Homework: help vs cheating

Where exactly is the line?

Banning AI from homework outright is both unenforceable and arguably bad preparation for a world that expects AI fluency. Most professional work in 2026 already involves AI tools to some degree, and the gap between students who can use them well and those who can't will only widen.

The useful line is this: AI may help your child think. It may not think for them.

The difference is not always obvious from the homework that lands on the kitchen table. A child can have AI explain a concept they didn't understand, then write the answer themselves — that's learning. A child can paste their assignment prompt into ChatGPT and copy the answer — that's not learning, and it doesn't matter how grammatically correct the output is.

The problem with relying on the homework artefact itself is that you cannot reliably tell which path produced it just by reading it. Both can look the same on paper. What's different is what's inside the child's head.

The "explain it out loud" test

You cannot reliably tell AI-written work by reading it, and neither can software (more on that below). But there is a two-minute test that works at any kitchen table: ask your child to walk you through their work.

Useful prompts:

• "Why did you start the essay this way?"
• "What does this word mean? Would you normally use it?"
• "Explain this step to me like I've never done this kind of maths."
• "If you had to argue the opposite, what would you say?"
• "What did you find interesting in what you wrote?"

A child who did the work — even with AI's help — can answer imperfectly but genuinely. They remember the choices they made, the parts they struggled with, the bits they're proud of. A child whose AI did the work cannot explain choices they never made. They will hesitate, contradict the page, or give a vague answer that doesn't match what's written.

Teachers increasingly use exactly this method, because asking about the work beats scanning it. If your child knows the out-loud test is coming, the incentive to outsource their thinking quietly disappears. That is the real point: not catching them, but making real learning the path of least resistance.

Why you should not trust AI detectors

Schools and parents are sold software that claims to detect AI writing. The pitch is appealing — paste in a paragraph, get a percentage, know whether to trust it. The reality is messier.

Independent studies have found that the leading detector missed more than a quarter of AI-generated text, while other detectors falsely accuse human writers — with non-native English speakers flagged most often, a finding with obvious unfairness built in. Meanwhile, students use "humanizer" tools specifically built to beat detectors. The result is an arms race in which the innocent get flagged and the well-equipped get away.

This cuts both ways for you as a parent.

A clean detector score does not mean your child's work is their own. The detection software has clear limitations. Your child running their AI-generated essay through a humanizer tool before submitting will produce a clean score and is well within the technical capabilities of any 14-year-old who can search YouTube.

And if your child is accused by a school on the strength of a detector score alone, you are entitled to push back. Ask for the evidence beyond the score. Ask the school to do what you can do — have the child explain the work out loud. Drafts, version history (Google Docs keeps one automatically), and the out-loud test are all stronger evidence than any detector. The unfairness of false-positive accusations falls most often on students whose first language is not English; if that describes your child, the burden of pushback is greater but the case is stronger.

House rules that actually work

The rules that hold up aren't about which apps your child can or can't use — those will change before you finish typing the policy. The rules that hold up govern the outcome: what learning happened, and what your child can demonstrate.

A practical three-rule policy that works for a 9-year-old or a 17-year-old:

1. AI is allowed for explanation. It is not allowed for production. "Ask the AI to help me understand why this maths step works" is fine. "Ask the AI to write my essay" is not. This rule is enforced by the out-loud test, not by surveillance.
2. You may need to show your work, including your AI conversations. Save the chat. Keep your drafts. If a teacher asks how you arrived at an answer, you have the trail. This rule reduces the temptation to copy-paste because the receipts can be requested.
3. You will explain your finished work to a parent on request. Not every assignment. Not as punishment. As normal practice, so the out-loud check stops feeling like an accusation when it happens. Parents who already chat about schoolwork over dinner have this rule informally and don't have to invent it under stress.

Notice what these rules do not require: any technical skill, any monitoring software, any guesswork about which app was used. They regulate the only thing that matters — whether learning happened — and they work whether your child uses ChatGPT, a calculator, or a tutor.

How to use a homework helper well

If you've chosen to use a parent-facing AI homework helper yourself (like Toolie's), the same principle applies. The point is not to extract an answer your child copies. The point is to have the concept explained in plain language so you can teach your child through it. The four-section Toolie output — what the question is asking, how to walk through it, where children typically get stuck, and a practice example to try together — is designed for exactly that flow.

A useful pattern for primary-school homework: your child attempts the question first, gets stuck, you read the Toolie explanation, then you walk through it together using the practice example. The AI has explained it to you. You explain it to your child. The thinking work happens in your child's head, the same as it would with a textbook or a tutor — just faster, and at 8pm on a Tuesday when no tutor is available.

Talking to schools and teachers

Schools are in different places on AI. Some have detailed policies. Some have a vague policy. Many have nothing yet. If your child's school has no policy, you have an opportunity (not a problem): you set the house rules, and they become the rules that matter.

If your school does have a policy, read it once with your child. Two things to look for: (1) does it distinguish between using AI to learn and using AI to produce? Good policies make this distinction; weaker ones don't. (2) what's the consequence for a first offence? You want to know before a crisis, not during one.

If your child's school is using AI detectors and treating high-percentage scores as proof of cheating, you should know now what you would say in the meeting where that comes up. The detector evidence is unreliable. Drafts, the version history, and the out-loud test are stronger.

Where to verify the research

The 54% and 1-in-10 statistics on teen AI homework use come from the Pew Research Center's "How Teens Use and View AI" (February 2026). The reasoning crisis warnings from teachers are documented in Fortune's February 2026 reporting. The independent evaluation showing AI detectors' false-positive bias against non-native English speakers has been published in multiple academic studies — search "Liang Stanford AI detector bias" for the most-cited source.