Your department chair forwarded a one-line email: “Please add your own AI policy to your syllabus this term.” No template attached. No workshop. No teaching center on speed-dial. Just you, a blank syllabus, and a class that starts soon.
So you did the obvious thing. You searched “AI policy for college syllabus,” and every result turned out to be some university writing for its own faculty, backed by a teaching center you don’t have. Helpful for them. Useless at 9pm when you just need a paragraph you can paste and get on with your night.
This is that paragraph. Three of them, actually: ready-to-paste syllabus statements at every level of strictness, one question to pick between them, and the single trap (AI detectors) that lands adjuncts in real trouble. If you’ve never opened ChatGPT yourself, you’re still fine. Writing a ChatGPT policy for your course doesn’t take AI expertise. It takes one honest decision about what each assignment is for.
First, pick your stance (there’s one question)
Skip the big AI debate for a second. You don’t need a philosophy of technology. You need to answer one question, per assignment:
What is the one skill this assignment is meant to build, and would AI doing that task rob the student of the learning?
That’s it. That question sorts almost everything you assign into one of three stances.
If AI never touches the core skill (say the assignment is about analyzing a dataset, and AI only tidies your grammar), you can allow it. If AI helps at the edges but the student still does the real thinking, you land on limited-with-disclosure. And if the assignment is the skill, where the whole point is that they can build the argument themselves, that’s a no-AI task. You’ll enforce it through how you design the work, not through a detector. Why that last part matters so much comes up in a minute.
One caution before you reach for the ban. Duke’s teaching center and others have flagged something that surprises people: defaulting to “No AI” everywhere, out of fear, can actually raise the number of integrity violations without buying you much. Students use the tools anyway, quietly, and now more of them are technically in breach. So choose your stance on purpose. Don’t grab the strictest one just because it feels safest.
The three ready-to-paste statements
Here they are. Pick one, paste it, swap the tool names if you like. Each is written in plain language you could hand a first-year student.
Stance A — No AI (the ban)
Every word you submit must be your own. Do not use ChatGPT or any other AI tool on any assignment or exam in this course.
What it means for grading: any AI use counts as misconduct. There’s no disclosure path, so there’s nothing for a student to declare. Clean and simple.
The catch, and Teaching@UW is blunt about this, is that a ban only holds if it rests on how you design the work, not on catching people after the fact. Build in some in-class writing. Ask a student to walk you through their draft out loud. Because, in their words, no technology can reliably tell you whether AI wrote something. If your whole ban leans on a detector score, it’s built on sand.
Stance B — Limited, with disclosure (the one most centers recommend)
You may use AI only for the specific tasks I name in each assignment (for example, brainstorming or grammar). If you use it, add one or two sentences saying which tool and how. Any use I haven’t named is unauthorized.
What it means for grading: now there are two separate lines a student can cross. One is using AI beyond the tasks you named. The other is using it and not telling you. Keeping those two apart is the whole trick, because it puts the weight on disclosure instead of detection. You’re not playing detective. You’re asking for honesty, and honesty is something you can actually grade.
That’s why it’s the default most teaching centers point to. It sidesteps the detector mess entirely.
Stance C — AI welcome (encourage it)
You’re encouraged to use AI in this course. When AI contributed to what you turn in, cite it like any other source (APA or MLA) and be ready to explain your process.
What it means for grading: the line is passing AI work off as fully your own, or going past whatever bounds you set. Add one sentence of “why,” something like, because you’ll be using these tools in the work this course prepares you for. That little rationale does more than it looks like it does. Hold that thought for the next section.
The four pieces every good policy has
Whichever stance you picked, a policy that holds up has four parts. Stanford, Teaching@UW, and Tufts all converge on the same short list:
- What’s permitted or prohibited. The specifics, not “use AI responsibly.”
- Whether and how to disclose or cite it. Hand them the exact sentence to write.
- The why. One line tying the rule to what the course is teaching. Stanford leans hard on this one, because when a student hits an edge case you never wrote down, the “why” is what lets them reason it out instead of guessing.
- The consequence. What happens if they cross the line.
Miss the third piece and your policy is just a list of rules. Include it and students can carry your intent into situations you never predicted. That’s the difference between a rule and a rule they understand.
The detector trap (read this before you threaten anyone with Turnitin)
Here’s where new instructors get burned. It’s tempting to close your policy with “AI use will be detected and penalized.” Don’t. The detectors don’t work the way the marketing implies, and leaning on them can put you in a genuinely bad spot.
The research isn’t subtle. A 2023 Stanford study (Liang et al., published in Patterns) ran essays through seven popular GPT detectors. Essays by native English speakers came back nearly clean. Essays by non-native English speakers, real TOEFL essays with zero AI involved, got flagged as AI-generated about 61% of the time on average. Not because of anything the students did. Because their writing had lower “perplexity,” which is the exact pattern the detectors read as robotic. Point one of these at your international students and you’ll wrongly accuse them at a rate you would never sign off on.
It gets thornier with neurodivergent students. There are now court cases over false accusations, including Newby v. Adelphi University and an ADA suit against the University of Michigan. I’m not going to hand you a percentage for that group, because there isn’t a trustworthy one. But documented lawsuits are their own kind of warning sign.
“But Turnitin says its false-positive rate is under 1%.” That’s Turnitin’s own number, about its own product. Independent testing has landed all over the map, some of it far higher. And enough schools lost confidence that Vanderbilt, Michigan State, and UT Austin all switched Turnitin’s AI detector off. The International Center for Academic Integrity says it plainly: enforcement by detection is “not a promising approach,” and AI use “should always be disclosed or acknowledged.”
So here’s the takeaway to build your whole policy on. Don’t threaten a detector score. Rest the policy on assignment design and disclosure instead. That’s the version that survives the day a student pushes back, sometimes with a parent or a lawyer behind them.
What a syllabus line can’t do
Let me be straight about the limits, because a paste-able paragraph can oversell itself.
A policy won’t stop a determined student. Someone set on cheating finds a way; the policy just makes the expectation unmistakable for the other 95%. It also can’t replace assignment design. If a take-home essay can be written start to finish by a chatbot, no sentence in your syllabus fixes that. Only changing the task does. And it isn’t permanent. The tools shift every few months, so plan to revisit the wording next term. A policy is a starting agreement, not a force field.
The bottom line
Pick the stance that matches what your assignments are actually for. Write the four pieces. Skip the detector threat. You can have this done before your coffee goes cold, and it’ll be clearer than half the official policies floating around your inbox.
If you want to move from “I set a rule” to “I can actually use these tools well enough to teach with them,” that’s the natural next step. Our Teaching with AI course is built for exactly this, and if you’re still finding your own footing, AI Fundamentals starts from zero. (Teaching K-12 instead? Here’s the K-12 version of this guide, with a ready parent email.)
Sources
- Tufts CELT — AI Syllabus Statements
- Teaching@UW — Developing AI course policies
- Stanford Teaching Commons — Creating your course policy on AI
- Duke Learning Innovation — AI Policies in Syllabi
- ICAI — Statement on Academic Integrity and AI
- Liang et al. 2023 — GPT detectors are biased against non-native English writers (Patterns)
