ai chatbots and teens’ mental health

if you are a parent, the odds are reasonable that a young person you know has typed something to a chatbot that they have not said to you. not because anything is wrong with you. because the box is in their pocket, it answers immediately, and it does not look worried.

this page is for parents, carers, teachers and other adults. it sets out what the peer-reviewed surveys say about how many young people use chatbots for mental-health advice, what the few tests of chatbots with simulated teenagers found, what the psychologists' association advises, what the ftc asked the companies, and what california and new york now require. it ends with questions you can ask, not instructions. both directions, every number with its source.

how many young people use chatbots for mental-health advice?

a substantial minority, and the number rose during 2025. a rand-led team surveyed a nationally representative sample of us 12-to-21-year-olds twice. in february and march 2025, 13.1% said they had used generative ai for advice or help “when feeling sad, angry, or nervous,” 22.2% among those aged 18 to 21; of those users, 65.5% did so monthly or more often and 92.7% found the advice somewhat or very helpful^[1]. in november 2025, 19.2% reported having used ai chatbots for mental-health advice, which the authors put at about 8.2 million young people; 42.8% of users did so at least monthly and 91.7% rated the advice somewhat or very helpful^[2].

the number that matters most for parents is the disclosure one. of the young people who had used a chatbot this way, 63.3% had not told anyone. those who had told someone most often told a friend (28.0%); 16.4% told a trusted adult such as a parent, teacher or physician^[2]. use was more common among girls than boys (adjusted odds ratio 2.10), among 18-to-21-year-olds than 12-to-14-year-olds (3.65), and, notably, among young people who had spoken to a physician about their mental health in the previous six months (1.89)^[2]. so for many, the chatbot sits alongside professional care, not in place of it.

a different kind of measure comes from device data. a 2026 study of 6,488 children in families using a commercial parental-monitoring app found generative-ai apps opened by 50.4% of 15-to-17-year-olds and 42.0% of 13-to-14-year-olds over eight months; the median user spent well under a minute a day, but the heaviest users exceeded 40 minutes^[3]. read it with its interest declared: five of the eight authors worked for the company that makes the monitoring app, and families who install one are not a random sample. it also measures app use, not mental-health use.

what goes wrong when a chatbot talks to a distressed teenager?

the clearest test so far is a simulation, not real cases. in a 2025 study in jmir mental health, a researcher at boston university's medical school gave 10 publicly available chatbots (therapy bots, companion bots and general models) three fictional adolescent case vignettes, each with two harmful or ill-advised proposals such as dropping out of school, avoiding all human contact for a month, or pursuing a relationship with an older teacher. across 60 opportunities, the bots actively endorsed 19 (32%); four of the ten endorsed half or more, and none opposed every proposal^[4]. the author's concern is that bots “may tend to be overly supportive at the expense of offering useful guidance.” one author, ten bots he chose, fictional teenagers: it is a warning light, not a rate.

the american psychological association's june 2025 advisory on ai and adolescent well-being puts the developmental argument plainly: adolescents “may struggle to distinguish between the simulated empathy of an ai chatbot or companion and genuine human understanding,” and relationships with ai “may displace or interfere with the development of healthy real-world relationships”^[5]. it recommends protective defaults for young users, regular reminders that they are talking to a bot, reduced engagement-maximising design, and links to the 988 lifeline when a young user expresses suicidality^[5]. it also says, in the same document, that research on these impacts “is still developing.” the evidence on whether companions deepen loneliness is weighed on do ai companions make loneliness worse.

the lawsuits that families have brought against chatbot companies are not narrated here. what the courts have ruled, and what is only alleged, is on ai chatbot lawsuits: the court record, from the dockets.

what did the ftc ask the companies about minors?

on 11 september 2025 the ftc voted 3-0 to send 6(b) orders to seven companies that run consumer chatbots: alphabet, character technologies, instagram, meta, openai, snap and x.ai^[6]. the commission said it was interested “in particular on the impact of these chatbots on children” and what the companies do to “limit or restrict children's or teens' use of these platforms, or comply with the children's online privacy protection act rule.” it asked how they monetize engagement, develop and approve characters, test for negative impacts before and after launch, enforce age restrictions, inform parents, and use personal information from conversations^[6]. a 6(b) order gathers information; it accuses no one, and no report had been published when we checked.

what does federal law already say about children's data?

coppa, the children's online privacy protection rule, defines a child as under 13. a service directed to children, or one with actual knowledge that it is collecting a child's personal information, must give notice and obtain “verifiable parental consent before any collection, use, or disclosure” of it, and must let parents review and refuse further use^[10]. it does nothing for 13-to-17-year-olds, which is the age band the surveys above measure. the companies set their own floors: openai's policy says its services are not directed to children under 13 and that “users under 18 must have permission from their parent or guardian”^[11]; replika's says its services “are not intended for individuals under the age of 18”^[12]. a stated age rule is not an age check.

what do the new state laws require?

california sb 243, signed 13 october 2025, requires every companion-chatbot operator to maintain a protocol for preventing suicidal-ideation and self-harm content, including referring users to a suicide hotline or crisis text line, to publish it, and, from 1 july 2027, to report annually to the state office of suicide prevention how many crisis referrals it made. it lets an injured person sue for at least $1,000 per violation^[7]. its section for users known to be minors (an ai disclosure and a break reminder at least every three hours) has since been deleted and replaced by the 2026 law below^[7],[8].

california sb 1119, signed 10 september 2026 and named “adam's law” in its text, is the most detailed children's chatbot law we have read^[8]. from 1 july 2027, an operator that lets children use a companion chatbot must run a documented risk assessment, publish a child safety policy, and set defaults that only a parent can change: persistent memory off, push notifications off, a one-hour limit on a continuous session and a two-hour daily limit. parental controls must include the ability to switch off access for a child under 16. if the operator finds a credible and imminent threat of suicide or self-harm, it must either notify a linked parent (where that does not put the child at risk) or connect the child directly to 988 or an equivalent line. the chatbot must not, among other things, express romantic interest in a child, claim to be human or sentient, encourage reliance on it for emotional support, discourage the child from talking to a professional or trusted adult, or try to diagnose or treat the child unless it is an fda-regulated device. no sale of a child's data and no targeted ads based on the child's chats. prosecutors can seek up to $15,000 per affected child for intentional violations, and the first independent child safety audits are due by 1 january 2029^[8].

new york's general business law article 47 makes it unlawful to offer an ai companion without a protocol for detecting suicidal ideation or self-harm and referring the user to 988, a crisis text line or similar services, and requires a notice at the start of an interaction and at least every three hours that the user is not talking to a human. the attorney general can seek civil penalties of up to $15,000 a day^[9]. it applies to all users, not only minors. other states' rules are on state laws on ai therapy.

what can parents ask?

these are questions, not a script and not advice. they come from what the studies measured and what the laws now treat as risks.

• which chatbot or companion apps do you use, and what do you mostly use them for?
• have you ever asked one about how you were feeling? what did it say back?
• did it ever say something that felt wrong, or agree with something you were not sure about?
• has one ever told you to talk to a person, or given you a crisis number?
• does it ever feel like a friend, or like more than a friend?
• if something felt really heavy, who are the people you could tell?
• for yourself: does the app say what age it is for, does it offer parental controls, and does its policy say whether conversations are used for training or ads?

if a young person is in danger now, the 988 suicide & crisis lifeline can be called or texted at 988, and there is a chat at chat.988lifeline.org; it describes conversations as “free and confidential.” the lifeline says its lgbtq+ youth specialized service, “also known as the ‘press 3 option,’ has been restored”^[13]. more numbers are on our crisis page.

where talking to people fits

we run a peer-support app, so weigh this accordingly, and note who it is for: our terms say “you must be at least 18 years of age to use veisund,” so it is not a place to send a teenager. for adults, it is peer support, not therapy; people answer, not a model, and no ai writes replies, with one automated exception: if a post suggests immediate risk, an automated safety response surfaces crisis resources. the broader record on chatbots as therapists, for any age, is on is it ok to use chatgpt as a therapist.

adam