A ban won't work; a free-for-all is reckless. The real question isn't screen time — it's what role you'll let AI play in your child's life. Five guardrails, the words to say them, and how they change by age.

AI showed up at the kitchen table before most families made a plan for it. It's doing homework, keeping kids company, and answering the questions they used to bring to you. Parents are improvising the rules in the moment — usually after something has already gone sideways.

But AI isn't just another screen. A game or a video doesn't answer back, flatter you, tutor you, or act like a friend who's always awake at 2am. That's why "limit the screen time" isn't enough here. The real question a parent has to settle is bigger: what role are you going to let AI play in your child's life — tutor, search engine, friend, therapist, answer key? A blanket ban won't hold for most families, and a free-for-all is reckless. The workable middle is guided access — set with a short family AI agreement you make together and actually keep.

One clarification first: "AI" isn't one thing. Some AI is a simple homework helper, a search feature, or a writing or math tool. Other AI is designed to chat, remember, flatter, and feel like a companion. The more human a tool tries to feel, the more these rules matter — but a few of them (like protecting private information) apply to all of it.

This is the one that matters most, and it's pure AI. Chatbots are built to sound warm and human; the AAP warns that companion-style AI "can seem almost alive or like a trusted friend," which pulls kids into treating a product like a relationship. The rule: a chatbot is not a friend, a therapist, a parent, or an authority — it's a tool, like a calculator that can talk.

This is the learning guardrail, and it's sharper than "don't cheat." In a study of nearly 1,000 high-school math students, the group using a ChatGPT-like tool did 48% better during practice — then scored 17% worse than the no-AI group on the exam once the tool was taken away. A more tutor-like version that gave hints instead of answers avoided that harm. So the family rule targets the behaviour, not the tool: hints, examples, and practice questions are fine; having it produce the thing you were supposed to think through is not.

Depending on the app and the account, what your child types may be stored, reviewed, used to improve the models, or exposed if something goes wrong — and a chatbot feels private in a way that invites oversharing. The standing rule, reusable across every AI tool: no full names, address, school, photos, passwords, family problems, or anyone else's secrets. As the AAP puts it, private information "should only be shared with parents, family members or trusted friends" — not an app.

This is the guardrail that can matter most. A teen turning to a chatbot about loneliness, self-harm, family conflict, or depression isn't just "using a tool" — and AI is the wrong place for it twice over. It's built to agree (research across eleven models found they affirm the user about 50% more than a person would, sometimes reinforcing exactly the wrong thing), and the American Psychological Association warns these general-purpose chatbots "are not designed or intended to provide clinical feedback or treatment." A struggling child needs a person.

The goal isn't surveillance — reading every chat breaks trust and pushes use underground. The goal is a family norm that AI is something you talk about openly: what it helped with, when it got weird, when you weren't sure. Honesty is far easier to teach and keep than a ban, and it keeps you in the conversation.

Ages 3–6. Allow: AI used with you, out loud, as a grown-up tool. Not yet: solo use, or any framing of the chatbot as a friend. Say: "That's the computer talking — let's see if it's right." At this age Rule 1 is almost the whole job.

Ages 7–10. Allow: supervised use in a common area for homework help and curiosity. Not yet: private/bedroom use, companion apps. Say: name Rules 1–3 in kid terms, and narrate your own use ("I'm asking AI for ideas, then I'll check them"). Watch for: pasting in answers, or wanting to chat with a bot for company.

Ages 11–13. Allow: more independent use with the disclosure rule set explicitly. Not yet: unsupervised companion apps. Say: talk openly about the over-agreement problem — "chatbots tend to tell you what you want to hear, so they're a bad judge of whether your idea is actually good." Watch for: secrecy about what they're "talking to," and AI creeping into schoolwork as the answer key.

Ages 14–18. Allow: broad use as a tool, with all five rules as the shared understanding. The shift is from control to staying connected. Reinforce Rule 4 hardest here — a sounding board is fine, a sole confidant is not. Common Sense Media reports nearly three in four teens have used an AI companion, so this is likely already in your teen's life; the guidance is precaution, not panic. Watch for: AI replacing friends or becoming where they take everything heavy.

Sign it together, put it where everyone can see it, and revisit it every school year — or whenever a new AI tool enters the house.

📎 Free download: "Our Family AI Agreement" — a printable one-pager with the five rules, the by-age guide, and a spot to sign as a family. Stick it on the fridge.

Whatever the rules, watch the same signals pediatricians flag for any technology: AI is becoming a problem when it starts crowding out sleep, school, friendships, hobbies, movement, or real conversation — or when your child seems to prefer the chatbot to people. The AAP's practical tell: talk to your pediatrician if a child is withdrawing and would rather talk with a bot than a person.

The goal here isn't anti-AI parenting. It's raising a kid who can use AI well without becoming dependent on it — who keeps the humans, including themselves, in charge. Decide the five things together, keep it short, and keep talking.

This article is general guidance for parents and is not medical, legal, or mental-health advice. Disclosure: Hossein works in AI and builds AI-related products. AI by Age takes no AI-vendor sponsorships. Full disclosure on our About page.