A parent watching their child use an AI assistant for homework usually asks the wrong first question: “is this cheating?” That question has no fixed answer, because the same action — asking a chatbot to explain a concept, check an answer, or draft a paragraph — can be entirely legitimate on one assignment and a direct substitution for the graded skill on another. The right first question is narrower: what is this specific assignment trying to measure, and does this specific use of AI replace that measurement or support it?
This article gives you a decision tree to answer that question per task, plus the one rule that overrides every household judgment call: the school’s or teacher’s stated policy controls, even when it is stricter or looser than what you would have decided on your own.
School rules control, not household preference
Before any decision tree, one fact has to be settled: does the school, the specific teacher, or the specific assignment have a stated AI-use policy? If yes, that policy is the actual rule your child is bound by, regardless of what your family considers reasonable. A household that is comfortable with AI-assisted brainstorming still has a child bound by a teacher’s “no AI on this essay” instruction, and a household that is AI-skeptical still has a child permitted to use a school-approved AI tutor if the teacher has assigned one.
Practically, that means the first item on every homework AI decision is a lookup, not a judgment call: check the syllabus, the assignment instructions, or ask the teacher directly if the policy is unclear. Assume no policy stated does not mean no policy exists — it more often means “ask,” not “default to yes.”
Do not help a child conceal AI use on an assignment that is being individually assessed and where disclosure is expected or required. Helping a child hide AI assistance from a teacher is teaching academic dishonesty, not helping with homework — even when your intent is to help them succeed.
The four-question decision tree
For any specific homework task, ask these four questions in order.
1. What is this assignment actually measuring? Not the subject — the skill. A math worksheet on long division is measuring whether the child can execute the algorithm, not whether they can produce a correct final number. An essay assigned after a unit on argument structure is measuring whether the child can structure an argument, not whether a document exists with the right conclusion. Name the skill in one sentence before doing anything else.
2. Does the intended use replace that skill or support it? “Explain long division to me differently than my teacher did” supports the skill — the child still has to execute it afterward. “Solve these ten problems for me” replaces it entirely. “Help me find three counter-arguments to my thesis” supports argument-structure practice; “write my thesis paragraph” replaces it.
3. Does the assignment require disclosure? Some teachers require students to state whether and how AI was used, similar to a citation. If disclosure is required and skipped, the issue is not the AI use itself but the concealment — which is a different, more serious problem. See the family agreement’s disclosure norm in the family AI agreement for how to set this expectation at home before it comes up at school.
4. Can the child independently demonstrate the skill without the tool? This is the test that matters most and is checked least. If a child used AI to help draft an essay, can they explain, unaided, why each paragraph is ordered the way it is? If AI helped with a math concept, can they solve a similar problem from scratch, on paper, without help? If the answer is no, the AI use has substituted for learning regardless of how the first three questions were answered — the grade may say the assignment was completed, but the skill it was meant to certify was not actually acquired.
Applying the tree: three worked examples
Example 1 — vocabulary practice. Skill measured: recall and usage of assigned words. Use: asking AI to generate five example sentences using each word, which the child then studies. Replace or support? Support — the child still does the recall work when tested. Disclosure required? Check the assignment; likely not applicable to practice tools. Independent demonstration? Test this directly — can the child use the word correctly in a new sentence without the AI-generated examples in front of them?
Example 2 — a history essay with a stated thesis. Skill measured: constructing an argument from evidence. Use: asking AI to write the full essay from an outline the child provided. Replace or support? Replace — the actual construction work, which is what’s graded, happened in the AI’s process, not the child’s. This is the clearest cheating case in this article, independent of any disclosure question. See make AI disagree with you for how to use AI to pressure-test an argument the child actually wrote, instead of writing it for them.
Example 3 — a coding assignment with a debugging component. Skill measured: locating and fixing a specific class of bug. Use: pasting the broken code and asking AI to identify what’s wrong, then reading the explanation before fixing it themselves. Replace or support? Genuinely ambiguous — depends on whether “identify what’s wrong” or “fix it” is the graded skill, and whether the explanation is understood or just copied. This is exactly the case where question 4 — can they independently debug a similar problem afterward — is the deciding test, not question 2 alone.
Verification habits that transfer either way
Whether or not AI use is permitted, a child who uses it for any part of homework needs the same verification habit taught in a child’s first AI conversation: check outputs against a second source before treating them as correct, because chatbots produce confident, fluent, sometimes wrong answers by default. Automated AI-detection tools are not a reliable enforcement mechanism on the school’s side either — OpenAI discontinued its own AI-text classifier in 2023 over an unacceptably low accuracy rate, and has stated plainly that no tool, including its own, reliably distinguishes AI-generated text from human writing (Ars Technica, “OpenAI confirms that AI writing detectors don’t work,” Sept. 2023). Detector error is also uneven: work published in Patterns has shown higher false-positive rates for non-native English writing (Liang et al., Patterns, 2023). Claims about neurodivergent writers are less settled, so do not treat detectors as a fair substitute for the disclosure-and-demonstration approach in this decision tree.
If your child’s school does use an AI-detection tool as part of its policy, that is a fact worth knowing, but it does not change the decision tree above — teach the actual skill and require the independent demonstration regardless of whether detection software exists, because a false negative from a detector is not the same as genuine learning having occurred.
The escalation rule
If you are genuinely unsure whether a specific use crosses the line for a specific assignment, the fallback is not a household guess — it’s asking the teacher directly, with a specific example: “for this assignment, is it acceptable if my child used AI to check their grammar after writing the draft themselves?” Teachers would almost always rather answer a concrete question in advance than discover an ambiguous case after grading.
Try it today
Pick one homework task due this week. Run it through the four questions above before your child starts, out loud if they’re old enough to participate in the reasoning rather than just receive the verdict. Then check, after it’s done, whether question 4 — independent demonstration — actually holds. If it doesn’t, that is useful information about where the learning gap is, whether or not any rule was technically broken.
The homework AI-use decision tree turns this into a printable flowchart with space to log the skill measured, the intended use, and the independent-demonstration check for each assignment, so the same four questions get applied consistently across a whole school term.
