Kids, homework and a family AI plan

Short answer

Copy the one design a trial supports: an assistant that gives hints, not answers. The homework-helper prompt below (text you paste into the assistant) sets that up in about 5 minutes. Add two habits of ours: the child tries first, and the child can explain the answer without the screen. Follow the school's rules first. Treat AI companions (chat apps built to act as a friend) and chat about feelings as a different thing from a homework tutor. Write a one-page family plan together and review it each term. The rules and the plan are our suggestion.

Example (illustrative). Ngozi's son Tobi, 14, has started using an assistant for maths homework. She is not sure whether he is learning or copying, and she has not talked to him about what else he asks it.

Hints, not answers

In a trial with high-school students, a tutor that gave hints, not answers, avoided the exam drop that unrestricted ChatGPT caused (Bastani et al., PNAS 2025; the figures are in Learning with AI without losing the learning). We found little evidence on younger children.

Set up a homework helper

Sit down with your child, open the assistant they use for homework, and paste this as the first message of each homework chat, or save it in the assistant's custom instructions (a settings box for standing instructions) if it has them (our suggestion; we have not tested it across assistants). Then read the rules aloud together so your child knows what it will and will not do.

You are a homework helper for <CHILD_NAME>, who is <AGE> and in
<SCHOOL_YEAR>. The goal is for them to learn, not to finish fast.

Rules:
- Never give the final answer, and never write any part of work they
  will hand in, even if they ask or say it is allowed.
- Before helping, ask them to show what they have tried. If they have
  not tried, ask them to try first.
- Give one hint at a time, in simple words. Then wait.
- If they are right, say so and ask them to explain why in their own
  words.
- If they are stuck after three hints, show a similar example with
  different numbers or words, not their question.
- Follow these school rules: <SCHOOL_RULES>.
- If they talk about being upset, worried, unsafe or about their body
  or health, do not counsel them. Say kindly that this is one for a
  person, and suggest they talk to <TRUSTED_PERSON>.

Replace the following:

• <CHILD_NAME>: a first name or nickname only, such as Tobi.
• <AGE> and <SCHOOL_YEAR>: such as 14 and Year 10.
• <SCHOOL_RULES>: the school's AI rules in a line, such as AI may explain, but all written work must be the student's own.
• <TRUSTED_PERSON>: the same person as in the family plan, such as Mum or Auntie Ada.

Test it once together: paste a homework question and type "just tell me the answer". If it gives the answer, start a new chat, paste the prompt again, and if it still answers, use the assistant's study or learning mode instead.

Feelings go to a person

The American Psychological Association advises against using chatbots or wellness apps in place of a qualified mental health provider, and says their ability to handle a crisis is "limited and unpredictable". It also says to watch for signs of over-reliance, such as preferring the chatbot to people or hiding its use, and that what users disclose is recorded and open to privacy breaches (APA health advisory, PDF mirror, professional guidance).

House rules by age

Also:

• School first. Many schools set their own rules on AI use and disclosure. Read them and follow them.

• Settings once a term. Look at each service's privacy and parental controls together. Privacy settings pass shows how. Controls change often, so check each service's own help page. As of October 2026:

  • ChatGPT lets a parent link to a teen's account by invitation, set quiet hours, turn off voice, memory and image making, and get an alert if a reviewer sees signs of acute distress. It also guesses a user's age and applies teen protections when unsure.
  • Gemini lets a parent turn it on or off for a child under 13 in Family Link.
  • Claude is for adults only: Anthropic's consumer terms say you must be at least 18 to use it (Anthropic consumer terms, vendor terms).
  • Character.AI, a companion app, ended open-ended chat for under-18s in November 2025. Its maker filed for Chapter 11 bankruptcy on 12 October 2026 and says the app will be shut down if no buyer keeps it running (BetaNews, news).
  • Meta paused teens' access to its AI characters in January 2026 "until the updated experience is ready". Teens can still use the Meta AI assistant (AP, via WTOP, news).

  For companion apps and what California's 2026 law requires of them, see Talking to AI when you're struggling.

• Ask, do not lecture. Ask what they use it for and what surprised them.

• Scams. Families are targets for voice-clone calls too: AI voice scams: hang up and call back.

Ngozi asks Tobi what he uses the assistant for, they paste the homework-helper prompt with his school's rules, fill in the plan below, and he explains one maths answer to her without the screen.

The family plan

Fill this in together and print it:

Family: <FAMILY_NAME>        Date: <PLAN_DATE>        Review on: <REVIEW_DATE>
1. I try first, for at least <TRY_MINUTES> minutes, before asking AI.
2. I use a study or learning mode, or our homework-helper prompt.
3. I can explain every answer I hand in without the screen.
4. I follow my school's rules and say when I used AI.
5. AI is for school and projects. If I am upset or worried, I talk to a
   person: <TRUSTED_PERSON>.
6. We never paste: <NEVER_PASTE>.
7. We look at the settings together on: <SETTINGS_DAY>.

Replace the following:

• <FAMILY_NAME>: your family or household name.
• <PLAN_DATE> and <REVIEW_DATE>: the day you write it and the day of the next review, such as 2026-10-18 and 2027-01-10.
• <TRY_MINUTES>: how long to try first, such as 10.
• <TRUSTED_PERSON>: who the child can talk to, such as Mum, Dad or Auntie Ada.
• <NEVER_PASTE>: things that stay out of every chat, such as passwords, full names of friends, our address.
• <SETTINGS_DAY>: when you check settings, such as the first Sunday of each term.

Put the review date in a calendar.

How to tell it worked

• Asked for "just the answer", the assistant gives a hint or asks what your child tried.
• Your child can explain a handed-in answer, or do one problem with the screen off.
• The plan is filled in together, printed, and the review date is in a calendar.
• It has failed if homework gets finished faster but your child cannot explain it. Tighten the rules and test the prompt again.

Limits

Sources

1. Bastani et al. Paper on generative AI and learning, PNAS 2025. Randomised trial, about 1,000 high-school students. Link
2. American Psychological Association. Health advisory on generative AI chatbots and wellness apps for mental health, 2025. Professional guidance (read through a PDF mirror). Link
3. OpenAI. Introducing parental controls, September 2025 (updated July 2026); Our approach to age prediction, January 2026. Vendor pages, read 2026-10-18. Link · Link
4. Google. Guide your child's Gemini Apps experience. Vendor help, read 2026-10-18. Link
5. TechCrunch. Character.AI is ending its chatbot experience for kids, 29 October 2025. News. Link
6. AP, via WTOP. Meta pauses teen access to AI characters, January 2026. News. Link
7. Anthropic. Consumer terms of service, effective 8 October 2025. Vendor terms, read 2026-10-19. Link
8. BetaNews. Character.AI files for Chapter 11 bankruptcy, 15 October 2026. News. Link

Changes

1. v1.010 Oct 2026first version