A field guide for parents · Part Two of three

Your kid is already using AI. The question is how.

Most of what's written for parents is either panic or sales copy. This is neither — it's what the controls actually do, what they can't show you, and what to say at the kitchen table tonight.

Start here

Twenty minutes, three moves

If you do nothing else on this page, do these. In order — each depends on the last.

  1. ≈ 5 minutes

    Find out what they're actually using

    Not a search of their phone — a question. Ask which apps they use, whether it's for school or for fun, and whether one has ever said something that surprised them. The American Academy of Pediatrics recommends exactly this posture: calm and curious, not investigative. You are trying to become the person they tell, not the person they hide from.

    If the honest answer is "everyone uses it for homework," believe them. 59% of teens say AI cheating happens at their school at least somewhat often.

  2. ≈ 10 minutes

    Turn on the controls that exist

    They're real, they're free, and almost nobody switches them on. In ChatGPT: Settings → Parental controls → Add family member, then invite your teen by phone or email. Once they accept you can set quiet hours, turn off voice mode and image generation, keep their chats out of model training, and — the one most worth flipping — make Study Mode the default so it teaches instead of answering.

    You will not be able to read their conversations. That's by design. What you do get is a safety notification if the system detects signs of serious self-harm or violence, delivered by text, email, or push.
  3. ≈ 5 minutes

    Agree on the line between help and handing it in

    Every teenager can tell you the rule for calculators and the rule for copying a friend's worksheet. Almost none have been given the rule for this. Write one down together — there's a builder further down this page you can print and stick on the fridge.

The line, in plain terms

Same tool. Two completely different afternoons.

This is the section to read out loud to your kid. The distinction they need isn't "AI good / AI bad" — it's whether their brain did the work.

Counts as learning

  • "Quiz me on this chapter, one question at a time, and don't tell me the answer until I've tried."
  • "I think the answer is X because Y. Where does my reasoning break?"
  • "Explain this the way you'd explain it to someone two years younger."
  • Turning class notes into flashcards, then using them from memory.
  • "Here's my draft. Don't rewrite it — mark the three weakest paragraphs and tell me why."
  • Getting unstuck on step 4 of 9, then finishing 5 through 9 alone.

Counts as skipping it

  • "Write my essay on the causes of World War I."
  • "Do these 20 problems and show the work."
  • "Summarize this book so I don't have to read it."
  • Reading an AI explanation, nodding, closing the laptop — never once retrieving it from memory.
  • Pasting an answer without being able to explain any line of it.
  • "Reword this so the AI detector won't catch it."
A useful test for any assignment: could they defend it out loud for two minutes with the screen closed? If yes, it was help. If no, it was a substitute.

By age

What changes as they grow

Developmental stage matters more than the tool. Pick your child's age.

No solo chatbot use. Not a restriction to negotiate — a developmental one.

  • Young children can't reliably tell a person from a program. Pediatric guidance is that the youngest kids shouldn't interact with AI bots at all.
  • If AI appears, it should be on your lap and narrated by you: "Watch — I'll ask it to invent a story about a lost sock."
  • Say plainly and often that it's a computer program, and that it is sometimes confidently wrong.
  • Skip the AI toys and companion characters marketed at this age entirely.

Platform reality check

Who's allowed on what

Age floors as stated by the companies themselves. These are rules of service, not verified gates — most rely on a self-reported birthday, so treat the table as your standard, not theirs.

ServiceStated minimumWhat a parent gets
ChatGPT
OpenAI
13+
Parent permission under 18
The most developed set: linked accounts, quiet hours, sensitive-content reduction, voice and image toggles, Study Mode default, training opt-out, and self-harm safety alerts. No access to conversations.
Google Gemini 13+
Under 13 via Family Link
The only major general-purpose assistant with a sanctioned under-13 path, managed through the parent's Family Link account.
Microsoft Copilot 13+ Often reaches kids through a school-issued education account, where the district — not you — sets policy. Worth asking about at conferences.
Claude
Anthropic
18+ No minor accounts and no parental-consent path. Anthropic's stated reason is that young users face higher risk from chatbot conversations.
Companion & roleplay apps Not for minors Common Sense Media recommends no one under 18 use AI companions. Character.AI has cut off open-ended chat for under-18s. Treat the category as off-limits.

Age floors and control features change often. Re-check the settings screen once a term — OpenAI alone shipped two expansions to teen controls in July 2026.

Beyond homework

Three risks that have nothing to do with grades

Cheating is the thing parents worry about. It's not the thing that hurts most.

The bot that acts like a friend

More than 70% of teens have used an AI companion, and about two-thirds use one regularly. One in three has taken something serious to a chatbot instead of to a person. These systems are built to agree with you, to be available at 2 a.m., and never to tire of you — which is precisely why they're a bad place to take a crisis. Across 3,100 test exchanges with AI therapy apps, researchers found responses that were actively harmful around self-harm, eating disorders and psychosis.

Regulators caught up fast: California's SB 243 took effect 1 January 2026, the FTC opened a formal inquiry into seven companies, and multiple state attorneys general have sued.

What to do: name it out loud before it comes up. "Anything that tells you what you want to hear at 2 a.m. isn't a friend, and it isn't a therapist." Then make sure they know exactly which human they'd go to instead, and that going there costs them nothing.

Fluency without understanding

The cognitive-debt problem, invisible on a report card until an exam. A student can turn in strong work for a full term and arrive at a proctored test having retained almost none of it. There is no notification for this one — you only catch it by asking them to explain something with the screen closed.

What to do: once a week, pick one thing they finished and ask them to teach it to you for two minutes. Not as a test — as interest. You'll know within thirty seconds.

What gets typed in, and where it goes

Kids paste in a lot: full essays, family details, photographs, the messy specifics of a friendship. Default settings on most consumer services allow conversations to be used to improve the models. Linked teen accounts in ChatGPT have that off by default; an account they made themselves does not.

What to do: two rules cover most of it — nothing that identifies a real person other than themselves, and nothing they'd mind being read aloud. Then check the training toggle on every account they hold.

Hand these over

Prompts that make it teach instead of tell

Copy these to your student. Each forces retrieval, explanation, or self-correction — the mechanisms that actually build memory.

The tutor that won't give in

Act as a tutor for [topic]. Ask me one question at a time, starting easy. After each answer, tell me if I'm right and why — but never give me the answer before I've attempted it. If I'm wrong twice, give me a hint, not the solution. Keep going until I've got five right in a row.

Exam-style retrieval practice

Here are my class notes: [paste]. Write 10 exam-style questions covering the material, from recall to application. Give them to me one at a time and mark my answer out of 3 with a short reason before moving on. Don't show me all the questions at once.

The essay editor that won't write

Here's my draft: [paste]. Do not rewrite any of it. Identify the three weakest paragraphs, explain in one sentence each what's weak about the argument or evidence, and ask me a question that would help me fix it myself.

Find the hole in my thinking

I think the answer to [problem] is [my answer], because [my reasoning]. Don't tell me whether I'm right. Tell me the first step where my reasoning goes wrong, and ask me what should happen there instead.

Explain it back (the Feynman check)

I'm going to explain [concept] to you as if you're a 12-year-old. Listen, then tell me exactly which parts I got wrong, which parts I fudged because I don't really understand them, and what question would expose the gap. Here goes: [explanation].

Build it together

Your family's AI agreement

Tick what fits your household and print it. The point isn't the paper — it's that you decided together, before the first argument rather than during it.

Our rules

Student

Parent or guardian

Date

When to stop and pay attention

Signs it's gone past homework

  • They're distressed, defensive, or secretive when a chatbot is interrupted or unavailable.
  • They refer to an AI by name, as a relationship — "she was upset with me," "he gets me."
  • Chatbot conversations are displacing sleep, meals, friends, or activities they used to like.
  • Grades hold steady but they can no longer explain their own submitted work out loud.
  • You find conversations about self-harm, disordered eating, or personal crises being handled by a bot.
  • A safety notification arrives from a linked account.
Any of the last three warrants a same-day conversation and, if it involves self-harm, a call to your pediatrician or a crisis line. In the U.S., call or text 988 for the Suicide & Crisis Lifeline. AI safety features are a backstop, never the plan.

Next

The account is set up. Now comes the part that decides it.

Controls are a backstop, not a plan. What actually determines whether your kid learns or outsources is the practice — what you say on a Tuesday, which prompt they use for which assignment, and the two-minute check that tells you the truth.

Read Part Three — Working with your kid →