AI companions and children: what parents should know

A friendly voice still needs boundaries

A chatbot can answer a homework question, invent a story and respond to 'I had a terrible day' in the same conversation. For a child, those activities may sit inside one familiar app.

That makes the product's label a starting point. Parents need to know what the tool does, how it responds when a conversation becomes personal and what happens to the information a child shares.

An assistant usually centres on tasks. A companion is designed around conversation or a simulated relationship. In practice, those uses can overlap. [C1]

What Australian children are reporting

eSafety's August 2026 report, Talking to machines, draws on a survey of 1,950 Australian children aged 10–17, conducted between 2 February and 4 March 2026. It found that 78% had used an AI assistant and 8% an AI companion. These are different categories and should not be added together. [C1]

Among children who had used either type, 20% reported potentially inappropriate or harmful interactions and 32% had shared personal or potentially sensitive information. Those percentages describe users in the survey, not all Australian children. These self-reported experiences do not establish what caused any individual child's wellbeing to change. The survey does not describe children under ten or constitute testing of KYREC. [C1]

A convincing reply is still generated

eSafety's February 2025 advisory identifies risks including inaccurate or harmful advice, sexualised content and unhealthy dependence on simulated relationships. A system can produce caring language without having human feelings or a reciprocal relationship with the child. [C2]

Explain the distinction plainly: the software generates replies. It can be entertaining or useful, and it can also be wrong. A confident or affectionate response is not proof that it understands the situation.

For a parent working away, reassurance needs to rest on understandable controls and a trusted adult available to help. A cheerful character or familiar voice does not establish that a service is suitable for a child.

Review the actual app together

eSafety advises adult supervision for children using generative AI tools and warns that content moderation may not provide sufficient protection. Its parental-controls guidance recommends combining settings with broader online-safety practices and reviewing them regularly. [C3]

The following is an editorial checklist, not a certification that a product is safe. Check the service's current age terms. An app-store rating or a 'family' description does not answer every question about permitted use.

Look at the conversation. Is the tool helping with a defined task, or encouraging an ongoing relationship? Can it introduce characters, images or topics the family did not choose?

Check the data. Find out whether chats are stored, used for training or reviewed by people. Locate memory and deletion controls. Consider what happens to uploaded photographs and voice recordings.

Find reporting, blocking, purchase restrictions and relevant parent settings. Work through how they operate rather than assuming they are switched on. Make sure the child knows how to leave a conversation and reach a trusted adult.

Try a conversation before setting rules

Ask what your child enjoys about the tool. They may be interested in storytelling, a game character, help understanding something or a place to ask questions without embarrassment.

Then discuss hypothetical situations. What would they do if a bot contradicted a teacher? What if it asked for a school name, a photograph or a secret? What if it said that nobody else understood them? These are conversation prompts, not a test.

Explain your boundaries too. You might allow a shared, supervised creative activity while deciding that an open-ended relationship chatbot is unsuitable. No household rule makes the permitted activity risk-free.

An AI character should not be presented as a substitute parent or a secret confidant who takes priority over real relationships. That is our editorial standard, not a claim that any particular product meets it.

Keep disclosure possible

If a child shows you a disturbing exchange, begin by listening. eSafety encourages open conversations and support rather than responses that discourage children from asking for help. Stop the interaction and use the service's reporting tools when appropriate. Seek help from a trusted professional if the child is distressed or you have concerns about their wellbeing. [C2]

Parents should not have to become full-time software moderators. Providers remain responsible for their services' design and safeguards; parent settings do not remove those responsibilities.

For a family operating system, or any product used around children, the questions should stay concrete: who can use it, what can it do, what information does it keep, and how does a person stay in control? Each specific product claim needs its own verification.