How to Write a Classroom AI Policy
The essay was never the point. The thinking was.
of those same teachers said their school has a policy on student AI use. Fewer than half.
Fraser Institute survey of 756 teachers, 2026of surveyed Canadian students aged 18 and over use generative AI for schoolwork, and 48% say their critical thinking has deteriorated since they started.
KPMG in Canada survey of 684 students 18+, October 2025of American students say their teachers never explicitly taught them how to use AI for schoolwork.
RAND survey panels, 2025of American teens say students at their school use AI chatbots to cheat at least sometimes.
Pew Research Center, February 2026Use climbed a lot faster than guidance did. But the gap in the numbers is not the real problem.
I teach high school sciences in Southern Ontario. Here is the truth about assessment in high school: I have never cared about the lab report itself. I care whether a student can think critically about an experiment and refuse to conclude more than the data shows. It is the same across the school. Nobody needs a high schooler's essay, we need students who can build an argument and support it. Nobody needs to know the tree is 32 metres tall, we need a kid who can solve problems and get to the answer from the length of a shadow and the angle of the sun. The product was never the goal. It was how we got students to practise the process, and how we checked that they had.
That worked for a long time, because there was no way to make the product without doing the process. Grade the essay and you were grading the thinking. AI cut that link. A chatbot can turn out the lab report in thirty seconds, the student never did any of the thinking, but the mark says it did. A student who hands that in is not cheating me. They are cheating themselves out of the exact skills the course exists to build. However, it is our job as teachers to ensure the integrity of the credits we give out and that we really are preparing students for the next step.
The fix is not a better detector, and it is not pretending AI will go away. And it is certainly not an all-out ban. It is assessment that looks at the process again, and in your classroom that starts with a clear policy: how students can use AI in your class, how they cannot, and why. That is what the rest of this page is: what a good policy contains, the truth about detectors, and a template you can copy today.
Why teachers need their own classroom AI policy
Your board already has an academic honesty policy. Growing Success requires one. But a board policy has to fit every subject in every school. It cannot tell your class whether using AI to check the reasoning in a lab conclusion is help or cheating. That is a call about your assignment and what it is actually assessing, and nobody above you can make it for you.
A written policy is also what makes enforcement fair: you cannot hold a student to a rule nobody said out loud. I include my policy with my syllabus for every course. It is written, I teach it, and I get the students to share it with their parents. That way we are all on the same page, and if an issue ever comes up, the conversation with the student and the parents is much more defensible. Not only that, the students know where they stand, and they do not have to wonder whether the way they use AI to study is allowed.
What a good classroom AI policy contains
One clear principle
This is the one sentence that anchors the policy. In the template below it is: the work you hand in has to come from your own thinking, in your own words. Write yours before you write anything else.
A plain definition
Say what academic dishonesty means in your class, in language your students understand without effort, and name AI inside that definition.
A list of what is allowed
Be specific, and be as generous as your subject allows: brainstorming, outlining, grammar checks, having a hard idea explained the way a tutor would, getting quizzed on the class notes before a test. Naming the allowed uses is what makes the banned ones enforceable.
A list of what is not
Just as specific. Generated text pasted into an assignment, AI summaries or paraphrases of sources, homework and quizzes done by a chatbot.
A fair, written process if there is an infraction
Decide now what happens the first time you are unsure, and write it down: look at the work, talk privately, then decide. In this section you also need to outline the exact consequences of an infraction. Make sure to get approval from your admin.
Honesty about detection
Say plainly what detection software does and does not do in your class, and that a score never decides anything on its own. But also that your professional judgment is the deciding factor.
AI detectors are not the answer
They are a signal. They are not evidence.
Turnitin says more than 16,000 institutions use its services, and in the first year of its AI detector it reported reviewing more than 200 million papers, with about 11% flagged as containing at least 20% AI writing. Its own fine print sets the limits: the under-1% false positive rate it claims applies only to documents with more than 20% AI writing, it has acknowledged more false positives below that mark, it no longer shows a percentage at all in the 1% to 19% range, and it will not score anything under 300 words.
Detectors are not evenly wrong, either. A Stanford study published in the journal Patterns tested seven publicly available AI detectors against 88 essays by US Grade 8 students and 91 TOEFL essays by non-native English speakers. Turnitin was not one of the seven. The Grade 8 essays passed almost perfectly. More than half of the TOEFL essays were flagged as AI-generated, an average false positive rate of 61%. In an Ontario classroom that is a specific kid, in a specific seat, who did the work.
Students already feel this. Half of the American students RAND surveyed said they worry about being falsely accused of using AI to cheat.
Detectors can start the conversation
A detector score can start a conversation. It can never finish one. Treat it as one signal among several, never as sole evidence, and never as the reason you give a student. The vendors say the same: Turnitin describes its AI writing report as a way to start a formative conversation, not a finding of misconduct, and OpenAI said of its own classifier that it "should not be used as a primary decision-making tool."
What actually holds up
- Drafts, notes and version history. Google Docs keeps it, and Word does too when the file lives in OneDrive or SharePoint. Missing history proves nothing, because a file saved to a laptop has none. But a student who did the work and has it can show you.
- A short private conversation. Ask them to walk you through their choices and explain their ideas in their own words. In my experience this is the step that settles it, in both directions.
- A baseline. One piece of writing done in front of you in the first weeks, so you know what this student sounds like.
- Checkpoints built into the assignment. If the entire goal is to assess the process, then do that, assess the student's process. Have a check-in with a rubric where they show you their outline or their sources.
What I do in my own class
None of this starts the day a suspicious report lands on my desk. It starts in week one. I talk with my students about their work all the time, so I know how each of them writes and thinks. And I try to design my assignments to be resistant to AI in the first place: writing done in class, checkpoints along the way.
When something still comes in that does not sound like the student, I go in order. First I put it beside their past work, to test my suspicion against something real. Then I run it through a detector and pinpoint the sections that look most questionable. The score does not decide anything, it tells me where to look. Then I set up the conversation.
I open the same way every time: "I wanted to talk to you about your report on..." Then I ask about the work itself. Can you explain what you meant in this section? Why did you choose this word here? Can you walk me through how you got from your data to this conclusion? A student who did the work can answer without reading it off the page. If they can, I am satisfied, the matter is closed, and we move on. If they cannot, I flag it as an infraction and follow the consequences in my policy. That process, look, then talk, then decide, is written word for word into the template below.
Two things not to do
- Do not paste student work into a public chatbot to "check" whether AI wrote it. That is a student's work going somewhere you do not control, and it does not work: OpenAI withdrew its own AI-text classifier in July 2023 because it caught only 26% of AI-written text while wrongly flagging human writing 9% of the time.
- Do not open the meeting with the number. Open it with the work, and let the student talk first.
Write yours in six steps
-
Read what already exists
Your school's academic honesty policy, any AI guidance your board has published, and your department's course outline template. Your policy sits underneath those, not beside them. If your board has said nothing about AI yet, say so in your policy: until there is board guidance, this is the rule in this class.
-
Write your one-sentence principle first
This might take a few tries and a few rewrites, but it really is the spirit of the policy and what AI use and academic integrity means to you. Here are some guiding questions:
- Do you want your students to use AI at home, in class, both, or not at all?
- Does the written output matter to you? Or do you want students to do the thinking and let AI do the writing?
- What about visuals? Can students use AI-generated images?
- What is it that you value in your class?
-
Make the two lists specific to your subject
Generic lists get lawyered. Subject-specific ones get followed.
Science
- Explain what a controlled variable is
- Quiz you on your experiment design before a lab
- Write your conclusion
- Interpret your data for you
English
- Brainstorm essay topics
- Build an outline
- Summarize or paraphrase your sources
- Write the paragraphs you hand in
Math
- Explain a method on a practice question
- Generate extra problems to drill
- Do your homework
- Produce the solution you hand in
French and other second languages
- Practise conversation with you
- Correct your grammar and explain the rule
- Translate your assignment
- Write your paragraph in French
History and social studies
- Quiz you on events
- Explain both sides of a debate
- Write your position
- Be cited as a source (it invents them)
Visual arts
- Help you find reference ideas
- Explain a technique like two-point perspective
- Generate the image you hand in
- Produce something you trace
Music
- Build a backing track to practise over
- Quiz you on theory
- Compose the piece you submit
- Write your analysis of a work
Health and phys ed
- Plan a training schedule
- Explain how a muscle group works
- Fill in your fitness log
- Write your health assignment
Business
- Play a tough customer for pitch practice
- Explain a term like cash flow
- Write your business plan
- Produce the market research you hand in
Computer science
- Explain an error message
- Help you trace a bug
- Write the program you submit
- Do your coding exercises for you
-
Write the process before you need it
Three steps is enough: look at the work, meet the student privately, then decide. The template below adds a fourth, an appeal, so a student who disagrees knows what happens next. Writing it while you are calm is exactly what keeps it fair on the day you are not. Say in the policy that detection software is one input into step one and never the decision.
-
Teach it in the first week, then once more
Fifteen minutes, in the first week. I run a short AI lesson at the start of every semester, and I am strict in the weeks after it, because the first assignment sets the tone. I tell my class straight: the lab report is not the goal of this course. The thinking behind it is, and if AI does that thinking for you, you are cheating yourself, not me. I show them some sample prompts they can use. They really like the crazy analogies and mnemonics AI can come up with, which honestly help. For most students it is the first time a teacher has ever talked to them about this.
-
Model it
If you use AI in your own work, say so and say how. Plenty of us do for the paperwork side of the job. The rule I ask of students is the one I hold myself to: the thinking stays mine, and I check everything that comes back. If the paperwork is what brought you here, our report card comment guides for elementary and high school teachers cover it. Or the Gradefully gradebook has report comment generation built in, using the data already in your gradebook.
Copy it, then make it yours
A complete classroom AI policy in student-facing language.
Academic Integrity Policy
How we work honestly in this course, including how you may and may not use AI.
Our commitment: honesty
In this course, honesty matters. Your work has to be your own, so that your marks actually reflect what you did and what you learned. That means taking responsibility for your work, giving credit to anyone who helped you, and citing your sources properly, in APA style.
What academic dishonesty is
Academic dishonesty means handing in someone else's work, or work you did not really do, as if it were yours. It includes:
- plagiarism (copying and pasting without citing),
- cheating on a test or an exam,
- working with someone else on something that was assigned as individual work,
- making up data or results, or changing your data so that it fits the answer you expected, or
- using artificial intelligence (AI) to do the work for you.
Using artificial intelligence (AI)
What is allowed: AI as a tool
You can use AI to help you:
- find ideas or build an outline,
- check your grammar and improve your writing style,
- understand a difficult topic, the way you would with a tutor, or
- format your citations (always check them, because AI gets citations wrong).
What is not allowed: academic dishonesty
You cannot use AI to:
- write text that you copy and paste into an assignment,
- summarize or paraphrase your sources for you,
- create, complete or "clean up" data or results for a lab, or
- do your homework, your quizzes or your tests.
The basic principle: the work you hand in has to come from your own thinking, in your own words.
Telling me how you used AI
If you use AI at all on an assignment, say so. Add a short note at the end of your work that names the tool and the model you used and explains what you used it for. One or two sentences is enough. Using AI in an allowed way is never a problem; hiding it is.
Example: "I used ChatGPT (GPT-5) to help me build an outline for this lab report and to check my grammar. All of the writing and the data are my own."
Example: "I used Google Gemini (2.5 Flash) to explain diffusion to me before I started. I did not use it to write any part of this assignment."
When in doubt, ask me
If you are not sure whether something is allowed, ask me before you hand the work in. You will never get in trouble for asking. It is much easier to sort out a question ahead of time than to sort out a problem afterwards.
Assignment instructions come first
Some assignments will have their own rules about AI, written in the instructions. Those rules replace this policy for that assignment. For example, an assignment may say that no AI may be used at all, for any part of the task. If the instructions on an assignment and this policy do not agree, follow the assignment instructions.
If I have concerns about your work
If I have a question about whether the work you handed in is your own, I will follow a fair and private process.
1. I look first
Before I talk to you, I look carefully at your work, at the sources you may have used, and at anything else I have, including reports from detection software. Detection software is only one signal, and it is never enough on its own. It can be wrong, so it never decides anything by itself. This step is about understanding the situation before we talk about it.
2. We talk, privately
Then I set up a time to meet with you. The goal is a conversation, not an accusation. I will ask you to walk me through how you did the work, to show me your notes or drafts, and to explain your ideas in your own words. This is the part that matters most, and it is the part that is fairest to you.
3. A fair decision
After we talk, I make a decision based on everything I know.
- If there is no sign of dishonesty, the matter is closed.
- If dishonesty is confirmed, the consequence depends on how serious it was, on how many times it has happened, and on your own circumstances. It may mean redoing the work, so that you actually learn what the assignment was for. In a serious case, or if it keeps happening, it can mean a mark of zero and a report to administration.
4. If you disagree
If you or your parents or guardians disagree with my decision, you can appeal it. Another teacher or an administrator will then review your work and my decision independently. After that review, the decision is final.
- Free to copy and adapt
- No sign-up
- Student-facing language
- From a real Ontario course
Adapted from the AI policy in a real Ontario Grade 9 science (SNC1W) syllabus, the course taught by Gradefully's founder. Free to copy, translate and adapt. No credit needed.
Line it up with your school and board first
This is a teacher's template, not legal advice and not board policy. Your school's academic honesty policy, your board's AI guidance and your own professional obligations all come first, and where anything here conflicts with them, theirs wins. It is worth showing your department head too. A department that says the same thing is far easier on students than six teachers saying six different things.
Four things to change before you use it
- Change the citation style. It says APA because that is what that course uses.
- Change the consequences. Mark penalties, reporting and escalation are usually set above your classroom, so match what your school actually does. Growing Success asks boards to build consequences as a continuum weighed against four things: the student's grade level, their maturity, how many times it has happened, and their individual circumstances. A zero is not a ministry default.
- Change the examples to your subject. Use the lists in step 3 above as a starting point.
- Change the voice to yours. Students can tell instantly when a policy was pasted in and never read aloud.
Classroom AI policy FAQ
Do I need a classroom AI policy if my board already has one?
Yes, because they do different jobs. The board policy, which Growing Success requires, sets the boundaries for every school and subject. A classroom policy tells students what counts as their own work in your assignments, and that is a call only the person who built the assignment can make. Yours sits under the board's and never contradicts it.
Should I just ban AI completely?
You can, and in some courses that is the right call. But be honest about where your rules reach: students will use AI at home if the work lets them, so a no-AI course means the writing that counts happens in class, in front of you. Make the ban specific, say what students should do instead, and enforce it: a blanket ban everyone knows is unenforceable teaches students that your rules are decoration. Most teachers get further by naming a few allowed uses, because that is what makes the banned ones clear.
Can I use an AI detector score as proof that a student cheated?
No. Treat it as one signal among several and never as sole evidence. Turnitin no longer shows a number at all below its 20% threshold, and a Stanford study published in the journal Patterns found that seven publicly available detectors misclassified more than half of a set of TOEFL essays by non-native English writers as AI-generated. A score is a reason to look more closely and to have a conversation. Drafts, notes, version history and a student explaining their thinking are what actually hold up.
What should I do the first time I suspect a student used AI?
Slow down and follow your process. Look at the work and the possible sources first, then meet the student privately and ask them to walk you through how they did it. Decide after that conversation, not before it. In my experience the meeting settles most of what a percentage cannot, in both directions.
Can I copy this template into my course outline?
Yes. It is free to copy, translate and adapt, and you do not need to credit anyone or ask permission. Please read it against your school and board policies first, and change anything that does not match how your class actually runs.
Where the claims on this page come from
The documents behind the claims on this page. Page numbers are the printed page.
- Growing Success: Assessment, Evaluation and Reporting in Ontario Schools, First Edition, Covering Grades 1 to 12
- The Ontario Curriculum, De-streamed Course: Science — Grade 9 (SNC1W)
- GPT detectors are biased against non-native English writers
- Fabrication and errors in the bibliographic citations generated by ChatGPT
- AI Use in Schools Is Quickly Increasing but Guidance Lags Behind: Findings from the RAND Survey Panels
- How Teens Use and View AI
- Survey of Canadian Grades 6–12 Teachers on AI Training and School Policies for Its Use
- Generative AI boom among Canadian students raises dilemmas
- Turnitin's AI writing detection capabilities FAQs
- Turnitin marks one year anniversary of its AI writing detector with millions of papers reviewed globally
- Turnitin admits there are some cases of higher false positives in AI writing detection tool
- New AI classifier for indicating AI-written text
- View previous versions of Office files
The Fraser Institute study is a non-probability panel survey, and the Turnitin figures are the vendor's own. Links checked July 2026. If a source here does not say what we say it says, tell us at /contact/ and we will fix it.
Report cards next? Gradefully drafts the comments in your own voice.
The gradebook this template's author built: your marks, your class lists, and the report card comments in one app.