goes outside the rule. Do that before the project starts, while it is a planning question. If you cannot state the individual contribution a member is accountable for, grade the task as group-only work rather than adjudicating authorship after the fact.

Disclosure you cannot verify. Some declarations will be incomplete and some omissions unfalsifiable. Bassett et al. (2026) argue AI detection should not be used in education at all, on three grounds that bear on wording. Detector output is a probabilistic estimate that cannot be independently verified, because real-world text origin is unknown, so validation runs on circular reasoning. Detector scores do not meet the balance-of-probabilities standard an integrity investigation requires. And the human-versus-AI dichotomy is meaningless for work created with rather than by AI. Their conclusion is that detection "does not safeguard academic integrity; it undermines it."

They flag a drafting problem worth copying into your revision pass. Rules restricting AI use "in assessment" fail to specify when an assessment begins, so AI-assisted research, planning or editing may or may not be a violation depending on who is reading. Name process evidence you can observe instead: a submitted prompt and interaction log, an exercise produced in the room, an oral explanation. Where your institution runs a detector anyway, their warnings about data storage, retention and commercial use of student work apply, and the Privacy objection two preservice teachers raised is the objection your students will raise.

What instructors get wrong about this

A stricter policy is a safer policy. Being stricter than your institution is common and defensible, but the enforcement cost is yours, and a rule you cannot apply consistently across a marking cycle is an equity problem. Cut the rules you cannot run.

A ban is the shortest message available. In the 27 preservice policies, the ambiguity sat mostly in the permissive ones, and 22 of 27 never addressed reading, so students were left to infer. Kim et al.'s participant obeyed a ban by breaking it and feeling bad about it.

Disclosure is a compliance mechanism. It is a signal, and signals depend on trust. Qu and Wang found peer norms and comfort with the instructor predicting disclosure better than moral reasoning did.

A detector score settles a disputed case. It does not reach the standard an integrity investigation has to meet, and it cannot distinguish work made with AI from work made by AI. Build the case on work you observed.

Writing the policy is the communication. Kim et al. found five sites where the real policy lives, and your document is one of them. The version students act on is assembled from what you say in class and what you do with your own materials.

What will come back at you

"Students will just hide it." Some will, and the design question is which students your rule tempts. Deterrence runs through a detector's discrimination between hidden use and legitimate work rather than its raw catch rate. When extra sensitivity produces more new false positives than new true positives, stronger monitoring can make concealment relatively more attractive, because honest students are penalized faster than hidden users are identified. Permission and disclosure are separate levers: permission moves the boundary of acceptable use, disclosure changes visibility. Diagnose where your task pulls the student most tempted to conceal, and design for that student.

"I do not want to ban tools I cannot detect." You do not have to. Prohibition is one of four design responses, alongside monitoring, permitted use with disclosure, and redesign. Redesign changes what the task rewards, and stays cosmetic if the rubric still grades mainly the final product. Pisan (2026) shows the levels made explicit: AI is barred in the introductory programming course, because outsourcing the first loops removes the thing being taught; encouraged on projects in data structures (Copilot permitted) and barred from pen-and-paper examinations; a study and review aid in the upper-division systems course; and required in the AI course itself, where one boundary carried most of the weight. The model may write code, and reflections must be in the student's own voice. Examinations were removed entirely, and assessment moved onto work a model cannot quietly ghost-write.

"My institution's policy already covers this." Less than you think. The gap runs both ways: 92% of syllabi give explicit guidelines, local rules drift from institutional guidance, and only 47 institutions had both levels detectable. Procurement, data and review cycles belong to How Do We Write and Implement an Institutional AI Policy?, but the classroom translation is yours. Nash and Burriss conclude that institutions must equip teachers to resist as well as adopt, with the guidance and professional development that let an instructor decline a specific AI use without being framed as behind the times. Their participants' technodeterminism contradicted their own pedagogical commitments. An institutional policy is not cover for your judgment call.

"I do not have time to rewrite the syllabus this month." Run the inventory once, on the assignment that worries you, and take the output straight into the assignment sheet. The syllabus sentence can wait for the next revision cycle. Students act on the assignment sheet anyway.

What to do this week

1. Inventory the graded tasks and write "what, specifically, am I assessing?" beside each one, before you decide anything about AI.

2. Decide task by task, and check which response each decision makes most attractive to the student most tempted to conceal.

3. Write the disclosure sentence: one place, one format, one worked example of adequate attribution.

4. Name the evidence you will accept in a dispute (prompt log, in-class work, an oral account), and drop any rule you cannot test against it.

5. Say the rule out loud in the next class, at the moment it first applies, with a rationale beside each restriction.

Where to go next

Governance, data policy, procurement and review cycles belong to How Do We Write and Implement an Institutional AI Policy?; mandated institutional and district rules sit with Educational AI Policy and AI Governance. Rebuilding the task so a grade still supports a defensible inference sits with How Do I Redesign Assessment So That a Grade Still Tells Me Something Defensible About What the Student Knows or Can Do?, the enforcement argument is in How Can I Reduce AI Cheating in My Course?, the detector decision in Should We Use AI Detectors?, and the label scheme in A bit of chaos and madness: The AI Assessment Scale and the work of assessment reform.