The honest line with AI and homework isn't whether AI was involved at all — it's whether the student's own understanding grew as a result of using it. That single question resolves almost every case parents and teachers worry about, and it's a more useful test than any blanket rule about which tools are "allowed," because those rules are already unenforceable and everyone secretly knows it.
Why "did AI touch this" is the wrong question
A student who looks up a word, checks a formula, or asks a chatbot to explain a concept they've half-grasped is doing something closer to consulting a dictionary or a textbook than to cheating. Nobody accuses a student of academic dishonesty for using a dictionary. The tool isn't the issue.
What actually matters is what happens to the student's head in the process. Did they finish that interaction knowing something they didn't know before, in a form they could reproduce without the tool in front of them? Or did they finish it holding a piece of text they didn't really generate, about to be submitted as if they had?
That's the real distinction, and it maps directly onto the Answer Trap: the trap isn't using AI, it's stopping at the first answer instead of treating it as the start of the next question.
The Ask Ladder test, applied to homework
Every use of AI on a piece of schoolwork sits somewhere on the Ask Ladder:
| What the student asked | Ask Ladder level | Integrity read |
|---|---|---|
| "What does 'photosynthesis' mean?" | Locating | Fine — this is a dictionary lookup |
| "Why does this algebra step work?" | Clarifying | Fine — this is asking for the rule, not the answer |
| "Where did my reasoning go wrong here?" | Diagnostic | Fine, and genuinely useful — this is debugging your own thinking |
| "Write me a paragraph comparing these two poems" | — | Not on the ladder at all. This isn't a question, it's outsourcing the task |
| "Solve this equation for me" | — | Same problem — no question was asked, an answer was requested to be handed over whole |
Notice the pattern: every genuinely defensible use of AI is a question about something specific. The indefensible use isn't a question at all — it's a request to skip the thinking and receive the output.
Why undefendable work is a practical problem, not just an ethical one
Parents and teachers often frame this as a matter of honesty, and it is. But there's a second, more immediate consequence that has nothing to do with getting caught: work the student can't explain doesn't hold up.
A GCSE coursework paragraph lifted whole from a chatbot collapses the moment a teacher asks "why did you choose that word there?" in class. A maths answer copied without working collapses the moment the exam asks the same method applied to slightly different numbers. This isn't a hypothetical — it's how KS3 and GCSE assessment is actually structured: follow-up questions, varied numbers, "explain your reasoning" marks. Undefendable work fails on contact with almost any assessment that isn't pure multiple choice.
That's a strong practical argument for doing the asking properly even if nobody is checking for integrity at all. It's also why aitutors.me's own tutors — Professor Pi's hint ladder, Professor Quill's PEEZL structure — are built to never hand over a finished answer. It isn't a policy choice bolted on afterwards; it's what makes the tutoring actually work.
The self-test: could you explain it unaided?
Before submitting anything a student has used AI to help with, there's one question worth asking honestly:
Could I explain, unaided, why this answer is right — to someone who then asked me a follow-up question?
If yes, the AI use was research, in the same category as a textbook or a teacher's explanation, and the work is genuinely the student's own even though a tool was involved in producing it.
If no, the work isn't ready yet. That doesn't mean starting over — it means going back and asking another question. What does this bit actually mean? Why does the argument work this way? What would I say if someone asked me to justify this sentence? Ask-Try-Ask — ask a specific question, attempt the work, ask a sharper follow-up based on what happened — turns an act of copying into an act of learning, using the exact same AI tool, with the exact same amount of time spent.
What this means in practice
- Looking things up is fine. Definitions, formulas, "what's the rule for this" — all Locating or Clarifying level, all normal.
- Debugging your own attempt is fine, and better than fine. "Where did I go wrong" is Diagnostic-level questioning and is exactly the kind of question a good tutor would want a student asking.
- Asking for a finished piece of work isn't a grey area. It skips every rung of the ladder at once, and it's also the version that fails the self-test almost automatically, because there was no thinking to be able to defend.
- The self-test beats any list of banned tools. Tool bans age badly and are trivially worked around. "Could I explain this to a follow-up question" doesn't — it's asking the same thing whether the tool is a chatbot, a sibling, or a study guide.
For more on how this connects to the wider theory, see what Learn by Asking actually is and why asking matters more than answers in the AI era. For the parent-facing version of the cheating conversation, see the AI homework cheating trap.
FAQ
Is it cheating to use AI for homework? Not automatically. It depends what you asked it for. Using AI to answer a Locating-level question — what does this word mean, how does this formula work — is normal research. Using it to produce a finished piece of work you then submit as your own is the Answer Trap, and it is a practical problem as much as an ethical one: you can't defend work you can't explain.
What's a simple test my child can use before submitting work? Could I explain, unaided, why this answer is right, in my own words, to someone who asked me a follow-up question? If the honest answer is no, the work isn't ready — ask another question of the material, or of yourself, before it goes in.
Why does it matter if a teacher can't tell the difference? Because the exam can. Coursework that isn't backed by real understanding falls apart the moment a teacher asks a follow-up in class, or an exam asks the same idea from a different angle. The gap shows up later, at a worse moment, whether or not anyone catches it at submission.