FR Newsletter

AI for Life

Helping your child with homework using AI, without it becoming cheating

The question is not whether your child will use an AI assistant. It is whether they will use it to avoid thinking or to think more. Schools have started to answer officially, with an age, a rule and a word: fraud. Here is what the rules say, and four uses that make a child work harder rather than less.

What the school rules actually say

France published a national framework for the use of AI in education in June 2025, after a consultation held from January to May 2025 that gathered more than 500 contributions from teachers, head teachers, inspectors, unions, parents and pupils. It is useful to any parent, in any country, because it states the questions plainly.

Its principles: make sure there is real educational value in using AI, protect the data entered into consumer tools, be aware of the environmental impact of generative AI, be transparent about its use, and keep a critical mind about what it produces.

Then two concrete rules. Autonomous use of these tools is allowed only from the fourth year of secondary school (classe de 4e in the French system, around 13 to 14 years old). And the sentence that matters most at the kitchen table: any use of generative AI to produce a piece of schoolwork, without explicit authorisation and without personal work of appropriation, will be considered fraud.

UNESCO reached a similar place two years earlier. Its guidance for generative AI in education and research sets an age limit of 13 for the use of AI tools in the classroom, and calls for teacher training on the subject, among seven steps it asks governments to take.

Two things follow for a parent. The permission is the teacher's to give, not yours. And "my child only used it a little" is not a category: either the use was authorised and there is personal work on top, or it is fraud.

Screenshot coming after our hands-on test
A worked example: the same exercise, with and without the assistant. Screenshots coming with our hands-on tests.

Before the first use: three settings and one conversation

The framework asks for vigilance about data entered into consumer tools, and the CNIL's position on generative AI services is to submit only information you are allowed to share and to avoid entering personal data as far as possible. For a child, that is simple to turn into rules:

  • No identity. No full name, no school, no class, no address, no photograph of a face, no medical or family detail.
  • Your account, not theirs, for a younger child, with history and data reuse turned off in the privacy settings where the service allows it.
  • In the open. Homework with an assistant happens in a shared room, not behind a closed door. This is the single most effective rule.

The conversation matters more than the settings. Say out loud what the school says: the work has to be theirs, and the assistant is allowed to help them understand, not to produce the answer.

Use 1: make it ask the questions, not give the answers

The default behaviour of an assistant is to answer. That is exactly what you do not want. Reverse it:

You are helping a 14-year-old understand a maths exercise. Never give the answer or the full method. Ask one question at a time to find out where they are stuck, then give one hint. If they give a wrong answer, do not correct it directly: ask a question that makes the mistake visible. Stop and tell them to ask their teacher if they are still stuck after five hints.

Write this once, keep it, and paste it at the start. It changes a cheating machine into something closer to a patient tutor who refuses to do the work.

Use 2: generate exercises, not answers

This is the use with the most evidence behind it, and it does not involve the assistant producing any schoolwork at all. Testing yourself on a lesson is one of the most reliable ways to remember it, which is why we gave it a whole guide on revising for an exam with an AI assistant.

Here is my lesson, pasted below. Write 10 questions on it, from easiest to hardest, using only what is in the lesson. Do not give the answers. Number them. After I answer, tell me which ones are wrong and point me to the part of the lesson to reread, without giving the correct answer.

Nothing produced here goes into a notebook, so nothing is fraud. The child does the recall work, which is the part that makes the lesson stick.

Use 3: explain the correction, after the attempt

Order matters. The child attempts the exercise, gets it marked or checks it against the textbook, and only then asks why. An assistant is genuinely good at reformulating the same idea five different ways until one lands, which a tired parent at nine in the evening is not.

I got this exercise wrong. Here is the exercise, my answer, and the correct answer. Explain only what I misunderstood, in simple terms, then give me a similar exercise to try alone. Do not solve the new one.

Use 4: written work, where the line is thinnest

For an essay or a report, the framework's "personal work of appropriation" is the whole test. Two uses stay on the right side of it:

  • Before writing: ask for questions the essay should answer, not for a plan and certainly not for paragraphs.
  • After writing: ask for a critique of their own text, with the instruction not to rewrite it. "Point out three places where my argument is not supported, and ask me a question about each. Do not rewrite anything."

Anything that produces sentences the child then submits is on the wrong side, authorisation or not.

What to check, and what we do not claim

  1. Ask the teacher what is allowed for their class. The rule above is the national frame; the authorisation is local.
  2. Read the output together at least the first few times. Assistants state wrong things with total confidence, and a child has no way to spot it in a subject they are still learning.
  3. Watch the dependency, not just the rules. If the first move on every exercise is to open the assistant, the tool has replaced the effort rather than supporting it.

We have not tested any tool marketed specifically for schoolchildren, so this guide names none and ranks none. The ages and the fraud rule come from the official texts; the four uses are the ones that put the work back on the child.

Tools in this guide

Questions

From what age can a child use a generative AI tool?

UNESCO's guidance sets an age limit of 13 for the use of AI tools in the classroom. In France, the national framework published in June 2025 allows autonomous use of these tools only from the fourth year of secondary school, with younger pupils introduced to the basics without handling generative AI directly.

Is using AI for homework cheating?

The French framework is explicit: any use of generative AI to produce a piece of schoolwork, without explicit authorisation and without personal work of appropriation, will be considered fraud. The authorisation comes from the teacher, not from the parent.

What should never go into the chatbot?

Personal details. The French framework asks for vigilance about data entered into consumer tools, and the CNIL advises submitting only information you are allowed to share. A full name, a school, a class, a photo or a medical note have no reason to be there.

Sources

  1. Ministère de l'Éducation nationale (France), Publication du cadre d'usage de l'intelligence artificielle en éducation
  2. UNESCO, Guidance for generative AI in education and research
  3. UNESCO, Governments must quickly regulate Generative AI in schools (age limit of 13)
  4. CNIL, Les questions-réponses de la CNIL sur l'utilisation d'un système d'IA générative