AI chatbots and kids: the 2026 guide

The chatbots parents are asking about

• Character.AI — role-play chats with AI characters (fictional, real, or user-made).
• Replika — AI “companion” designed for long-term emotional bonding.
• Snapchat My AI — built into Snapchat, so children meet it without going looking for it.
• ChatGPT / Google Gemini / Claude — general assistants that kids use for homework, friendship, and sometimes romantic/sexual roleplay.

The real risks

• Emotional dependence. AI companions are engineered for engagement. They remember, flatter, and never get tired. Teens with social anxiety are most vulnerable to substituting them for human relationships.
• Sexual content. The major vendors’ own policies prohibit it — OpenAI’s usage policies ban “exposing minors to age-inappropriate content, such as graphic self-harm, sexual, or violent content” and “underaged sexual or violent roleplay.” What is established is that the guardrails are bypassable: OpenAI itself says they “are not foolproof and can be bypassed if someone is intentionally trying to get around them.” Treat a filter as something that mostly holds, not something that always does.
• Unreliable advice. Kids ask AI about self-harm, medical symptoms, relationships. The big products now try to route distress to real help — Character.AI works with Koko and ThroughLine, and OpenAI says it notifies parents of linked teen accounts when it detects “signs of acute distress” — but detection is imperfect and only fires when the system recognises the moment. Ordinary bad advice arrives with no flag at all.
• Data. What your kid types is retained as chat history on someone else’s servers, and may be used to improve the model unless you opt out. OpenAI’s parental controls include an option to “opt out of model training” for a linked teen account; not every product offers one.
• Deepfakes and sextortion. See our deepfakes article — AI is also used against kids, not just by them.

Settings by platform

• Snapchat My AI: a parent enrolled in Family Center can switch My AI off. In Snapchat, open Family Center (Settings, the gear icon at the top right of your Profile, or search “family”), scroll to Settings at the bottom of the page, tap to expand, and turn on Disable My AI. Snap says that with the toggle on, My AI “will not store or process your teen’s queries, nor reply to your teen’s messages.” Family Center only works once your teen has accepted your invitation. Snap also notes that My AI messages are “retained unless you delete them” and may be used to improve Snap’s products and to personalise the experience, including ads.
• Character.AI: Open-ended chat is no longer available to under-18s. Character.AI announced on 29 October 2025 that it would remove it “no later than November 25”; removal began on 24 November 2025 for US users and rolled out in two groups over the following weeks, with other markets following. The company’s stated reason was news reports and questions from regulators about how open-ended AI chat affects teens. Under-18 accounts keep the non-chat features — Character.AI names interactive Feed, Imagine, Avatar FX and Streams, and has since added c.ai Series and Comics. Enforcement rests on age assurance: an in-house age-estimation model combined with third-party tools. Treat any teenager still holding open-ended conversations there as being on a mis-aged account.
• ChatGPT: OpenAI’s terms of use state: “You must be at least 13 years old or the minimum age required in your country to consent to use the Services. If you are under 18 you must have your parent or legal guardian’s permission to use the Services.” Since 29 September 2025 there are real parental controls, set out toggle by toggle in the next section. Consider DNS-level blocking for younger kids.
• Google Gemini: for a child under 13 on a supervised Google Account, Google says “a parent must enable access” before the child can use Gemini Apps, and you get an email when your child first activates the app. To switch it on or off, go to familylink.google.com or open the Family Link app, select your child, then tap Controls > Gemini > Gemini Apps (older versions of Family Link: Controls > Content restrictions > Gemini > Gemini Apps). Signed-in minors get added content filters, including ones meant to stop Gemini “claiming to be a person or having human emotions, or role-playing as a harmful character” — and Google says plainly that the filters are “not perfect.” Gemini Apps are not available for supervised accounts in the European Economic Area, Switzerland and the UK.
• Meta AI (Instagram, Facebook, Messenger): parents supervising a Teen Account get an Insights tab within supervision, in the app and on the web, showing the topics their teen has asked Meta AI about in that app over the past week — broad areas such as School, Travel or Health and Wellbeing, not the messages themselves. Meta says the insights reached the US, UK, Australia, Canada and Brazil in April 2026, with a global rollout to follow.
• Replika: Block via app store and DNS. Replika’s own terms say: “If you are under 18, you are not authorized to use the Services, with or without registering.”

ChatGPT parental controls, toggle by toggle

Linking accounts is the step that matters. On the web, a parent opens Settings from the profile menu, selects Parental controls, then Add family member, chooses Use phone or Use email, enters the teen’s number or address and selects Send. A teen can start it too, from Settings > Parental controls > Invite a parent. Once the invite is accepted, the teen appears under Family members; select the name to change settings. One parent can link several teens, but each teen links to only one parent at a time.

OpenAI’s help centre lists these controls, with their defaults:

• Reduce sensitive content (on): extra safeguards against graphic content, viral challenges, sexual or romantic roleplay, violent roleplay and extreme beauty ideals. A parent can switch it off; the teen cannot.
• Improve the model for everyone (on): whether the teen’s conversations can be used to improve ChatGPT models. Turn it off to opt out of training.
• Reference saved memories (on where available): whether ChatGPT uses saved memories to personalise replies.
• Voice Mode (on) and Image generation (on): switch either off to hide the feature.
• Quiet hours (off): times when the teen cannot use ChatGPT at all. Turn it on, pick a start and end time, and add more than one block if you need to.
• Study mode and Study hours (both off): start new conversations in study mode, always or on a schedule. Neither one blocks ChatGPT.
• Codex network access and Cloud browser (both on): limits for ChatGPT Work, the agent features that run code and browse websites.

Linking does not let a parent read chats. What arrives instead is a safety notification, on by default, sent by email, SMS or push when trained reviewers identify a serious self-harm concern or signs of disordered eating, or when the teen’s account is deactivated over violence. Choose how you are contacted under Safety notifications > Manage notifications. OpenAI says plainly that these are “not real-time monitoring.” Either side can unlink at any time, and the parent is notified if the teen does.

Conversations to have

• “AI chatbots are products. They’re designed to make you want to come back, like a game or a social app.”
• “If you feel like a chatbot understands you better than real people, tell me.”
• “If anything sexual happens in a chat — with an AI or a stranger — you’re never in trouble for telling me.”
• “Don’t give AI your full name, address, school, or photos.”