Using AI for homework without cheating: a practical line

Students get contradictory advice about AI. One teacher bans it, another demonstrates it in class, and the school policy — if it exists — is written vaguely enough to cover both. Meanwhile the tools are on every device.

There is a workable line underneath all of it, and it is not about which tool you opened. It is about which part of the work the tool did.

The line

If AI helped you understand something, that is studying. A tutor, a worked example, a friend explaining a concept and a language model explaining the same concept are the same category of help.

If AI produced the artefact you handed in, that is submitting someone else’s work — regardless of how much you edited it afterwards. The assessment was measuring whether you could produce it.

Almost every real school policy resolves to that distinction, even when it is phrased in more words.

Seven uses that are study, not cheating

1. Ask for the concept a different way. When a textbook explanation does not land, ask for it explained at a different level, or with a different analogy, until it clicks. Then close it and write from your own understanding.
2. Generate practice questions. Feed in the chapter you are revising and ask for questions on it. Answer them yourself first. This is retrieval practice, the single best-evidenced study technique there is.
3. Check your reasoning, not your answer. Write your working, then ask where the reasoning breaks. You learn far more from a located error than from a correct answer you did not produce.
4. Stress-test an argument. Ask for the strongest objection to the thesis of your essay. Then answer the objection yourself, in your own essay. This makes the essay better and is entirely your work.
5. Turn messy notes into a structure. Extract text from photographed notes and organise them into topics. Organising your own material is not the thing being assessed.
6. Learn the vocabulary of a subject. Ask what a term means in this specific context, where it differs from the everyday use, and where students commonly confuse it.
7. Produce media, when media is not the assessment. If the assignment assesses your history argument and asks for a video, generating a title card is not cheating. If it assesses your editing, it is.

What to avoid

• Submitting generated prose, even heavily edited, as your own writing.
• Pasting the assignment prompt in and working backwards from the output.
• Using AI on a coursework component that carries a signed authenticity declaration.
• Trusting a citation you have not opened. Models still fabricate references, and a fake source is the fastest way to fail an academic-integrity check.

On detectors

AI-detection tools are unreliable in both directions. They flag non-native English writers disproportionately and they miss lightly edited generated text routinely. Do not treat "the detector will not catch it" as a plan, and if you are wrongly accused, your draft history is the strongest evidence you have.

Working in a studio that keeps your drafts and your canvas locally, in one project, means that history exists by default.

A note on privacy

Your homework is your work. Pasting it into a cloud tool means handing a copy to a company whose retention and training policies you probably have not read. An offline studio computes the answer on your own machine, so the question does not arise.