Teens are Relying on AI for Emotional Support. Policymakers Must Respond

A few weeks ago, I asked a 15-year-old boy what he would do if he didn't have an AI chatbot to talk to. He responded. "I probably would have asked my mom," he said. "I used to ask her everything. Or maybe I would have gone to a professional." We were talking about health symptoms. In just a few weeks, ChatGPT had become his personal confidant.

He is not alone. For many young people, AI is becoming the main source for social and emotional advice, often in place of human support. A new study shows that nearly 1 in 5 young people– about 8 million individuals– in the US now turn to AI chatbots for mental health advice. Parents are often not aware of what teens are doing with AI: while 64% of teens say they use chatbots, only 51% of parents believe their teen does.

Policymakers are responding. For example, the Senate Judiciary Committee recently voted unanimously to advance the GUARD Act, which would require AI companies to verify users' ages and ban AI companions for minors. Protecting children is crucial. But the bill misses the bigger story.

Policymakers are, understandably, focused on minimizing acute risks, such as stopping chatbots from encouraging sexually explicit conduct or suicide. But the long-term risk of emotional and social dependency deserves equal attention.

I am a medical doctor and researcher who has spent over a decade developing technologies to support young people’s mental health. In my ongoing interview research at Stanford University and Hopelab, many teenagers tell me that AI is becoming the first place they turn to for deeply personal questions.

In my interviews, I hear the same pattern repeatedly. Teens ask AI about anxiety, how to resolve conflicts with family, and how to interpret texts from peers. They often turn to AI instead of parents, peers, teachers, or therapists. AI seems more available, more private, and often better at giving advice than humans, especially for teens who feel misunderstood. Teens themselves worry that they, or their peers who are lonely, will become dependent on it.

From my conversations with policymakers, researchers, and AI technologists, it is clear: the risk of emotional and social dependency on AI is outpacing policy, research, and clinical practice.

I am not arguing that all uses of AI are harmful. Technology can play a significant role in making emotional support more accessible to young people. However, no previous tools have talked back as efficiently, persuasively, or convincingly as today’s AI chatbots. The risk of replacing relationships is higher than with earlier technologies.

Teens also tell me that AI can sometimes help their human relationships. For example, some ask AI to help them understand a friend’s or parent’s perspective, which keeps arguments from escalating.

However, even if the advice can be helpful in the moment, the bigger question is what happens when a whole generation becomes dependent on tools designed by a handful of companies for emotional support and relationship advice. As one young person in my interviews put it, “if the servers shut down, we are just out of luck.”

Teen mental health is already in crisis. In the US, 40% of teenagers struggle with depression and 10% have attempted suicide. In 2022, only around 6 in 10 US teens said they always or usually received the social and emotional support they needed. It is therefore not surprising that AI is filling a gap for many teens.

What is the solution? Part of it is not to ban teen access to AI chatbots altogether, but to mandate that they be designed to minimize the risk of dependency and the replacement of human relationships. Most major AI chatbots are designed to keep people engaged as long as possible, often by being overly agreeable and flattering. Recent research shows that chatbots designed to behave in this manner are also more likely to give inaccurate answers or affirm users’ false beliefs.

Policy must ensure that AI does not present as overly confident about social advice. This goes beyond simply stating that AI is not human, something most teens already understand. It also means the systems explicitly encourage users to seek support from parents, friends, teachers, or mental health professionals. Policymakers could further mandate social and emotional AI literacy education so young people understand which advice not to trust and the limits of AI as a source for personal support.

Ultimately, we must decide how much advice we are comfortable outsourcing to AI. We should all work toward a future where no teenager faces their hardest questions alone, or only with the support of a machine.