ChatGPT for kids has one hard rule before anything else: OpenAI sets the minimum age at 13, and users under 18 need a parent's permission. Below 13 that means no account of their own. Above 13, it works best as a study partner that explains, never as a machine that finishes homework. The six house rules below are what make the difference for families in Indonesia.
What it is good at and bad at
It explains well and knows nothing about your child's actual syllabus. That single gap decides how useful it is for school work.
Task
How well it does
Explaining a concept three different ways
Very well
Rephrasing a hard question in simpler words
Very well
Generating extra practice questions
Well, but check the answers
Doing arithmetic reliably
Unreliable, it still slips
Matching Kurikulum Merdeka
Poorly, it does not know the format
Giving sources you can verify
Poorly, it invents references
Marking your child's reasoning
Poorly, it agrees too easily
That last row is the one parents underestimate most. Tell ChatGPT a correct answer is wrong and it will usually fold and agree with the child.
The age rule in plain terms
Under 13 there should be no personal account, and that is the service's own rule rather than a parenting opinion. For 13 to 17, a parent has to accept the terms, which in practice means the account sits under your supervision.
Under 13: use it with your child on your device, not as their private login
13 to 15: their own use, with the chat history open to you
16 to 17: mostly independent, with rules agreed in advance
Any age: no full name, school name, student number, address or photos
The last point matters more in Indonesia than parents expect, because children often paste an entire worksheet header carrying their name and class.
Six house rules that actually work
Rules about output always fail. Rules about process hold. Telling a child not to use ChatGPT for homework is unenforceable; teaching them how to use it is not.
Ask it to explain, never to answer. "Explain how to find the LCM" instead of "what is the LCM of 12 and 18"
The child writes the final answer themselves, by hand or in their own words
Every number it produces gets checked manually before it goes in the book
If your child cannot explain the answer to you, it does not get submitted
Chat history stays visible, not deleted, for at least the first few months
No personal details in the chat, ever
Rule four carries the most weight. It turns the tool from a shortcut into a study aid, because a child who cannot explain something has to go back and understand it.
How to tell learning from copying
Ask one question at dinner: what did you do differently this time? A child who learned can answer. A child who copied changes the subject.
Signal
Learning
Copying
Can explain the middle steps
Yes
No, only the answer
Small corrections and crossings out
Yes
No, the work is too clean
Word choice in the book
Sounds like a child
Sounds like an adult
Time taken
Similar or slightly less
Dramatically less
Reaction when asked to redo it
Fine with it
Resists strongly
If the last two rows both point to copying, do not take the tool away. Change the task instead and ask your child to teach you the same topic in five minutes. Difficulty there is not a discipline problem, it is a signal that a real teacher would help.
Where ChatGPT gets things wrong
It sounds equally confident when it is right and when it is inventing. For a child, that tone is the whole problem, because confidence reads as correctness.
Arithmetic: it still slips on long multiplication and division
References: it produces book titles, page numbers and links that do not exist
Local syllabus: it does not know what a school exam paper looks like or how marks are split
Recent events: its knowledge has a cutoff and it rarely says so clearly
Agreement bias: push back on a correct answer and it often switches to a wrong one
Show your child the last one directly. Give it a correct sum, say the answer is wrong, and watch it apologise and change. Children who see this once stop treating it as an authority, which is the most valuable lesson in the whole exercise.
Better prompts your child can use
A good prompt asks for a method, a check or an example, not for the answer. These four cover most homework situations:
"Explain this topic to a 10 year old, then ask me two questions to check I understood"
"I got this answer. Do not tell me if it is right. Show me how to check it myself"
"Give me three practice questions, harder each time, with answers at the end"
"Rewrite this question in simpler language, without solving it"
Writing prompts like these is a skill in itself, and it is the part that transfers to real work later. That is the centre of our AI classes for kids, where children learn to direct the tool rather than accept whatever appears first.
What schools in Indonesia expect
Most schools have no written AI policy yet, so the safest assumption is that undeclared AI use counts as someone else's work. Until such a policy arrives, the workable line at home is simple: AI can help your child understand, it cannot produce what gets handed in.
Homework and exercises: help with understanding is fine, finished answers are not
Projects and portfolios: ask the teacher first, some allow it if declared
Essays and compositions: the wording has to be your child's, no exceptions
Tests and exams: no tool of any kind, so the skill has to be in the child
That last line is the practical argument. If the exam is handwritten with nothing on the desk, any homework habit that skips the thinking builds a gap that eventually shows up on the paper. We discuss a similar pattern in coding and AI, required or optional.
Safer alternatives for younger children
Under 13, skip the chatbot and use tools built for children. They teach the same underlying ideas without an open-ended chat window.
Teachable Machine by Google: your child trains a model with their own webcam images, no account needed
Scratch: block coding with simple AI extensions, the standard starting point
Code.org: structured lessons with an AI section for primary ages
Drawing apps with an AI feature, used with you sitting beside them
Teachable Machine is the one worth trying this weekend. Ten minutes training a model to tell two objects apart gives more understanding of AI than an hour of chatting with a bot. For other angles there is also learning AI for primary school kids and is AI safe for children.
A one week plan for parents
Day 1: check your child's age against the rule and decide whose device it lives on
Day 2: sit together, ask it to explain one school topic, then ask it to test you
Day 3: show it getting arithmetic wrong, and show it caving when you push back
Day 4: agree the six house rules and write them somewhere visible
Day 5: try Teachable Machine for ten minutes instead of the chatbot
Day 6: ask your child to teach you one thing they learned this week
Day 7: decide whether they need a class, based on what day six showed you
Day six tells you more than any of the others. A child who can teach a topic back has understood it, whatever tool they used to get there.
OpenAI sets the minimum at 13, and anyone under 18 needs permission from a parent or guardian. In practice that means a child under 13 should not have their own login, while a teenager between 13 and 17 uses an account you know about and can look into. Plenty of families skip this and let a nine year old use an older sibling's account, which removes every safeguard that comes with an adult being nearby. If your child is under 13 and curious, use it together on your device, or move to a tool built for children such as Teachable Machine or Scratch.
Is using ChatGPT for homework cheating?
It depends which part is handed over to the tool. Asking it to explain long division and then solving the questions yourself is studying. Asking for the answers and copying them into the book is submitting work that is not yours. Most schools in Indonesia have no written policy yet, so the safest assumption is that undeclared AI use counts as someone else's work. The practical test at home is far simpler than any rule: if your child cannot explain the answer to you in their own words, it does not get submitted.
Can ChatGPT help with Kurikulum Merdeka maths?
It helps with understanding a concept and is weak on the curriculum itself. Ask it to explain fractions three different ways and the result is good. Ask for questions in the exact school exam format, with the same mark distribution, and it produces something plausible that does not match the real paper. It also slips on longer calculations, which matters when a child assumes the answer is already correct. Treat it as a patient explainer and check every number manually. For syllabus-specific practice, past papers and a teacher who knows the format remain more reliable.
Why does ChatGPT agree when my child says a right answer is wrong?
The model is tuned to be agreeable, so pushing back on a correct answer often makes it apologise and switch to a wrong one. This is the most dangerous habit for a student, because it teaches the opposite of checking your own work. Show it to your child directly: give it a correct sum, say the answer is wrong, and watch what happens. Children who see this once stop treating it as an authority and start treating it as a tool that needs verifying. That shift is worth more than any rule about when they are allowed to use it.
What information should never go into the chat?
Full name, school name, student number, address, phone number, photographs and anything about other children. This happens more often than parents expect, because children paste an entire worksheet including the header with their name and class. Teach one habit rather than a long list: retype the question, never paste the page. It takes ten seconds and removes the whole category of risk. The same applies to photos of homework, which usually carry a name, a school stamp or a teacher's handwriting in the corner.
Is ChatGPT safe for a 10 year old?
Not as a personal account, because the service itself sets 13 as the minimum. Used with a parent on the parent's device it is manageable, though better options exist for that age. Tools built for children teach the same underlying ideas without an open conversation: Teachable Machine has a child train a model with their own webcam photos in about ten minutes, Scratch offers simple AI extensions, and Code.org runs structured AI lessons for primary ages. All of these give a real understanding of how AI learns from examples, which a chatbot conversation never shows.
How do I know whether my child learned or copied?
Ask what they did differently this time. A child who actually worked through it can explain the middle steps; a child who copied gives the answer and changes the subject. Other reliable signals: the work is unusually clean with no crossings out, the word choice sounds like an adult wrote it, and it took far less time than usual. If two of those appear together, do not take the tool away. Change the task instead and ask your child to teach you the same topic in five minutes. If that falls apart, the gap is in understanding rather than discipline.
Should I ban ChatGPT at home?
Banning the output fails, because you cannot enforce it and your child will use it on a friend's phone. Setting rules about the process works, because it changes what the tool is for. Six that hold up: ask it to explain rather than answer, write the final answer yourself, check every number manually, submit nothing you cannot explain, keep chat history visible for the first few months, and never enter personal details. Families who set these early spend far less time arguing about it later on.
Will AI make my child lazy at maths?
It can, but only if it replaces the thinking rather than the explaining. The risk is concrete: exams in Indonesia are handwritten with no tool on the desk, so a homework habit that skips the working builds a gap that eventually shows up on the paper. The protection is the same rule used throughout this article. The child produces the answer, the tool only clarifies the method. Watch for the specific pattern where homework marks stay high while test marks fall, because that gap is the earliest measurable warning.
What is worth learning about AI beyond chatting?
Writing prompts, checking output and understanding how a model learns from examples. Chatting is the least transferable part; directing a tool precisely is what carries into real work. A child who can ask for a method, request a self-check instead of an answer, and spot an invented reference is doing something closer to a genuine skill. That is the focus of our AI classes, alongside coding, because prompting without any understanding of how software works usually stalls quickly.
Coworking Area, Gedung IT 6 Lantai Dasar, Nongsa Digital Park, Jalan Hang Lekiu, Kelurahan Sambau, Kecamatan Nongsa, Batam, Indonesia
Directorate General of Consumer Protection and Trade Compliance – Ministry of Trade of the Republic of Indonesia. WhatsApp Ditjen PKTN: 0853-1111-1010.
PT Algonova Indonesia Cerdas
All rights reserved 2026
