Children and AI: what parents should know

Separate the tool from the relationship

“AI use” can mean autocomplete, an image generator, an adaptive tutor, a general chatbot, or a companion designed to simulate friendship. Those systems present different developmental, privacy, and learning questions. A useful family response begins with what the child is using, why, what information it collects, and whether it is replacing an activity the child needs to learn or a human relationship the child needs to maintain.

The evidence is still developing. The American Psychological Association’s health advisory says AI is neither uniformly good nor bad for adolescents and that effects depend on the system, user, context, and safeguards. That supports active supervision and literacy, not either panic or automatic trust.

Start with a product inventory

Ask the child to show you which AI tools they use rather than asking only, “Do you use AI?” Check school platforms, game and social apps, search features, writing tools, and companion products. Read the service’s current age requirements and privacy notice together. Do not enter a false birth date to bypass an age limit; the restriction may reflect safety, consent, or data-practice concerns.

For a child under 13, the federal Children’s Online Privacy Protection Act applies to certain child-directed services and services with actual knowledge that they collect a child’s personal information. The FTC’s COPPA page explains the rule’s scope, but COPPA compliance is not a general safety certification. A product can comply with a privacy rule and still produce inaccurate, manipulative, or age-inappropriate responses.

Before allowing regular use, answer four questions:

1. Are conversations stored, reviewed by people, or used to improve the system?
2. Can the child delete history and the account, and what remains afterward?
3. Does the tool include ads, in-app purchases, public characters, direct messages, or location sharing?
4. What does it do when a young user mentions abuse, self-harm, eating disorders, sex, violence, or another crisis?

If those answers are unavailable or evasive, limit the information shared and reconsider whether the benefit justifies the uncertainty.

AI companions require a higher bar

A companion is not merely a reference tool. It may use a name, persistent memory, affection, role-play, rewards, notifications, or apparent empathy to sustain engagement. A young person may understand intellectually that it is software while still experiencing the interaction as emotionally meaningful.

The APA warns that adolescents may have difficulty distinguishing simulated empathy from genuine understanding and that AI relationships may displace real-world relationships. In 2025, the FTC opened an inquiry into companion chatbots, including how companies test harms, enforce age restrictions, disclose data practices, and protect children and teens. An inquiry is not proof that every companion causes harm, but it shows that major safety questions remain unresolved.

Co-create boundaries instead of conducting secret surveillance by default. Decide which products are allowed, when they may be used, whether companion or romantic role-play is permitted, and what kinds of conversations must move to a trusted adult. Explain any monitoring honestly unless doing so would create an immediate safety problem. Review the arrangement as maturity and evidence change.

Watch for functional changes: lost sleep, withdrawal from friends, distress when separated from the bot, escalating sexual or violent content, spending that is hidden, or treating the system as the only trustworthy adviser. Those signals merit a calm conversation and, when persistent or serious, professional support. Do not shame the attachment; shame can drive it underground.

Keep health and crisis decisions human

General chatbots can sound clinically confident without knowing a child’s history, observing nonverbal cues, checking facts reliably, or carrying a clinician’s duty of care. The APA’s mental-health chatbot advisory states that there is no consensus that general generative chatbots have the qualifications required to provide mental-health care or diagnosis.

Set an explicit rule: AI may help formulate questions for a clinician or explain a term, but it does not diagnose, change medication, create a diet for an eating disorder, or decide whether a crisis is serious. Verify health claims with a pediatrician, pharmacist, or authoritative public-health source. Teach the child never to paste medical records, therapy notes, identity numbers, or another person’s private information into a general chatbot.

If the child discloses abuse, self-harm, exploitation, or danger in a chat, focus first on safety and connection. Preserve relevant evidence if doing so is safe, but do not interrogate the child or threaten punishment for using the tool. Seek professional guidance about reporting and care.

Protect against synthetic-image abuse and extortion

AI can generate or alter intimate-looking images without a real photograph of the depicted body. Children should know that creating, requesting, possessing, or sharing sexual images involving minors can cause severe harm and legal consequences even when content is synthetic or altered. Family rules should prohibit using anyone’s face or voice without permission and prohibit requests for intimate material.

Teach a non-negotiable response to sextortion: stop contact, do not pay, do not send more images, save evidence, and tell a trusted adult. Payment rarely creates control over copies. The FBI’s guide to financially motivated sextortion describes offenders using manipulation and altered images and provides reporting resources. A child who comes forward needs protection, not blame.

Preserve learning when AI helps with schoolwork

Begin with the teacher’s or school’s current policy. Acceptable use can differ by class and assignment: brainstorming may be allowed where generated prose is not; translation or accessibility support may be permitted with disclosure; citation requirements vary. U.S. Department of Education guidance emphasizes parent and teacher engagement, privacy, and AI literacy.

Use a “think first, tool second, verify last” routine:

• The student attempts the problem or outlines an answer before asking AI.
• The prompt requests coaching, questions, or feedback rather than a finished submission.
• The student checks every factual claim against course materials or reliable sources.
• The student can explain the final work without the tool and discloses AI use when required.

For mathematics or coding, ask the tool to identify where reasoning went wrong instead of supplying the next answer. For writing, compare the student’s draft with suggestions and require the student to decide what to keep. For research, never cite a paper until the student has opened it and confirmed that it exists and supports the claim. AI can fabricate citations and quotes.

Accessibility deserves nuance. A learner with a disability may use AI-supported transcription, simplification, organization, or language assistance under an education plan or school policy. Do not remove an accommodation under a general “no AI” household rule; coordinate with the student and school.

Teach AI literacy through small experiments

Children need neither a technical lecture nor blind fluency. They need repeatable questions: Who made this system? What is it optimized to do? Where could its answer come from? What might be missing? Who benefits if I keep engaging or buy something? What information am I giving it?

Run age-appropriate demonstrations. Ask the same factual question in different wording and compare answers. Request sources, open them, and check whether the claim appears. Generate two persuasive arguments for opposing positions. Examine a synthetic image for both obvious and invisible uncertainty. This shows that fluent output is generated content, not an authority signal.

Model the behavior adults expect. Say when you used AI, where you verified it, and what you chose not to share. Avoid posting a child’s face, voice, school, and routine by default while demanding perfect privacy from the child.

Write a short family agreement

Keep it understandable enough to use:

• We follow school rules and disclose AI help when required.
• We do not share passwords, addresses, schedules, health records, intimate information, or other people’s private data with a bot.
• AI companions do not replace sleep, schoolwork, friends, family, or professional care.
• We verify important claims with a human or reliable source.
• We never create deceptive or intimate media of another person.
• If an interaction becomes frightening, sexual, coercive, or secretive, a child can tell an adult without losing all device access automatically.

Review the agreement every few months because products and a child’s needs change. The aim is not perfect control. It is a home where the child can use helpful tools, recognize simulation and persuasion, preserve the difficult work of learning, and seek a real person when stakes become personal.