Your Child Calls an AI Chatbot a Friend. What Happens Next?

Your child calls an AI chatbot a friend. The useful response is neither to panic nor to treat the conversation as harmless because it happens on a screen. Ask what the child uses the bot for, look at how the product behaves, and make sure real people remain available when the subject is serious.

A companion bot is designed for continuing conversation, sometimes with a name, a personality and a memory of earlier exchanges. That can feel different from asking a search engine a question. UNICEF's 2026 policy brief describes children using such systems for information, advice, support and relationships. It also stresses that responsibility for safety cannot rest on parents alone: developers, platforms and regulators have roles.

Notice what the bot is becoming to your child

Begin with ordinary curiosity: “What do you like asking it?” and “Has it ever said something that felt strange?” A child may be using it to practise a language, invent stories or ask embarrassing questions. Those are different situations. The issue becomes more consequential when the bot is a principal source of reassurance, asks for repeated attention, encourages secrecy, or supplies advice about health, relationships or safety.

Do not assume that a friendly tone means the system understands the child. It produces responses from patterns and product rules. It cannot know the whole family situation, notice everything a person in the room would notice, or take responsibility for a harmful suggestion. That is why the question is about the role the tool is playing, rather than the mere number of minutes spent chatting.

Check the product, not just the child's behaviour

Look together at the account's age rules, privacy controls, chat history, memory and reporting options. Does the bot say clearly that it is AI? Can it initiate a conversation or send notifications? What happens when someone asks about self-harm, bullying or sexual topics? You need not stage a risky test with a child present. Product documentation, a parent account and a careful review of safety settings are better starting points.

UNICEF's guidance on AI and children, updated in 2025, specifically addresses companion systems, child rights and the need for safeguards. The UNICEF document describes those responsibilities in more detail. A visible safety label is useful, but it does not prove that every response will be safe. Settings can differ by product and account, so check the version your child actually uses.

Agree on a rule for high-stakes conversations

A simple family rule is easier to remember than a ban on every question: use AI for ideas and practice, but bring anything about your body, money, threats, relationships or urgent distress to a trusted person. Name those people together. A child who worries about getting in trouble for using the app may hide the very exchange you need to see.

Consider a hypothetical example. A teenager asks a companion bot whether to stop talking to a friend after an argument. The bot gives a confident answer based on one short account. The safer next move is to ask the teenager what happened, what they want to repair, and whether anyone is at risk. The bot can help put feelings into words; it should not decide the relationship.

If a conversation suggests immediate danger or serious distress, respond to the child directly and use appropriate local emergency or professional support. No article or chatbot can assess that situation from a few messages.

Keep the conversation open

Try a short shared exercise rather than a lecture. Ask the bot a low-stakes question the child knows something about, then compare its response with a book, a trusted site or a person. Talk about what it got right, where it guessed and how its friendly wording affected your first impression. The goal is not to catch the child out. It is to make checking and disagreeing with the tool feel normal. If the app lets you turn off notifications or memory, decide together whether those features help the child or merely make returning to the conversation more tempting.

Revisit the subject as the child grows and the product changes. Ask them to show you a useful answer and an unhelpful one. Explain that a bot can be warm and wrong at the same time. Invite them to challenge it, close it, and come back to a person.

This is not a verdict on every companion app. The evidence on long-term effects is still developing, and individual experiences vary. What parents can do now is understand the tool's design, establish a route to human support and make it safe for the child to tell them when an AI conversation goes badly.