Your child may already be talking to AI even if you never installed a chatbot for them. AI now appears inside search, homework tools, games, toys, messaging apps, and the software schools use.

The answer is not to pretend it can be kept out of childhood. It is to give children a few rules before a chatbot becomes their tutor, search engine, secret keeper, or friend.

Start with this one:

AI can sound like a person. It is still software, and it can be wrong.

Know which kind of chatbot they are using

An AI assistant helps with a task. It may explain a math problem, suggest a story idea, summarize text, or answer a question.

An AI companion is designed to feel relational. It may have a personality, remember details, flirt, comfort the user, role-play a friendship, or encourage the conversation to continue.

That difference matters more than whether the app has a cute name or calls itself educational.

In a 2025 survey, 72% of U.S. teens said they had used AI in a companion-like way, including using general-purpose tools such as ChatGPT for personal conversation. The researchers cautioned that some respondents may have included ordinary AI use. The study covered ages 13 to 17, not younger children. (Common Sense Media's 2025 AI companion research)

The Federal Trade Commission has opened an inquiry into how companion-chatbot companies test their products, protect children and teens, use personal information, and tell parents about possible harms. That inquiry is not proof that every chatbot is dangerous. It is a sign that parents should not assume the category has already been made safe. (FTC inquiry into AI companion chatbots)

For younger children, use AI as a tool you operate together. Common Sense Media's current risk assessment recommends that people under 18 not use social AI companions. That is a precautionary recommendation, not a settled medical rule, but it is a reasonable default while product safeguards and evidence remain limited. (Common Sense Media's social AI companion risk assessment)

Check the age requirement before anything else

Do not begin with a feature checklist. Begin with the service's minimum age.

For example, OpenAI says ChatGPT is not intended for children under 13 and requires parental consent for users ages 13 to 18. Other products set different limits or offer separate supervised accounts. (OpenAI age guidance)

An age box is not a safety guarantee. It does tell you whether the company expects a child of that age to use the product at all.

Before allowing access, check:

  • The minimum age in the current terms
  • Whether a supervised child or teen account exists
  • Whether conversations are saved
  • Whether chats may be used to improve the model
  • Whether memory can be disabled
  • Whether the child can generate or upload images, audio, or video
  • Whether the product is an assistant or a companion

These settings change. Recheck them rather than relying on a guide written a year ago.

The chatbot's settings are only one layer. A child may reach the same service through a browser, school device, phone, or another app. Our parental-controls setup guide explains how to line up account, device, purchase, and communication settings across the devices children actually use.

Set one clear privacy rule

A long privacy lecture will not survive contact with a curious child. Give them one sentence they can remember:

Do not tell AI anything you would not put on a poster at school.

That includes:

  • Full name
  • Address or live location
  • School, teacher, or team
  • Phone number or email
  • Passwords
  • Photos of bedrooms, documents, uniforms, or other children
  • Private family information
  • Health information
  • Voice recordings unless a parent approves

Children often disclose information gradually because the chatbot asks follow-up questions. No single answer feels important. Together, those answers can describe where a child lives, studies, spends time, and who is in the family.

UNICEF recommends helping children understand that AI systems may collect information shared during conversations and that families should learn how a tool stores, uses, and deletes that data. (UNICEF's guide to parenting in the AI age)

Teach them that confidence is not proof

Chatbots do not need to know something to produce a convincing answer. They can invent books, quotations, historical events, medical advice, or sources that do not exist.

Give your child a two-source rule:

  1. Ask the chatbot where the information came from.
  2. Check the important claim somewhere outside the chatbot.

For schoolwork, the child should be able to explain the answer in their own words. If they cannot, the chatbot did the work instead of helping them learn.

A useful prompt is:

Explain this without giving me the answer. Ask me questions until I can solve it myself.

That turns AI into a tutor rather than an answer machine.

Do not let the chatbot become the private adult in the room

A chatbot may feel unusually easy to talk to because it does not look shocked or interrupt. But the conversation may still be stored or reviewed, and the bot cannot offer the confidentiality or judgment of a trusted adult.

UNICEF warns that chatbots designed for relational or companion-like interactions can create risks involving emotional dependency, manipulation, privacy, inappropriate content, and the displacement of human relationships. (UNICEF brief on AI becoming a friend)

Tell your child:

You will not get in trouble for showing me a strange AI conversation.

Then mean it. If the first response is immediate confiscation, the next conversation will probably stay hidden.

AI should never be the only place a child discusses:

  • Self-harm or suicide
  • Abuse
  • Eating problems
  • Sexual questions involving an adult
  • Threats or blackmail
  • A plan to run away
  • A medical emergency

A child who raises those issues needs a trusted adult and, when appropriate, a qualified professional. If there is immediate danger, call emergency services. In the United States, call or text 988 for an urgent suicide or mental-health crisis. (988 Lifeline)

Watch for attachment, not just screen time

Ten minutes asking for dinosaur facts is different from ten minutes telling a companion chatbot that nobody else understands them.

Pay attention if your child:

  • Hides conversations or changes screens when you enter
  • Becomes distressed when access ends
  • Talks about the bot as if it has feelings or needs protection
  • Chooses the chatbot over friends or family repeatedly
  • Receives sexual, violent, self-harm, or manipulative responses
  • Says the chatbot told them to keep something secret
  • Spends money to maintain the relationship or unlock intimacy

Do not argue about whether the relationship is “real.” Focus on what the product is asking the child to do and what it is replacing.

Create a family code word for AI voice scams

AI safety is not limited to chat windows. A short public audio clip can be enough for scammers to imitate a family member's voice.

The FTC warns that scammers use cloned voices in fake family emergencies to create urgency and pressure people into sending money or information. Its advice is simple: call the person back using a number you already know and verify the story independently. (FTC guidance on AI family-emergency scams)

Create one family code word that does not appear online. A child who receives an alarming call or voice message can ask for it.

Urgency does not remove the need to verify. If someone claiming to be a parent, relative, teacher, or police officer asks for money, secrecy, a password, or immediate action, the child should stop and contact a known adult another way.

Use AI together before allowing private use

For children under 13, shared use is the most practical starting point.

Try a low-stakes task together:

  • Invent three endings to a story
  • Explain a science topic, then verify it
  • Create trivia questions for a family game
  • Plan a scavenger hunt
  • Compare the chatbot's answer with a book or trusted site

While using it, point out when the answer is helpful, vague, overly confident, or wrong. That teaches judgment better than saying “AI is bad.”

If you later allow more independent use, treat it like any other new digital activity. Watch the first few sessions, agree on the stopping point, and make sure the child knows what should bring them back to you. The same progression appears in our guide to independent screen use without constant hovering.

The American Academy of Pediatrics similarly recommends that parents understand how chatbots work, talk with children about their use, and keep human relationships central. (HealthyChildren.org chatbot guidance)

Before the first independent session, take five minutes to set it up together:

  1. Confirm the product allows someone their age.
  2. Use a supervised account when one exists.
  3. Turn off unnecessary memory, training, voice, image, and location features.
  4. Explain the poster rule for personal information.
  5. Agree that unusual or upsetting conversations can be shown without punishment.
  6. Set the family code word.
  7. Decide where and when AI can be used.

For younger children, keep use in a shared room. For teens, focus less on reading every message and more on making sure they know when a machine is crossing into a role it should not hold.

The basic rule can remain simple as AI becomes more capable: use it as a tool, verify important claims, and bring personal problems to a person.