How do you teach students to use AI honestly?

Most students are not trying to cheat. They are trying to finish work under pressure with tools that sit one tab away, and many honestly don't know where the line is. A course rule that says "use AI responsibly" leaves each student to guess, and different teachers in the same school may guess differently. Honest use starts with clear expectations, set assignment by assignment.

Detection is not a reliable alternative. OpenAI's own guidance for educators states that AI detectors, including its own, have not proven able to tell AI writing from human writing reliably, and that they can wrongly flag students who write in a second language or in a formulaic style. Asking ChatGPT "did you write this?" doesn't work either; it has no way of knowing.

Use levels, not a single rule

A helpful tool here is the AI Assessment Scale, developed by researchers Mike Perkins, Jasper Roe and Leon Furze and revised in late 2024. It gives five levels a teacher can assign to a task:

The authors say plainly that the scale is a starting point to adapt, not a rulebook, and that higher levels are not better than lower ones. They also make a practical point: a "No AI" task is only believable if it happens where the teacher can see it. For take-home work, it is more honest to allow some AI and assess what the student does with it.

A worked assignment

Here is an example for a high-school history class. It assumes students are old enough to use the tool under its terms: ChatGPT requires users to be at least 13, with a parent's permission under 18.

Task: Write a 600-word argument on whether the railway changed your region more than the automobile did.

AI level: 2, AI-assisted planning.

Allowed:

• Asking an AI to suggest angles, counter-arguments or questions to research.
• Asking it to explain a term or event you met in your reading.
• Asking it to quiz you on your notes.

Not allowed:

• Having AI write any sentence that appears in your final text.
• Using facts from AI that you have not checked in a source from the reading list or the library.

Disclosure line, at the end of the essay, in one of these forms:

AI use: None.
AI use: I used ChatGPT to brainstorm counter-arguments and to explain "land grant." All wording and sources are my own. Link to the conversation: [shared link].

Process evidence, worth 20% of the grade:

• Your outline, showing at least one idea you rejected and why.
• The shared link to any AI conversation you used.
• Two sources you checked a claim against, with page numbers.

OpenAI's educator guidance suggests this kind of shared conversation link as a way for students to show their work and for teachers to see how they think. It also turns the chat itself into something to teach from: which questions were good, and where the student accepted an answer too quickly.

Teach the skill, not just the rule

Rules work better when students practice the honest version. Spend one lesson early in the term doing an assignment together at the chosen level: brainstorm with an AI on the projector, then show students how to check one of its claims and why some suggestions are weak. The revised scale's authors argue that the real goal at the middle levels is critical judgment: students who can spot errors, question an answer that sounds final, and keep their own voice.

Students should also see that disclosure is ordinary. Teachers can model it: "I used AI to draft the quiz questions and then rewrote four of them." When honesty is normal and costs nothing, there is little reason to hide.

What should an assignment's AI policy say?

A good policy fits on half a page and answers five questions a student would ask:

1. What level is this task? Name it and explain it in one sentence.
2. What may I do with AI? Give concrete examples, written as actions ("ask for counter-arguments"), not general values.
3. What may I not do? Again, specific actions.
4. How do I say what I did? Provide the exact disclosure line to copy, including "None."
5. What happens if I get it wrong? Distinguish an honest mistake, such as forgetting the disclosure line, from deliberately hiding AI use, and say how each is handled.

Add one line on why the level was chosen, for example "This task checks whether you can build an argument from sources yourself." Students accept limits more readily when they see the purpose. Finally, make sure the policy matches your school's academic integrity rules, and revisit it each term, since the tools students use keep changing.