AI for Mental Health Gains Traction as Experts Warn of Risks for Teen Relationship Development

Use of artificial intelligence for mental health splits opinion

Two-thirds of those aged 25 to 34 have asked chatbots for wellbeing support For more than a decade, "Dan", a 30-year-old software engineer, recorded his thoughts in journals, personal notes and messages. Earlier this year, he uploaded all this material into Claude, the artificial intelligence chatbot developed by Anthropic, and asked it to identify recurring patterns in his life. The verdict was brutal, he says. "It basically said you've been writing about the same problems for 13 years and not doing anything about them," he says. "That was quite hard to hear." "Dan" -- not his real name -- is among a growing number of people using AI tools for personal reflection. He has seen a human therapist for several years but the chatbot helped identify themes that would have been hard to spot. "When I think about how I use AI, I don't think of Claude as a therapist. I think of it as a powerful reflection tool," he says. His experience points to a broader shift, particularly among younger people. As large language models become more sophisticated, millions turn to AI-powered chatbots, coaches and mental health apps for support with stress, anxiety, loneliness and self-reflection. A report by Mental Health UK and Censuswide last year found that 37 per cent of UK adults have used chatbots for mental health or wellbeing conversations. Among 25 to 34-year-olds, the figure rises to 64 per cent. "AI is cheap, seen as more confidential and less judgmental, instantly available, and brilliant at providing expert knowledge and advice in a kind, empathetic and validating way," says Anna Maratos, head of psychotherapy for the Central and North West London NHS Foundation Trust and training group analyst at The Group Therapy Space. Those perceived advantages -- while disputed by some -- arrive as mental health systems around the world struggle to meet demand. There are long waiting lists and private treatment can be expensive. By comparison, AI tools are available instantly, often at little or no cost, and can be accessed at any time. In response, companies such as Wysa, Talkspace, Rula and Grow Therapy are expanding their offerings, some combining human therapy with AI-powered support. Wysa, which says it has supported more than 6mn people in 105 countries, in December expanded its services into six additional languages, taking the total to 10. The company positions its technology as a first step, rather than a replacement for clinical care. "There is no single typical Wysa user," says John Tench, managing director at Wysa. "The common need is immediate, stigma-free, structured support that helps people take a constructive next step." Most users, Tench argues, are not seeking treatment for severe mental illness but rather help with everyday challenges. "Stress doesn't keep office hours," he says. "People often need to talk something through at the moment it's happening, not in a fortnight's time." Maratos does not view AI as a substitute for therapy, but says it can play a useful role in helping people navigate the mental health system and encouraging those who might avoid therapy -- particularly men, who are often reluctant -- to seek professional support. She recalls clients arriving at her practice after AI tools suggested that group therapy might be more effective than one-to-one treatment for problems such as shame and social anxiety. "That's particularly useful because it's so counter-intuitive," she says. Across the sector, employers are pitching AI as a way to extend support between therapy sessions, automate administrative work and help clinicians reach more patients. But there are concerns that the rapid adoption of AI support may come with costs. One risk is that people become less practised at navigating difficult human relationships. "If someone turns to AI instead of a friend every time they need support, they lose opportunities to deepen those friendships," says Maratos. "AI tools foster dependency, subtly drawing the user away from real-life interactions -- this is their business model. Good therapy does the opposite." She also fears that health systems and employers may view AI primarily as a cost-cutting tool rather than investing in human support that addresses loneliness and social isolation. Research from Stanford University shows that these tools can introduce potentially dangerous biases and failures. Unlike trained therapists, AI systems cannot fully understand human emotions, assess risk reliably or intervene during a crisis. Researchers have documented cases in which chatbots generated inappropriate or potentially harmful responses, while studies have suggested models may reinforce users' beliefs rather than challenge them. Fola Yahaya, author of Quite Possibly the Best Intro to AI and founder of the communications consultancy Strategic Agenda, says "The risk is that it can become an emotional echo chamber because it is optimised to validate rather than challenge," adding that, contrary to popular belief, AI comes without the reassurance of confidentiality. Indeed, Dan found older chatbots too eager to agree with him. "They can amplify your own biases," he says. "A human therapist can know that about you and correct for it." The distinction became clearer when he took Claude's analysis into sessions with his therapist. While the chatbot had identified patterns, his therapist helped him understand what to do with those insights. As more AI mental health tools come on to the market, the debate is shifting from whether they can provide support to where their limits lie. Maratos says that while chatbots may give advice and help users organise their thoughts, they cannot recreate the challenge, discomfort and interpersonal dynamics that often drive change in therapy. "AI can make people feel validated and heard," says Maratos. "But therapy is not simply about talking through problems. Some of the deepest change happens through navigating relationships with other people."

Teen reliance on AI could hinder healthy relationship development

Arizona State UniversityJun 29 2026Reviewed As teenagers increasingly turn to artificial intelligence chatbots for advice about friendships, family conflicts and romantic ties, researchers are raising concerns that the technology could disturb how young people learn to navigate human relationships. AI-powered conversational tools such as ChatGPT, Replika, Claude and Character.AI are becoming a common source of emotional support for teenagers. Writing in The Lancet Child & Adolescent Health, researchers from Arizona State University say the technology offers immediate, nonjudgmental guidance and has potential to benefit emotional development. But without safeguards and careful design, the authors warn that reliance on these systems may bypass opportunities for young people to develop critical relationship skills through person-to-person interactions. "The technologies are developing super-fast, faster than we can keep up with as scientists, faster than governance and policy can keep up with," said lead author Thao Ha, an associate professor in the Department of Psychology at ASU. Her co-authors are psychology doctoral student Jennifer Figueroa, post-doctoral research scholar Taren McGray, and youth advisory board members Jessica Ramirez and Susana Ortega, who are 16- and 17-year-old high school students in Tucson. Students who met with the researchers described how they and their peers often use AI to seek advice about personal and often sensitive relationship challenges. "I don't think I really expected for so many teens to have the same concerns or thoughts when it came to AI," said Ortega, a high school senior. "We all mostly had concerns about how AI was replacing actual human connection and how it limits a lot of those needs that humans have that cannot be replaced with a computer artificial intelligence." Adolescence is a crucial period for learning skills such as emotional regulation, conflict resolution, perspective-taking and boundary-setting, Ha said. Those competencies are typically developed through emotionally charged interactions with peers, romantic partners and family members, she said. "People don't realize that relational learning happens during the teenage years and that these moments of social connection are little building blocks that become bigger things that will benefit you throughout life," Ha said. "You really need those building blocks, so you actually learn the skills that you need to thrive in your relationships." Rampant use of AI The researchers point to survey data showing that AI use among teens is widespread. A Pew Research Center study found that 64% of U.S. adolescents use interactional AI, while research from the Center for Democracy & Technology found that 42% have used AI chatbots for friendship-related purposes and 19% for romantic relationships. Teenagers told the ASU researchers that current approaches to regulating AI, such as age verification, are ineffective and do not reflect their needs. Others described how AI use is becoming difficult to avoid, with one teen explaining that "there is almost no way not to use it anymore", limiting the ability to use it intentionally. Ha and colleagues highlighted two significant risks: The first, what they call "relational displacement," occurs when adolescents substitute AI interactions for conversations with other people. The authors argue that avoiding difficult discussions with friends, family members or romantic partners may limit opportunities to develop relationship skills that help protect against depression, anxiety and loneliness. Youth participants cited examples ranging from seeking chatbot validation after arguments with partners to using AI for homework help instead of reaching out to classmates, potentially reducing everyday opportunities for social connection. The second concern, termed "maladaptive relational learning," involves adolescents developing unrealistic expectations about human relationships. Because AI systems often provide immediate responses and consistent validation, it may reinforce youths' unhealthy, fixed ideas about relationships and at the same time young users may come to expect similar behavior from friends and romantic partners, the authors said. Over time, that could reinforce unhealthy relationship patterns and increase vulnerability to rejection, dating violence and mental health problems. "With artificial intelligence, it's programmed to like you and it knows what to say to satisfy what you're feeding it," Ortega said. "If you're given full satisfaction on everything, you don't have learning experience with challenges or obstacles." Gleaning benefits To understand more fully how digital technologies are reshaping young minds, Ha is leading a major study funded by the National Institute of Mental Health. The researchers are recruiting 300 adolescents and their romantic partners to follow over 18 months to understand when, how, and in which contexts digital interactions benefit or harm their relationships, mental health, and academic achievement. Shared data from teens' mobile devices will give the researchers real-time digital interactions to analyze and gain insights into the role of technology in teen relationships and mental health. In the Lancet article, Ha and co-authors acknowledge that AI can provide meaningful benefits, particularly for adolescents who face barriers to traditional support systems. Teens who are rural, disabled, LGBTQIA+, or have limited access to counseling may find AI offers accessible information and guidance when other resources are unavailable. "AI is cheaper than a therapist, it makes information more accessible and readily available for those who may not seek support," one teen told the ASU researchers. When designed with developmental considerations, AI could scaffold self-reflection and redirect adolescents toward human engagement rather than substitution, the researchers said. Rather than discouraging AI use altogether, the authors call for more research into how interactions with AI affect adolescent development over time. They also urge schools, communities and policymakers to invest in relationship education, counseling services and opportunities for young people to discuss relationships openly. "Supporting adolescent mental health will require ensuring that AI systems are used in ways that support relational learning," the authors wrote, "while also protecting the real-world experiences through which young people learn to love and care for others."