School · For parents

Three AI habits to change at home, and what to do instead.

Copying answers, trusting unchecked facts and oversharing with a chatbot can hurt your child's learning or privacy. A safer swap for each.

30 Oct 2026 · 6 min read · sources read up to 30 Oct 2026

It's 10 pm and the homework is done

Here's an example we made up. No real family is involved. It's 10 pm. Your child says the homework is finished, a little faster than usual, and you can't tell how. You're not sure whether to feel pleased or worried. You've heard that AI is cheating. You've also heard that it's the future. Neither line helps you work out what to say right now.

We wrote this for that moment, and we aren't here to tell you to ban the chatbot. There are three habits worth changing. One can cost your child some learning, one can hand them a wrong fact, and one can give away something private. Each has a swap that takes about as long as the habit.

We haven't run any of this ourselves. It comes from one 2025 study ^[1], from the chatbot makers' own rules and help pages, and from UNESCO's guidance page ^[11], all of which we read on 30 October 2026. The sources list at the end has every page.

1. When the chatbot does the thinking

Let your child try the problem first. If they get stuck, they can ask the chatbot for a hint on the first step, or ask it to quiz them. This is our own advice, and the study below doesn't say it. We can't promise a hint from an everyday chatbot works the way the researchers' tutor did, but the order matters, and the child's own attempt comes first.

Here is why we say that. In one 2025 study, nearly 1,000 maths students in grades 9 to 11 at one school in Turkey practised with GPT-4, a chatbot. The students who used the ordinary chatbot did much better on the practice problems, with their grades up 48%. Then the researchers took it away and set a later test. That group scored 17% lower than students who had never had the chatbot ^[1].

A second group used a tutor that gave hints instead of full answers. Their practice grades rose 127%, and they largely avoided the drop on the later test. The researchers built that tutor with the teachers' own solutions, so it isn't a setting you can switch on in a public chatbot ^[1].

Why did it happen? The authors' own reading is that students leaned on the chatbot instead of working things out, and they say that mattered more than the chatbot's mistakes ^[1].

Two of the makers' rules touch this too. OpenAI's usage policies list academic dishonesty, which means passing off work you didn't do as yours ^[3]. Anthropic's usage policy forbids handing in AI work without permission or credit ^[2]. On graded work, your child's school decides what is allowed, and if a teacher asks, your child should say how AI helped.

One caution. This was one study, in maths, at one school, with GPT-4, and the later test was taken without the chatbot. It says nothing about board exams or about today's chatbots ^[1].

2. When the chatbot sounds very sure

In that same study, the chatbot's own maths answers were right only about half the time. That was 51% on average across 57 problems, each asked 10 times. It was GPT-4 on those problems only, and we aren't saying it describes any chatbot today ^[1].

The makers say something similar in their own words. OpenAI's terms say the chatbot is not a sole source of truth and ask you to check what it says before you use it ^[4]. Anthropic's terms say it can contain serious mistakes, so check elsewhere ^[5]. Google says its chatbot, Gemini, can state wrong things as if they were true ^[6].

The swap is old-fashioned and simple. If the chatbot says something that matters for a test, look it up in the textbook. If the textbook and the chatbot disagree, ask the teacher which is right.

3. What your child types in

The swap first. Keep your child's name, photos, school and phone number out of any chat. If a tool has a Temporary chat (a chat kept out of your history), use it. If it has a setting that lets your child's chats be used to teach the AI (on ChatGPT it's called "Improve the model for everyone"), switch that setting off. These are our own advice, on top of what the pages say.

We'll keep the reason calm, because none of the pages we read say chats are published. What they say is that chats may be kept, may be read by staff, or may be used to teach the AI. Each tool does it a little differently ^{[6, 8]}.

• Gemini. Google says people may read some chats, so it advises you not to enter confidential information. On a personal Google account, it also keeps chats for 18 months unless you change the setting ^[6].
• ChatGPT, OpenAI's chatbot. On the Free, Plus and Pro plans, chats can be used to teach the AI unless you switch that setting off. A Temporary chat isn't used that way ^[8].
• Claude. Claude is Anthropic's chatbot and it is for adults only (18 and over), so it isn't a tool for your child ^{[5, 10]}.

Before anyone signs up

Every tool has its own age limit, and they differ. ChatGPT is for ages 13 and up, with a parent's permission under 18, and its terms also refer to your country's minimum age ^[4]. Gemini is for ages 13 and up as well, and on a school account the school has to switch it on. Accounts made through an employer need 18 or older ^[7]. Claude, as above, is for adults only, and it switches off accounts that look like a child's until age is verified ^[10].

UNESCO's guidance asks governments to set an age limit and to protect users' data. Its page gives no number, so we won't quote one ^[11].

Our advice: set the account up yourself, on a type that the age rule allows, and sit beside your child the first few times. Rules change, so read the current terms before you sign anyone up.

Three questions for tonight

You don't have to watch the screen. Three questions do most of the checking, and a child can answer them in a minute.

1. What did you try first?
2. Where did you check that?
3. What did you share about yourself?

If the answers come easily, the chatbot was a helper. If they don't, that's useful to know, and nobody needs to be in trouble. Another check we like is to ask your child to explain the topic back to you with the chatbot closed.

How we checked

We read the study on PubMed Central, a free archive of research papers, because the journal's own page wouldn't load for us. We read the chatbot makers' own pages too, all on 30 October 2026. For UNESCO we read the guidance page itself, not the longer document behind it. We found nothing we could read on how many students copy from AI or how many parents worry, so we've left those numbers out. We also say nothing about CBSE's rules or any other board's on AI help, because your child's school decides. Pages change, and privacy settings change most, so check them again if you're reading this a while after 30 October 2026.