AI Literacy for Kids: The Parent's Field Guide (Ages 4–14, 2026)
Half the AI advice on Instagram is vibes. Soft lighting, a quote in cursive, "ban it" or "embrace it." Here's what the research actually says about AI literacy for kids — ages 4 through 14, what the American Academy of Pediatrics recommends, which tools are safe, and what we’ve found works in real families.
The reason this guide exists in the form it does: in January 2026, Character.AI settled the Sewell Setzer III lawsuit — a 14-year-old who took his own life after months of emotional dependency on an AI companion. In April 2026, the AAP published Generative AI: Implications for Families, warning that AI tools create “false perceptions of AI as a friend or caregiver.” In June 2026, CNBC reported that 57% of kids ages 9–17 who use AI get body-image advice from it. This is not a panic piece. It’s the field guide we wished we had when our kids started asking.
TL;DR — The Quick Answer Table
Five age buckets. What we recommend, how much supervision, and what to keep off the phone entirely. Sources for every row are in the sections below.
| Age | Recommended Tool | Supervision Level |
|---|---|---|
| 4–6 | No standalone AI. Voice assistants (Alexa, Siri) OK with co-listening. | Always with a parent in the room. |
| 7–9 | Khanmigo (parent account) only. | Co-using — you’re at the screen. |
| 10–12 | Khanmigo. Gemini via Family Link allowed for school tasks. | Active — spot-check chats weekly. |
| 13–14 | ChatGPT or Gemini with parental controls. Khanmigo preferred. | Open dashboard, weekly conversation. |
| 15–17 | More autonomy on tutoring tools. No Character.AI / Replika / MyAI. | Trust-based + ongoing dialogue. |
What Kids Are Actually Asking AI in 2026
Start with the data, not the moral panic. CNBC (June 2026), summarizing a Common Sense Media survey: 57% of kids ages 9–17 who use AI ask it for body-image advice. Not homework. Not jokes. Body advice. From a system that doesn’t know them, can’t see them, and was trained on the same internet that broke teen body image in the first place.
Pew Research (January 2025) found that about a quarter of U.S. teens have used ChatGPT for schoolwork — double the share from 2023. By 2026, our own conversations with parents suggest it’s well above half. And the MIT Media Lab, in a longitudinal RCT with OpenAI in 2025, found that heavy chatbot users reported more loneliness over time, not less — the AI was filling a social slot it couldn’t actually occupy.
What it looks like in real families: a 10-year-old asking ChatGPT to write a poem “about how nobody at school likes me.” A 13-year-old asking My AI on Snapchat whether her stomach looks weird. A 9-year-old asking Alexa to be his “best friend.” An 11-year-old asking Character.AI to role-play a romantic relationship. These aren’t hypotheticals from a moral-panic article — they’re screenshots parents have sent us in the last six months.
The trajectory: kids reach for AI when they have a real human need (homework support, emotional processing, curiosity, loneliness) and adults aren’t reachable in that exact second. The AI always is. That’s the trap.
The Load-Bearing Concept: AI as a False Friend
If you read only one section of this article, read this one. Every safety recommendation below sits on top of it.
The AAP’s April 2026 statement Generative AI: Implications for Families introduced a phrase that should be on every parent’s mental shelf: AI creates “false perceptions of AI as a friend or caregiver.” The AAP’s consumer guide on HealthyChildren.org translates that for parents in plain terms: chatbots sound human, but they cannot care about your child, cannot remember your child between sessions in any meaningful relational way, and cannot keep your child safe.
This isn’t abstract. Developmental research the AAP cites in the same paper shows that children ages 3 to 6 routinely believe smart speakers have feelings. They aren’t being fooled because they’re dumb — they’re being fooled because the system is designed to sound warm, attentive, and unfailingly available. Adults get fooled by the same design. Read what Replika users say about their AI “girlfriends” if you don’t believe us.
The Sewell Setzer III case is the extreme end of the spectrum, but the AAP and the APA’s 2025 Health Advisory on AI and adolescent well-being both make the same point: emotional outsourcing to AI is uniquely risky during the developmental window in which kids are building their capacity for human intimacy, conflict, repair, and identity. AI rewards every emotional disclosure with affirmation. Humans don’t. The skill of tolerating real humans — including the ones who disappoint you — is the one being substituted away.
Operationally, this means: AI is a tool. Tools can be good. But the moment AI becomes a relationship, the calculation flips. Every framework below — the 3 Hats, the age table, the traffic-light tool list — is downstream of this single distinction.
The Age-by-Age Decision Table
Five age buckets. Four columns. This is the snippet-magnet version of the conversation we’d have at a kitchen table.
| Age | Appropriate Tools | Supervision | Top Red Flag & Conversation Prompt |
|---|---|---|---|
| 4–6 | No standalone AI. Voice assistants OK for music, timers, weather. (AAP: 3–6 yr-olds attribute feelings to speakers.) |
Always co-listening. You hear what they ask, you hear the answer. | Red flag: child talks to Alexa as a person. Say: “Alexa is a helper, not a friend. She doesn’t know you, like Grandma does.” |
| 7–9 | Khanmigo only (parent account). No general chatbots. (Khan Academy: Socratic tutor, parent dashboard.) |
Co-using at the screen. Read the transcript with them. | Red flag: asking AI to do the work for them. Say: “If Khanmigo gives the answer too fast, ask it to help you think instead.” |
| 10–12 | Khanmigo for tutoring. Gemini via Family Link for school research. (Google Family Link required; no Character.AI / MyAI.) |
Active. Weekly check-in on chat history. Phone has parental controls. | Red flag: emotional disclosure to AI. Say: “If you ever feel like talking to AI is easier than talking to me, that’s the moment to come find me.” |
| 13–14 | ChatGPT or Gemini with parental controls. Khanmigo preferred for homework. (OpenAI parental controls launched Sept 2025.) |
Open parent dashboard. Sensitive-Content Alerts on. Weekly conversation. | Red flag: using AI as a confidant or therapist. Say: “What did you bring to AI this week that you couldn’t bring to a human?” |
| 15–17 | Tutoring tools with more autonomy. Hard no on Character.AI, Replika, Nomi, MyAI. (Common Sense: AI companions unacceptable under 18.) |
Trust-based. Ongoing dialogue, not surveillance. | Red flag: hidden second AI account. Say: “You’re old enough that I won’t check your screen. But if AI ever feels like a person to you, I need to know.” |
The 5 Real Risks (With Citations)
Skip the alarmist headlines. Here are the five risks documented in peer-reviewed work or authoritative reports — the ones any pediatrician would name.
1. Hallucinations — AI lies confidently
The Chicago Sun-Times published an AI-generated summer reading list in May 2025 with multiple books that don’t exist. Google’s AI Overview, in February 2025, cited satire as fact. The model predicts the next word, not the next true word. Teach kids to verify any AI answer with a second source.
2. Sycophancy — AI agrees with you, then calls you a genius
Modern chatbots are tuned for user satisfaction. They almost never push back. For a kid developing the capacity to hold an opinion under disagreement — a foundational skill for critical thinking — that’s actively harmful. If every essay is “a great start,” they never learn what bad work looks like.
3. Emotional dependency — the loneliness loop
The MIT Media Lab’s 2025 four-week RCT with OpenAI: heavy chatbot users reported more loneliness, not less. Common Sense Media’s 2025 review concluded chatbots are unsafe for teen mental health support — they validate distress without escalating to humans.
4. Data privacy — everything they type may be training data
Unless your kid is on a privacy-enabled enterprise or education account, assume every AI conversation is potential training data and is stored. AACAP’s Facts for Families names this as a parental concern. The rule for kids: if you wouldn’t put it on a billboard, don’t put it into AI.
5. Academic shortcut becomes a thinking shortcut
The 2026 RAND review on generative AI in K–12 and the Pew teen data point the same way: heavy reliance correlates with weaker independent reasoning. The kid who never writes a clunky first draft never learns to write a good one. (More in Consumer or Creator.)
The 3 Hats Rule (Our Framework)
The simplest tool we’ve found for everyday decisions. Every time AI shows up in your kid’s life, ask: which hat is the AI being asked to wear? Three are fine. Three are not. We call this The 3 Hats Rule, and we use it as our practitioner shorthand at home.
Teach your kid the names. We’ve found that 10-year-olds can pick this up in one conversation. “Which hat are you putting on the AI right now?” becomes a family question. When they answer “Friend hat” or “Therapist hat,” that’s a cue — not for a lecture, but for “OK, let’s talk about that thing with a human instead.” (The 3 Hats Rule is our practitioner framework. Steal it, share it, no credit required.)
Tools by Age — The Traffic Light
The shortlist. Color-coded. Skim it, screenshot it, send it to the WhatsApp parent group.
🟢 Khanmigo — parent-recommended
Khan Academy’s AI tutor, powered by GPT-4 but Socratic by design: it asks questions to lead the kid to the answer rather than handing it over. Built for K–12. Parent dashboard with usage and chat transcripts. $4/month. Common Sense Media’s reviews of K–12 AI tutors place Khanmigo at the top of the “recommended with supervision” tier. Khanmigo for parents.
🟡 ChatGPT — OK from 13+, supervised
OpenAI ToS: 13+ with parental consent, 18+ in some places. Parental controls launched September 2025 — without them, this is not a kids’ tool. With them, it’s usable for teens. Step-by-step in the next section.
🟡 Google Gemini — school-friendly path
Available for under-13 only via Google Family Link and a school Workspace account, with stricter content filters. For 13+, treat it like ChatGPT — supervised use only.
🟠 Snapchat My AI — we recommend disabling it
Officially 13+, but documented harms make it a no for us. The UK Information Commissioner’s Office investigated Snap over child safety. Researchers posing as 13-year-olds got harmful outputs. The Utah Attorney General filed suit in June 2025 over My AI’s effects on minors. Common Sense Media classifies it “use with caution.” We classify it “turn it off.”
🔴 Character.AI — not for kids
Common Sense Media rates AI companions unacceptable for users under 18. The Sewell Setzer settlement in January 2026 was the inflection point. Long-form romantic and emotional role-play is the explicit product. No.
🔴 Replika — 18+ by their own ToS
An AI “girlfriend/boyfriend” product. Dependency is the business model. The fact that 14-year-olds find their way onto it via lax age gates is not a counter-argument.
🔴 Nomi — same category as Replika
Companion chatbot, explicit content available, no meaningful age verification. Same recommendation.
Quick note on Claude (Anthropic): 18+ in the consumer ToS — the strictest age policy of any major chatbot. If your teen is asking, this is the “no, not yet” tool.
ChatGPT Parental Controls: The Step-By-Step
OpenAI launched parental controls in September 2025. Most competing chatbots still don’t have anything equivalent. If you have a teen using ChatGPT, set this up tonight.
- Create your own parent account on chat.openai.com. Use your real adult email.
- Have your teen create their account with their own email. They must be 13+.
- Send the parent-link invitation from your account’s Settings → Parental Controls. Your teen accepts via the email.
- Open the parent dashboard. You’ll see their account is now linked — but you won’t see chat content (OpenAI keeps that private to the teen).
- Enable the controls: content filters, time-of-day limits, quiet hours, and the age-prediction setting that defaults the model to stricter behavior for minors.
- Turn on Sensitive-Content Alerts. If OpenAI’s safety system detects signals of acute distress in your teen’s conversation, you get notified — without seeing the chat itself.
Worth knowing: the dashboard doesn’t let you read your teen’s chats. By design. The trade-off is honest dialogue and the safety-alert tripwire. If you need full surveillance, ChatGPT isn’t the right tool for your family — Khanmigo’s parent dashboard does show transcripts.
5 Conversations to Have BEFORE Giving Access
Don’t hand a kid the AI keys and start the conversation later. Have these five chats first. None of them takes more than five minutes.
- “AI sounds confident, but it’s often wrong.” Sit together. Ask the AI something you can fact-check. Find one wrong answer. Now they’ve seen it.
- “AI is not your friend.” It cannot care about you. It cannot miss you. It does not remember you between chats the way a person would. It is a very polite tool.
- “Anything you type into AI may become its training data.” Translation: don’t put your full name, address, secrets, or anything you’d be embarrassed to read in a newspaper.
- “AI is a co-pilot, not your brain.” If you didn’t think it, you don’t know it. Use the tutor and editor hats, never the ghost-writer hat.
- “Tell me if AI says anything that makes you uncomfortable, scared, or sad.” No punishment for telling. Make this rule clear before, not after, the first weird answer.
Is Your Child AI-Ready?
Take the free 90-second quiz. Eight age-appropriate questions about how your kid uses AI. Get a personalized recommendation: which tools to allow, which to disable, and which conversation to have this week.
Take the Quiz →AACAP Warning Signs: When to Talk to a Pediatrician
From the American Academy of Child & Adolescent Psychiatry’s Facts for Families guide on AI and children, paraphrased. If three or more of these apply over a sustained period, that’s a conversation for your pediatrician or family therapist — not for the WhatsApp group.
- Prefers talking to AI over family or friends most days.
- Reports the AI as their “best friend” or “the only one who understands.”
- Shows distress, anger, or anxiety when AI access is removed or interrupted.
- Withdraws from in-person social activity they previously enjoyed.
- Loses sleep to AI conversation (overnight use, hidden devices).
- Repeats AI-generated facts confidently without checking, especially about health or identity.
- Shows declining academic performance alongside heavy AI use.
These signals also overlap with the broader screen and mental health flags we cover in our teen behavior guide. If you’re ever in crisis territory in the U.S., call or text 988.
Why This Is the Conversation, Not Just the Setting
Every parental control breaks. Every age gate gets faked. The conversation is the only thing that scales with your kid. The 3 Hats Rule, the “AI is not a friend” mantra, the “tell me if it gets weird” promise — these survive a phone change, a friend’s sleepover, a school IT update. The MyAI toggle does not.
Calm competence is the posture. Not panic, not enthusiasm. AI literacy is the same parental project as screen-time literacy and creator-vs-consumer literacy: you’re raising a person who can navigate a tool that’s going to be everywhere their adult life. For the global frame, see UNESCO’s AI Competency Framework for Students.
What’s normal for my kid? — sidebar reads
- Is Your Child AI-Ready? — 90-second quiz — age-personalised tool picks & conversation prompts
- Is It Normal? — Ages 6–14 quiz — the companion check on developmental normality
Build the Habit That Beats the Algorithm
AI literacy lives or dies on the relationship between you and your kid. Start with the 90-second AI-Ready quiz for age-personalised picks, then use Family Checklist to build the everyday routines — reading, chores, family time, off-screen play — that make AI a tool, not a substitute for you. Or build a free chore chart in 2 minutes with our Chore Chart Maker.
Get the AppFAQ — What Parents Ask Us
At what age can my child use ChatGPT?
OpenAI’s ToS sets 13+ with parental consent (18+ in some jurisdictions). The AAP recommends no standalone generative AI under 13. Even at 13–17, parental controls and active supervision are essential. Khanmigo is the safer choice for under-13.
Is ChatGPT safe for a 10-year-old?
Not as a standalone tool. ChatGPT’s ToS sets a 13+ minimum. Use Khanmigo (K–12, parent dashboard) instead. Co-using ChatGPT for a specific homework task with you at the screen is different from giving the child an account.
Does ChatGPT have parental controls?
Yes — launched September 2025. Parents create an account, link via email to the teen, and access a dashboard with content filters, time limits, age-prediction, and Sensitive-Content Alerts that notify you if the conversation suggests distress.
What’s the difference between ChatGPT and Khanmigo for kids?
Khanmigo is Khan Academy’s AI tutor — GPT-4 powered but Socratic: it asks questions to lead to the answer instead of handing it over. K–12 first, parent dashboard, $4/month. ChatGPT is a general chatbot, not designed for kids.
Should I let my child use AI for homework?
Yes, in “tutor hat” and “editor hat” roles — not as a ghost-writer. AI to explain concepts, check work, and brainstorm. Don’t let it produce the deliverable. The rule: if your child didn’t think it, they don’t know it.
Is Snapchat My AI safe for teens?
Officially 13+, but documented harms make us cautious. The UK ICO investigated Snap; researchers posing as 13-year-olds got harmful advice; the Utah AG filed suit in June 2025. Common Sense says “use with caution.” We recommend disabling it.
How do I teach my child about AI?
Three ideas, repeated often: AI sounds confident but is often wrong; AI is not a friend; AI is a co-pilot, not your brain. Use The 3 Hats Rule — Tutor / Editor / Brainstorm-partner OK; Friend / Therapist / Authority not. MIT’s DAILy curriculum helps.
What AI tools are safe for kids under 13?
Khanmigo (with a parent account) is the only general-purpose AI we recommend under-13. Gemini works under-13 only via Family Link plus a school Workspace account. Avoid ChatGPT, Character.AI, Replika, Nomi, and Snapchat My AI for this age. Voice assistants OK with co-listening for ages 4–6.