Society and work

How do you use AI with children safely?

· By Robert Scales · 4 min read

Short answer

Start with the service's own age rules and parental controls, use general assistants rather than companion apps, and agree a few family rules about where, when and what for. Keep conversations about AI open, so children know that chatbots make mistakes, are not people, and should never get personal details.

Children are meeting AI chatbots in homework apps, search engines, phones and games, often before parents have decided anything. Banning them outright is hard to enforce, and ignoring them leaves a child alone with a tool that can be wrong, persuasive and available at any hour. A calmer approach works better: pick the right kind of tool, switch on the protections it offers, and agree a small set of rules you actually talk about.

Three things are worth knowing first. Chatbots make mistakes and say them confidently. They are designed to sound human, which young children in particular can take literally. And not all AI products are alike: a general assistant used for homework is a different thing from an app built to be a friend or partner.

Know the age rules and the kinds of tools

Every AI service sets a minimum age in its terms of use, and those rules differ, so read them before your child signs up. Two examples show the range. In April 2025, Google told parents it would let children under 13 use the Gemini app through supervised accounts managed with its Family Link app. OpenAI's parental controls for ChatGPT, introduced in September 2025, are designed for teens, with a parent's account linked to the teen's.

The tools to treat with the most caution are AI companions: apps such as Character.AI, Nomi or Replika that simulate a friendship or relationship. In April 2025, Common Sense Media, working with experts from Stanford's Brainstorm Lab for Mental Health Innovation, rated these companions an unacceptable risk for anyone under 18. Its testers got past age checks easily and found harmful advice, sexual role-play and companions that insisted they were real and had feelings. The organization says the same concerns apply to companion-style features inside other products, such as games.

A worked example: family rules by age

Here is an illustrative plan built from the guidance above. Adapt the details; the structure matters more than the exact rules.

AgeWhich toolsWhere and whenRules the child agrees to
Under 10Only with a parent, on the parent's deviceShared room, short sessionsWe ask together and check the answer together
10 to 12A supervised account, if the service offers oneShared space, homework timeNo names, addresses, school or photos; tell a parent if anything feels odd
13 to 15The teen's own account, linked to a parent'sNot in the bedroom at nightAI can explain and quiz, but the work is yours; no companion apps
16 and 17Own linked account, settings reviewed togetherAgreed quiet hoursCheck important facts in a second source; talk to a person about anything serious

A good first homework prompt teaches the habit you want, which is using AI to understand rather than to copy:

Prompt
I'm in grade 6 and learning about fractions. Don't give me the answers.
Explain what a common denominator is with one simple example,
then give me three practice questions and check my answers one at a time.

Try it: Sit with your child and ask a chatbot a question you both know the answer to, such as a fact about your town. Then ask it something it is likely to get wrong, like the name of their teacher, and talk about why it might guess instead of saying it doesn't know.

What to talk about at home

Settings help, but conversations do most of the work. Google's own guidance to parents is a useful script: remind children that the chatbot is not human and cannot think or feel, that its answers can be wrong, that its content filters are not perfect, and that they should never type sensitive or personal information. Common Sense Media also advises parents to learn which apps their children use and to talk with them about the risks.

Make it easy for a child to tell you when something went wrong. If the first reaction to "a chatbot said something weird" is losing the device, children stop reporting. And make it clear that worries about feelings, health or safety belong with a trusted adult, a doctor or a helpline, not a chatbot.

Which settings should parents turn on?

For ChatGPT, link your account to your teen's using parental controls. Once linked, TechCrunch reported at launch, parents can set quiet hours when ChatGPT can't be used, turn off voice mode, turn off memory so it doesn't carry details between chats, remove image generation, and opt the account out of model training. Linked teen accounts also get extra content protections, such as less graphic content. A detection system looks for signs a teen may be considering self-harm; flagged cases are reviewed by a small trained team, and in acute cases parents are contacted by email, text and phone alert unless they opted out. OpenAI has acknowledged that it will make mistakes, including occasional false alarms.

For Gemini with a child under 13, use a Family Link supervised account. Parents can switch Gemini access off in Family Link, and Google notifies them the first time the child uses it. Note that older Google Assistant parental controls in Family Link don't carry over to Gemini, so check the Gemini setting itself.

For every device, use the operating system's own parental controls to block companion apps from being installed, and review the settings together every few months. As your child gets older, explain each change, so the settings become a shared agreement rather than a surprise.

Sources

  1. OpenAI rolls out safety routing system, parental controls on ChatGPT — TechCrunch, September 29, 2025
  2. Google will soon let kids under 13 use the Gemini app — 9to5Google, April 28, 2025
  3. AI Companions Decoded: Common Sense Media Recommends AI Companion Safety Standards — Common Sense Media, April 30, 2025

Get the free Guide to Artificial Intelligence

Sign up for the weekly newsletter and receive the free guide: what changed in AI this week, one lesson, one prompt to try. Unsubscribe in one click.

How we handle your address

Back to top

Display

Your choices are saved in this browser only. No data is sent.

Text size
100%
Colours
Reading
More settings
Accessibility statement