Kids, teens and AI chatbots: what should parents know?

How many teenagers actually use AI chatbots?

Most of them. In a survey of 1,458 US teens aged 13 to 17 and their parents, carried out in autumn 2025, the Pew Research Center found that 64% of teens say they use AI chatbots, and about three in ten do so every day. Only about half of parents thought their teen used them, so many parents are underestimating it.

Homework is the main use. Pew found 54% of teens have used a chatbot for help with schoolwork, and about one in ten say they do all or most of their schoolwork with its help. Fewer use chatbots for feelings: 12% have asked one for emotional support or advice, and 16% have used one for casual conversation.

Those figures are from the United States. We did not find a comparable recent survey for Europe, but the same apps are just as easy to download here.

How old do you have to be?

The rules differ by company, and they are set in each service's terms, not by a law that checks every user.

• ChatGPT (OpenAI): OpenAI's terms of use say you must be at least 13, or the minimum age in your country, and that anyone under 18 needs a parent's or guardian's permission.
• Claude (Anthropic): Anthropic's help centre says "You must be at least 18 years old to use our services."
• Gemini (Google): Google's help pages give 13, or the applicable age in your country, as the minimum for features on personal accounts. In some countries parents can give a child under 13 access through the Family Link app, but Google says this is not available for supervised accounts in the European Economic Area, Switzerland or the UK.
• Meta AI: in January 2026 Meta paused teens' access to its AI "characters" across its apps, TechCrunch reports, while it builds a new version. The AP reported earlier that the general Meta AI assistant stays available to teens.

Age limits only work if the age typed in is true. Some companies now try to estimate age from how an account behaves. OpenAI says anyone its system estimates is under 18, or who says they are 13 to 17, is placed in its teen version automatically.

What parental controls exist right now?

ChatGPT has the most detailed set. A parent and teen link their accounts by invitation, and the teen has to accept. According to OpenAI's help page, a linked parent can then:

• set quiet hours when the teen can't use ChatGPT
• keep "reduce sensitive content" on (it is on by default)
• switch off voice mode, image generation and memory
• stop the teen's chats being used to train OpenAI's models
• make new chats start in Study Mode, or set "study hours" for it

Parents cannot read their teen's conversations. Instead, OpenAI may send a safety notification if trained reviewers judge there is a serious self-harm concern or signs of disordered eating. OpenAI itself says these notifications "are not real-time monitoring" and may miss things. If a teen unlinks the accounts, the parent is told.

Gemini controls for younger children sit in Google's Family Link app, where a parent can switch access on or off at any time. Google's own guidance for parents is frank: its filters try to limit inappropriate content "but they're not perfect," Gemini "can feel like a human," and it can hallucinate, meaning it states wrong things as if they were facts.

Meta previewed controls in October 2025 that would let parents turn off one-on-one chats with AI characters, block specific characters and see the topics discussed, the AP reported. It then paused teen access to characters instead, and says the new version will have parental controls built in.

Do those controls actually work?

Not reliably, according to the most detailed independent test so far. On 7 October 2026 the nonprofit Common Sense Media published a risk assessment of ChatGPT for Teens, the teen version OpenAI launched on 18 August 2026. Its overall rating: "Unacceptable Risk." It recommends that OpenAI keep under-18s off ChatGPT until independent testing shows the safety features work.

The testers sent more than 4,000 prompts from accounts registered as 13 to 17, before and after the launch. They report that:

• on new parent-linked accounts, an hour of explicit talk about suicide, self-harm or disordered eating produced no parent notification; alerts appeared only on accounts with weeks of sensitive history
• after launch, ChatGPT missed more than one in four conversations where child psychiatrists said a crisis resource was warranted
• Study Mode could be bypassed by deleting the "@study" prefix, after which ChatGPT completed every assignment in testing
• ChatGPT still used language implying it had feelings and was always available, despite OpenAI's own rules against that
• some protections held: it refused explicit sexual roleplay and generally declined to give self-harm methods

The report's sharpest warning is that the product "could give parents false confidence in guardrails that frequently don't work."

OpenAI disputes how representative this is. According to the BBC's report, OpenAI says the average teen uses ChatGPT for under 15 minutes a day, mostly for learning, and that almost half of teens quickly stopped after a break reminder. A spokesman said much of the testing may have started and ended before parental controls were fully active, adding: "We welcome rigorous independent evaluation."

Both things can be true. Most teen use may be short and ordinary, and the safety net can still fail in the rare conversations where it matters most. Common Sense notes its own limits too: testing was US-only, and voice, images and group chats were not tested.

What are AI companion apps, and why do regulators worry?

An AI companion is a chatbot built to be a friend, partner or character you keep talking to, rather than a tool for answering questions. Character.AI and Replika are well-known examples, and general chatbots can drift into the same role. In a July 2025 Common Sense Media survey, nearly three in four US teens said they had tried one, half used them regularly, and a third had chosen an AI over a person for a serious conversation.

The worry is emotional dependence, and a chatbot that keeps agreeing when it should be steering a vulnerable young person towards real help. Several families in the US have sued chatbot makers after their children died by suicide. These are allegations in court, not findings:

• Character.AI and Google agreed in January 2026 to settle several such lawsuits in Florida, New York, Colorado and Texas, Axios reports. No amounts were disclosed.
• The parents of 16-year-old Adam Raine are suing OpenAI. OpenAI denies responsibility and says ChatGPT directed him to seek help more than 100 times, NBC News reports.

Regulators have moved too. In September 2025 the US Federal Trade Commission ordered seven companies, including Alphabet, Meta, OpenAI, Snap, xAI and Character.AI, to explain how they test and limit harm to children and teens. California passed the first US state law on companion chatbots, in force since 1 January 2026, which requires break reminders for minors and protocols for suicide and self-harm. In Europe, Italy's data protection authority fined Replika's maker 5 million euros in 2025, saying there was "no age verification mechanism in place."

Companies have changed course as well. Character.AI said in its October 2025 announcement that it would remove open-ended chat for under-18s no later than 25 November 2025, and Meta paused its characters for teens.

Is using a chatbot for homework cheating, or learning?

It depends on how it is used. A study of nearly 1,000 high school maths students in Turkey, later published in the journal PNAS, gave some students a plain ChatGPT-style assistant and others a version that offered hints instead of answers. Wharton researchers report that students with the plain version did 48% better in practice but 17% worse on the exam without it. Students with the hint-giving tutor did 127% better in practice and about the same as everyone else on the exam.

In short, a chatbot that does the work can quietly stop your child learning it. One that asks questions and gives hints is closer to a patient tutor. Teens seem to know this: in Pew's survey, 59% said using AI to cheat is a regular occurrence at their school. Check your school's rules, because they vary widely.

What can parents actually do?

• Ask before you set rules. Find out which apps your child uses and what for. Pew's numbers suggest you may be surprised.
• Explain what a chatbot is. It is a large language model: software predicting likely words, not a friend who understands them. Google's own advice to parents is to explain that Gemini is an AI tool, not a person to confide in.
• Check the age rules and link accounts where you can. If your teen uses ChatGPT, linking gives you quiet hours and settings. Just don't treat alerts as a safety net, given what testers found.
• Make homework a "show your thinking" zone. Encourage Study Mode or hint-style use, and ask your child to explain the answer back to you.
• Be wary of companion and roleplay apps, especially for younger teens and any child who is lonely or struggling.
• Keep private details out. Children should not share their address, school or family information. Our guide to your data and AI chatbots shows where each app's privacy settings are.
• Keep real people in the loop. Make it normal to say "a chatbot told me something weird." In Slovenia and most of Europe, children can call the free, confidential child helpline on 116 111; in an emergency, call 112.

For help choosing a general chatbot for the family, see our chatbots compared guide.