Nearly one in five U.S. adolescents and young adults have used AI chatbots for mental-health advice, according to a study published in JAMA Pediatrics. The finding lands as state legislators race to impose safety requirements on chatbot platforms, and as separate survey data shows 12 percent of U.S. teens ages 13 to 17 have already turned to chatbots for emotional support or advice. The gap between how fast young people adopt these tools and how slowly guardrails follow them is widening in real time.
Why youth chatbot use for mental health demands attention right now
The speed of adoption is the core problem. Young people are treating AI chatbots as informal counselors before any consistent safety standards exist for those interactions. The JAMA Pediatrics paper on youth chatbot use documents this shift with national-level data on chatbot use and disclosure patterns among people ages 12 to 21. That research confirms the “nearly one in five” figure and raises a pointed question about disclosure: how many of these young users tell a parent, doctor, or school counselor that they are relying on a chatbot for mental-health guidance?
California has responded with SB-243, a bill that would require companion-chatbot platforms to build in suicide and self-harm response protocols, provide crisis referrals, and display clear disclosures when minors are using the service. If signed into law, SB-243 would test a straightforward theory: that requiring platforms to disclose their limitations and route users toward licensed help will reduce the number of young people who silently depend on AI for mental-health support. Whether such a law produces measurable drops in undisclosed queries within 18 months could be tracked through repeated national surveys like the ones already generating data on this topic. But no state has yet produced compliance data or post-implementation survey results, so the theory remains untested.
The tension is not abstract. A teenager experiencing a mental-health crisis who turns to a chatbot instead of a crisis hotline or therapist is making a choice shaped partly by access barriers and partly by the frictionless availability of AI. Without disclosure requirements or crisis-routing protocols, that teenager may never encounter a prompt to seek professional help.
What JAMA Pediatrics and Pew Research data reveal about teen chatbot behavior
Two distinct datasets anchor the public understanding of this trend, and they measure slightly different things. The JAMA Pediatrics study covers adolescents and young adults, a group spanning roughly ages 12 to 21, and focuses specifically on mental-health use and whether users disclosed that behavior to others. Its central finding, that nearly one in five people in this age range used AI chatbots for mental health, captures a broader slice of youth than most prior surveys.
Separately, a Pew Research Center survey found that 12 percent of U.S. teens ages 13 to 17 used chatbots for emotional support or advice. The Pew figure is lower in part because it covers a narrower age band and uses different survey language, asking about “emotional support or advice” rather than “mental health” specifically. The distinction matters: a 16-year-old asking a chatbot how to handle a breakup and a 20-year-old asking one about persistent anxiety are both captured by the JAMA Pediatrics study, but only the younger user falls within Pew’s sample frame.
Together, the two datasets show that chatbot use for emotional and psychological support is not a niche behavior. It is a pattern visible across multiple research methodologies and age definitions. The JAMA Pediatrics paper adds a layer that Pew does not: disclosure. How many young users told anyone, a parent, a clinician, a friend, that they were getting mental-health advice from an AI? The paper’s title signals that disclosure gaps are a central concern, and the implications for clinical care are direct. A therapist who does not know a patient is also consulting a chatbot cannot account for the advice that chatbot has already given.
California’s SB-243 and the limits of current evidence
California’s attempt to regulate this space centers on SB-243, described in the bill text as targeting “companion chatbot” platforms that interact with users in a sustained, conversational way. The proposal would require these services to implement suicide and self-harm response protocols, provide crisis referrals, and add safeguards specifically designed for minors, including age-appropriate disclosures that the chatbot is not a human therapist and cannot replace professional care.
The bill addresses a real gap, but several questions remain open. No public records yet show how many chatbot companies have filed compliance plans or begun building the required protocols. The bill’s practical impact depends entirely on enforcement and on whether platforms treat California’s rules as a ceiling or a floor. If major chatbot providers adopt SB-243’s requirements nationwide to avoid managing state-by-state compliance, the law could have effects well beyond California. If they treat it as a local obligation, its reach will be limited.
The evidence base itself has clear boundaries. The JAMA Pediatrics and Pew surveys are snapshots, not longitudinal studies. They do not yet show whether chatbot use for mental health is rising year over year, nor do they measure clinical outcomes such as symptom improvement, worsening, or delayed access to care. Policymakers are therefore legislating in a space where the risks are plausible and the usage data are solid, but the causal links between chatbot interactions and mental-health outcomes remain uncertain.
Disclosure gaps and clinical blind spots
Disclosure sits at the center of the concern. When an adolescent uses a chatbot to talk through suicidal thoughts or self-harm urges and never mentions that behavior to anyone offline, clinicians and caregivers lose visibility into a critical part of that young person’s coping strategy. Even if the chatbot responds with cautious, supportive language, it cannot coordinate safety planning, monitor for escalating risk, or integrate its guidance with a broader treatment plan.
For clinicians, undisclosed chatbot use can create blind spots. A therapist might see a patient who appears to be stabilizing, unaware that the patient is simultaneously receiving inconsistent or unvetted advice from an AI system. In some cases, the chatbot may echo evidence-based practices; in others, it may offer generic reassurance that conflicts with the therapist’s recommendations or fails to recognize red-flag language. Without disclosure, there is no opportunity to reconcile those messages.
Families face a similar problem. Parents who assume their child would turn to them, a school counselor, or a crisis hotline in moments of distress may not realize that the first point of contact is now a chatbot open in a browser tab. This shift does not automatically mean worse outcomes, but it changes the information flows that safety planning has historically relied on.
Designing safer chatbot experiences for young users
In the absence of long-term outcome data, several design principles emerge from the current evidence and policy debates. First, chatbots that are likely to be used for emotional support should include clear, age-appropriate disclosures that they are not a substitute for professional care, delivered not just at sign-up but woven into conversations that touch on self-harm, suicide, or severe distress.
Second, crisis-routing protocols can be built into the interaction itself. When users mention suicidal thoughts, self-harm, or acute danger, the chatbot can respond with supportive language while also presenting direct links or phone numbers for crisis hotlines and encouraging the user to contact a trusted adult or clinician. SB-243 would make such responses mandatory for covered platforms operating in California; other jurisdictions may follow with similar requirements.
Third, systems can be designed to encourage, though not coerce, disclosure to humans. For example, when a young user repeatedly raises serious mental-health concerns, the chatbot could suggest bringing those concerns to a parent or therapist and offer guidance on how to start that conversation. This approach respects user autonomy while nudging toward offline support networks.
Finally, transparency about data use is essential. Young users may not realize that their most intimate disclosures to a chatbot could be stored, analyzed, or used to refine models. Clear, concise explanations of what happens to conversation data, presented in language accessible to teens, should be part of any responsible deployment.
What policymakers and researchers should watch next
As SB-243 and similar proposals move forward, two types of evidence will be especially important. One is implementation data: how many platforms adopt robust crisis protocols, how consistently those protocols are triggered in real-world use, and whether users report understanding the limitations and risks of chatbot advice. The other is longitudinal research that tracks not just how many young people use chatbots for mental health, but how that use intersects with clinical outcomes, help-seeking behavior, and disclosure to trusted adults.
For now, the main certainty is that AI chatbots have already entered the mental-health ecosystem for adolescents and young adults, whether or not that ecosystem is ready for them. Laws like California’s SB-243, coupled with careful monitoring of usage and disclosure patterns, will help determine whether these tools ultimately function as a bridge to human care or as a quiet, unregulated substitute for it.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
