A family AI agreement is a one-page set of house rules for AI at home, written with your teen and signed by everyone, parent included. It works when it is short, has a review date, and lets your child admit a mistake without losing the tool. It cannot override school rules or a tool's age limit, and no piece of paper makes a child tell the truth.
Why does an agreement work when a setting does not?
Settings stop things. They do not start conversations, and they only reach what you know about. Common Sense Media's guide to parental controls, last reviewed on July 14, 2026, puts the limit plainly: "Yes, kids can undo parental controls." It adds that the controls "can only keep track of accounts that they know your kid is using." If a setting is the whole plan, the plan ends the day your child finds the gap.
There is also a gap in what each side knows. Pew Research Center's survey of 1,458 US teens aged 13 to 17 and one parent each, published February 24, 2026, found that 64% of teens said they use chatbots, while 51% of parents said their teen does. About three in ten parents were not sure. (The survey ran from September 25 to October 9, 2025, and both figures are what people reported, not what they were observed doing.)
Here is my reading of that. Most families are not fighting about AI. They have never said the rules out loud, so each person is guessing. An agreement is the cheapest way to stop the guessing.
For the settings themselves, ChatGPT's parental controls and how to turn off AI on your child's phone cover what each one does and does not reach. Do those too. The agreement covers what they cannot.
What goes in a family AI agreement?
Five lines is enough. I call this the Five-Line Agreement: five short lines on one page, signed by everyone in the house including the parent, with a date to look at it again. More than a page and nobody reads it. Fewer than five lines and it misses something you will wish you had agreed.
1. Where. Which rooms and which devices. I would choose a shared room over a rule like "no AI at night," because a shared room needs no policing.
2. What for. The jobs AI may help with and the jobs it may not. This is the line that matters for schoolwork. The version I would use at home is: let AI help with the doing, keep the thinking with your child. Brainstorming, explaining a topic you have already tried, checking a draft you wrote: fine. Producing the answer or the essay: not.
3. Show your working. If AI helped, you say so, to me and to your teacher. It does not have to be a confession. It can be one sentence: "I used it to check my outline."
4. Tell me. If something odd, upsetting or inappropriate shows up, or if you broke a line, you tell me. And telling does not cost you the tool. This is the line the rest hangs on, so it gets its own section below.
5. Review date. We look at this again on a set date, and anyone can ask to change a line. Three months is a sensible first gap, because tools and school rules both change often.
Why is "tell me" the line that matters most?
Because children often do not. Common Sense Media's 2026 census of AI use, a US survey of 1,204 children aged 9 to 17 fielded March 18 to 26, 2026, found that 17% of those who use AI chatbots had seen something inappropriate. Of the ones who had, only 33% told a trusted adult, and 53% did not.
That is a survey of what children reported, and it does not say why they stayed quiet. My interpretation is that a child who expects to lose the app will weigh the cost of telling against the cost of hiding it, and sometimes hiding wins. So my suggestion is to make telling cheaper than hiding: put it in writing that coming to you about a slip, or something strange, will not by itself cost the tool.
You could say: "If something weird happens with it, or you break one of these, I would rather hear it from you on the day. Telling me will not get it taken away."
There is a limit, and it is worth saying to your teen too. If what they tell you involves a threat, a sexual image of them or of someone else, or a worry about their own safety, you will act on it, because that is what a parent has to do. You can promise not to punish the telling. You cannot promise to do nothing.
If your child tells you they have been asking a chatbot about hurting themselves, treat it as a conversation about your child, not about the tool. In the US, call or text 988. In the UK, Samaritans is on 116 123 and Childline on 0800 1111. Your pediatrician (GP in the UK) and your child's school safeguarding lead are also good first calls. A chatbot is not a clinician, and none of these services can promise how a situation will unfold.
What does a copy-and-paste agreement look like?
Copy this into a document, change the words to sound like your family, and fill the blanks together. It is deliberately plain.
Our Family AI Agreement
1. Where. We use AI in: ______________________ (rooms) on: ______________________ (devices). Not in: ______________________.
2. What for. AI can help me with: ______________________. AI does not do: ______________________ (for example, write work I hand in as mine).
3. Show your working. When AI helped, I say so, to my parent and to my teacher, in one sentence.
4. Tell me. If something strange, upsetting or inappropriate happens, or if I break a line here, I tell my parent. Telling will not, by itself, mean losing the tool. My parent will act if I or someone else is at risk.
5. Review date. We look at this again on: ____________. Any of us can ask to change a line before then.
My parent also agrees to: ______________________ (for example: ask before looking at my chats; learn how the tools I use work; keep to the same lines myself).
Signed: ______________ (teen) ______________ (parent) Date: ____________
Notice the parent line. A one-sided list of promises reads as rules handed down. A two-sided one reads as an agreement. The Family Online Safety Institute's free agreement for teens takes the same two-sided approach. It includes the teen line "Use AI tools to support my learning, not to do my work for me," and a parent commitment to "Stay informed about the AI tools my teen uses." Its other lines cover wider online safety, so it is a useful companion to this page rather than a replacement.
How do I get a teenager to actually sign it?
Do not hand it over finished. Ask them to write line 2 themselves, because the line about what AI is for is the one they know most about. Then offer to change anything that is unfair.
You could say: "I do not want to spy on you or ban anything. I want us to agree what is fine, so neither of us is guessing. Can you tell me what you actually use it for?"
If they push back on a line, ask what problem it creates for them. A real answer ("I need it at night for my coding project") is a reason to change the line. "Because you do not trust me" is a reason to start with the review date and try it for a month.
For the longer conversation that comes first, how to talk to your teen about AI homework without a row has more scripts, and what teens actually use AI for will help you ask better questions.
How does the agreement change with age?
Under 8. Skip the paper. Children this young should use AI with you beside them, on your device, for something you are doing together. Age limits differ by tool. OpenAI's teen safety post of September 16, 2025 says "ChatGPT is intended for people 13 and up." See what age kids can use AI for the rule on each tool, and follow it.
8 to 12. Three lines, spoken or drawn: where, tell me, and what for. You are still in the room. Use the review date to move up a step together. AI for 9 to 12 year olds goes deeper.
13 to 15. The full five lines, negotiated. This is the age when the "tell me" line is hardest to keep and most worth keeping, and when a teen leaning on AI for feelings is a real possibility.
16 to 18. Shift from rules to a shared review. Ask your teen to bring their own draft of lines 2 and 3, including how they plan to follow exam, coursework and application rules.
What I See in Schools
The pattern I see most often in secondary schools is that the rule is about the task, not the tool. A teacher says AI is fine for brainstorming on this assignment and not allowed on that one, and the rule changes by subject and by teacher. Students lose track, and parents rarely hear what the rules are.
Common Sense Media's census found that 73% of the children it surveyed said a school or teacher had communicated what they should and should not use AI for in schoolwork. That is the children's report, and it says nothing about whether parents were told. So my practical suggestion is to put the school's rule in line 2 of your agreement, in your teen's words, and to ask the school for the written version.
An agreement is not a contract. It is a reason to have the conversation again in three months.
What should I ask the school before I sign anything at home?
Ask whether there is a written AI policy parents can read, what counts as acceptable help in each subject, and what happens if a student is accused of misusing AI. Questions to ask your child's school about AI has the full list. If your child's school has not spoken to parents about AI yet, this is the work I do with schools, and it is worth sending to your head of year or principal: my work on school AI policy.
What can a family AI agreement not do?
It cannot make a child honest. It cannot reach what happens on a friend's phone, a school device or a tool you have never heard of. It does not replace platform age limits, which are often self-declared and weakly checked, and it does not replace advice from a professional if you are worried about your child's wellbeing. Where the school's rule and your agreement disagree, the school's rule comes first for schoolwork.
What it can do is make the next conversation easier to start. If you would like help keeping up as the tools change, I write a newsletter that parents are very welcome to join.
Dan Fitzpatrick is a Forbes contributor, keynote speaker and bestselling author who has trained more than 150,000 educators in 30-plus countries. He is a former teacher and assistant headteacher, and a school trustee. About Dan.
