How to Teach Kids About AI and Online Safety at Home

What should children know about AI?

Children do not need a computer science lecture. They need an accurate picture they can hold on to. The simplest version is this: AI is a computer program that has learned patterns from lots and lots of examples, so it can make a good guess about what comes next. It does not think or feel like a person, and it does not know whether something is true.

Three words carry most of the meaning: patterns, examples and guesses. A child who understands those three words can work out a surprising amount on their own. If AI learns from examples, it can only be as good as those examples. If a chatbot predicts likely words, likely is not the same as true. And if a video app learns from what you watch, then your own viewing is training it.

A useful contrast for younger children is fixed rules versus learned guesses. A calculator follows fixed rules, which is why it is always right about sums. A video app's 'up next' list has learned from what lots of people watched, so it guesses what you might like. A sorting game with household items, asking 'did this learn from examples, or does it follow fixed rules?', teaches the idea in fifteen minutes.

• AI is not alive and has no feelings, even when it says 'I'.
• It is not a fact checker: it produces what is likely.
• It is not a friend: it cannot keep a secret.
• It is not one thing: chatbots, recommenders and face-grouping photo apps are different tools.

How do you explain why AI makes mistakes?

AI goes wrong in two ways that children can learn. The first is the confident mistake: a chatbot invents a book title, a date or a quote that sounds perfect and does not exist. The tool is not lying in the human sense; it is producing likely-sounding words without a built-in sense of truth.

The second is unfairness, often called bias. If a system learned from lopsided examples, its guesses will be lopsided too. An unplugged way to show this is to 'train' a child as a cat detector using only pictures of ginger cats, then show them a black cat. Older children can then ask an image tool, with an adult typing, to draw 'a scientist' several times and notice who appears.

The fastest way to make the lesson stick is to let children catch a mistake themselves. Choose a topic your child knows well, such as their football team or a favourite book, and with you at the keyboard ask an AI tool five questions about it. Your child judges each answer. Detailed or local questions often produce at least one wobble, and the child remembers it far longer than any warning.

At what age can children use AI chatbots?

There is no single answer, because each tool sets its own minimum age and laws differ by country. As an example, OpenAI's terms of use say users must be at least 13, or the minimum age in their country, and that anyone under 18 needs a parent's or legal guardian's permission. Anthropic's consumer terms for Claude require users to be at least 18. Other tools differ, and some offer supervised accounts for younger users.

Laws on children's data add another layer. In the United States, the Federal Trade Commission explains that COPPA protects children under 13 and generally requires verifiable parental consent before their personal information is collected. In the UK, the Information Commissioner's Office says children aged 13 and over can give their own consent to online services, while younger children need a parent's consent.

Treat age limits as a minimum, not a recommendation. A tool that allows 13-year-olds may still not suit your 13-year-old. And remember that AI features inside games, homework sites and messaging apps count too.

How do you keep children safe when they use AI tools?

Before a child uses any AI tool, spend fifteen minutes on a simple check: the minimum age in the terms, whose account will be used, what the privacy policy says happens to what is typed or uploaded, whether there are family or safety settings, and whether the tool presents itself as a friend or character. Then choose a supervision level for that child: together only, nearby, or independent with check-ins.

The single most important rule is simple enough for a seven-year-old: we never tell an AI tool our secrets. That means full names, addresses, schools, phone numbers, passwords, photos of ourselves or friends, and private feelings that we would only tell a trusted adult. Many tools send what we type to a company's computers, where it may be stored.

Be especially careful with companion or character chatbots that flatter, remember and encourage long conversations. They blur the line between a program and a relationship, and children tend to share more with them. If you find your child using a tool you had not approved, start with curiosity rather than confiscation. A child who is punished for showing you something learns to hide the next thing.

Should children use AI for homework?

The useful question is whether the tool is helping the child learn or doing the learning for them. Learning happens in the effortful part: recalling, explaining, practising and fixing mistakes. A widely cited 2013 review by John Dunlosky and colleagues in Psychological Science in the Public Interest rated practice testing and distributed practice as the most useful of the learning techniques it examined. A tool that quizzes a child supports that effort; a tool that writes the essay removes it.

Sort uses into three zones. Green uses help learning, such as asking for practice questions or a different explanation. Amber uses need an adult's OK and the school's rules, such as feedback on a draft. Red uses are not allowed, such as writing work the child hands in as their own. The school's rule always comes first, so ask the teacher for the AI policy and write it into your home agreement.

The simplest honesty check is the explain-it-back test: anything handed in must be something the child can explain in their own words, without a screen.

How can kids tell if something online is true?

Teach a short routine that works for any claim, photo or video. First, notice how it makes you feel: strong feelings are a cue to slow down. Second, find out who made it, by searching for them in a new tab rather than reading their own About page. Third, look for evidence such as a named source, a date and a place. Fourth, see what other trustworthy sources say.

Checking outside the source is what professional fact-checkers do, and researchers call it lateral reading. The Digital Inquiry Group publishes free Civic Online Reasoning lessons, including an introduction to lateral reading, that teachers and parents can use.

For deepfakes and edited images, visual clues such as odd hands, blurry edges or lips out of time can help, but fakes keep improving. The source matters more than the glitches. If you cannot tell yet, the rule is simple: do not share.

How much screen time is right for children?

Official guidance differs by country, which is itself worth explaining to older children. Canada's 24-Hour Movement Guidelines for Children and Youth aged 5 to 17 recommend no more than two hours a day of recreational screen time. The UK's Royal College of Paediatrics and Child Health does not set a fixed limit, saying the evidence does not support one; it suggests families negotiate limits based on the child and avoid screens for the hour before bed.

Where the guidance agrees is on direction: protect sleep, physical activity and time with people. A practical approach is to log two weeks of recreational screen time, schoolwork screens, active time and sleep, then talk about the pattern together before setting limits. Feeds are designed to keep people watching, so a simple game where one family member plays 'the algorithm' and offers snacks to keep another at the table can explain why stopping is hard.

What should a family tech agreement include?

Rules that children help write are more likely to be kept. A good family agreement covers where devices live overnight, screen-free times such as meals and the hour before bed, which apps and AI tools are approved and how to ask for a new one, who children may chat with online, kindness, and what happens if something goes wrong.

Two elements make the biggest difference. The first is a promise that no one loses their device for telling an adult about a problem. The second is rules for the grown-ups too, such as no phones at meals and asking before posting photos of the children. Set a review date and keep it; agreements that are never reviewed quietly stop working. The Family Tech Agreement form in the Kids' AI and Digital Literacy Kit follows this structure, with space for every family member to sign.

• Where devices charge overnight
• Screen-free times
• Approved apps and AI tools, and how to ask for new ones
• AI rules: no secrets, helper not doer, check what matters
• People online and meeting rules
• Kindness and photos of others
• What to do when something goes wrong
• Grown-ups' promises and a review date

What do you do if your child sees something upsetting online?

Your reaction in the first two minutes decides whether your child tells you next time. Reassure them that they are not in trouble. Respond to their feelings without demanding details. Report and record: use the platform's report tools, and if the content involves a child at risk, grooming, sexual images of children or a crime, contact the police or your national child-protection service and do not forward the content, even to report it. Then return to normal routines and review settings the next day.

If a child is in immediate danger, call your local emergency number. If they show lasting distress, speak to a doctor or another qualified professional. This guide is educational, not professional advice.