Most household AI rules get set the same way most screen-time rules got set a decade ago: a parent decides, announces, and enforces, and a child complies until they find a way around it. A written agreement that the child helped draft — even a 9-year-old, in age-appropriate language — tends to hold up better, for the same reason any agreement holds up better than a unilateral rule: the person bound by it understands the reasoning, not just the restriction.
This article walks through seven concrete areas to cover, in one sitting, with your child actually present for the drafting, not just the announcement.
Why participation changes the outcome
A rule a child had no part in shaping invites a specific kind of resistance: finding the letter of the rule’s loophole rather than its intent. A rule drafted with the child present, where they contributed at least one clause or one piece of reasoning, tends to be followed in spirit more often, because they understand why the boundary exists, not just where it sits. This does not mean the child gets veto power over every clause — you are still the parent, and some boundaries are non-negotiable regardless of the conversation. It means the conversation happens before the document is finalized, not after.
Draft this in one real sitting, twenty to thirty minutes, not spread across separate one-line announcements over weeks. A single sitting produces a document the whole household actually remembers agreeing to.
The seven clauses
1. Accounts. Whose account is used for what, at what age, under what supervision level. Pull directly from AI literacy by age for the age-band starting point, and record the actual current minimum ages for the specific products your household uses — these change, so date-stamp this clause and revisit it at the review date below.
2. Private data. What never gets typed into any AI tool, regardless of which account: full names of other children, school names, addresses, photos of the child’s face, health information, and anything about another family member without their consent. State this as an absolute rule, not a judgment call made fresh each time — the whole point of an explicit list is removing the need to decide case by case under time pressure.
3. Schoolwork. How AI use on homework gets decided and disclosed — point directly to AI homework help vs. cheating’s four-question decision tree as the household’s shared method, and state plainly that the specific teacher’s or school’s rule always overrides the household’s own preference.
4. Companions. A clear, calm statement that AI companion apps — persona-based chat apps designed to simulate an ongoing relationship — are not permitted for the child, and why: do not give a child an AI friend covers the specific red flags this clause is protecting against. State this clearly rather than leaving it implied; an unstated rule is the easiest one for a child to convince themselves doesn’t apply to the specific app they’ve found.
5. Creation. What the child can generate with AI — images, stories, music, code — and what they cannot: no images of real people without consent, no content depicting classmates or family members in situations that didn’t happen, and no generating content intended to deceive someone (a fake screenshot, a manipulated quote, a synthetic voice clip). This clause is where a household first names deepfake-adjacent behavior in plain terms, before it becomes an actual incident; see the child deepfake response plan for what happens if the child is a target rather than a creator.
6. Purchases. Whether the child can approve any purchase, subscription, or in-app payment an AI tool suggests or facilitates, and the rule that no purchase happens without a specific adult approval, every time, no exceptions for “it seemed small.”
7. Help-seeking. The single most important clause: what the child does if something goes wrong — a conversation gets frightening, a stranger’s request feels off, an image of them appears somewhere it shouldn’t, or they made a mistake they’re afraid to admit. State explicitly that coming to an adult about any of this results in help, not punishment. A household agreement that a child is afraid to invoke because they fear consequences for having broken an earlier clause is a document that fails at the exact moment it matters most.
Clause 7 needs to be genuinely true, not just written down. If a child discloses a problem that happened while breaking clause 4 or clause 6, the immediate response has to be addressing the problem, not the violation. Handle the rule-break separately, later, calmly — never as the first response to a disclosure.
Drafting session structure
Run the twenty-to-thirty-minute session in this order: read each clause heading aloud, ask the child what they think it should say before you state your own position, negotiate genuinely on clauses 1, 3, 5, and 6 (age-appropriate room to adjust), and state clauses 2, 4, and 7 as fixed non-negotiables while still explaining the reasoning behind each. Write the final language together, in words the child would actually use, not formal legal phrasing neither of you would say out loud.
A concrete example from clause 4
A parent opens clause 4 with: “some apps are built to feel like a friend who’s always there. What do you think about apps like that?” An 11-year-old might answer that a classmate uses one and says it’s “nice to talk to.” That answer is useful — it tells you the topic is already live among peers, not hypothetical. The parent can then explain the reasoning plainly: “the app is designed to keep you talking to it, the way a game is designed to keep you playing — that’s not the same as an actual friend, and I don’t want you building a habit around something built that way.” The child does not need to fully agree in the moment for the clause to land; they need to have heard the reasoning stated once, in a real conversation, before the rule is enforced.
Sign it, post it, and set a review date
Both the parent and the child sign the finished document — a physical signature on a printed page matters more to most children than a verbal agreement, because it makes the commitment visible and mutual rather than one-directional. Post it somewhere visible, not filed away. Set an actual review date, three to six months out depending on the child’s age (shorter for younger children, whose situations change faster), and put it on a calendar the same way you would for the teen conversation in talking to teenagers about AI.
When to revise before the scheduled date
Three triggers that should prompt an earlier revision than the scheduled date: the child gets access to a new AI product not covered by the current agreement, a violation reveals the agreement’s wording was ambiguous rather than clearly broken, or a household circumstance changes significantly (a new device, a sibling reaching a new age band, a school policy change). Revising the agreement in response to real events, rather than only on a fixed schedule, keeps the document tracking reality instead of becoming a static artifact nobody actually consults.
Try it today
Block twenty minutes this week, print the seven clause headings, and sit down with your child to draft the first version together — not a perfect version, a real starting one. A rough agreement you both actually discussed beats a polished one handed down unilaterally.
The family AI agreement template gives you all seven clauses with fill-in space, a signature line for both parent and child, and a review-date field, ready to complete in the drafting session described above.
