Clinicians warn AI chatbots are replacing talk therapy without validation or risk safeguards

Key takeaways

A 2025 JAMA Network Open study found 13 percent of minors and 22 percent of adults have used chatbots for mental health advice, even though the tools lack clinical validation, regulatory approval and risk-assessment training. Grow Therapy’s analysis of 350,000 AI coaching conversations with Stanford researchers concluded AI should support rather than replace the clinician, with boundaries, real-time risk monitoring and clinical evaluation built in from the start. Providers say the tools are acceptable for low-risk tasks like journal prompts and session prep, but they cannot substitute for therapy or crisis intervention, and a 2026 study suggests heavy chatbot use may signal distress rather than relieve it.

In a world in which individuals are increasingly glued to their phones, it’s no surprise that loneliness and disconnection are prominent themes in an ever-present mental health crisis. It’s also not surprising that artificial intelligence is a key player in this crisis, as millions of people are willingly engaging with AI wellness applications, even swapping sessions with licensed therapists for chatbots largely due to lower costs and ease of access.

According to a 2025 study in JAMA Network Open, 13 percent of U.S. adolescents and young adults ages 12 to 21 have used chatbots to access mental health advice, and the rate rises to 22 percent among those 18 to 21. This is concerning to the medical community, as these tools are not clinically validated, have, in some cases, impersonated licensed therapists and may cause “AI psychosis,” or amplified delusions and disorganized thinking.

Clinicians are predictably responding to this trend with alarm, as these technologies are not designed to provide clinical feedback or treatment. In fact, according to the American Psychological Association, these apps and chatbots likely lack scientific validation and oversight. They also typically don’t have adequate safety protocols in place, nor have they received regulatory approval.

Grow Therapy, an online and in-person mental health platform, recently collaborated with researchers at Stanford University School of Medicine on an analysis of more than 350,000 AI coaching conversations. The findings showed why safety needs to be built in from the start, with clear boundaries, real-time risk monitoring and ongoing clinical evaluation. The research highlighted how, while AI can be a useful complement to mental healthcare, it should support, rather than replace, the clinician and the broader therapeutic relationship, says Matt Scult, PhD, principal of clinical AI at Grow Therapy.

“As clinicians, we know some of the most meaningful moments in therapy come from reflection, nuance and thoughtful challenge, and being agreeable isn’t always the same as being helpful,” Scult says. “In mental health support, some good friction matters, and the goal shouldn’t be an AI that simply validates everything you say, but one that knows when to reflect, gently challenge and bring a human into the conversation.”

Carolina Estevez, PsyD, a psychologist at Soba, a treatment center in New Jersey, echoes that sentiment, highlighting the potentially dangerous consequences that can come with easy access to AI tools for mental health support and advice. Providers aren’t just concerned about patients receiving bad advice. The bigger worry is a psychologically incorrect recommendation or diagnosis delivered in a way that comes across as confident and trustworthy. AI chatbots are also not trained to assess risk, so they may miss signs of suicidal thoughts or actions, or affirm an individual’s distorted beliefs, she says.

“Psychologists and mental health therapists have been trained to assess risk, read nonverbal cues and interpret what a patient is saying and experiencing using individual context and background knowledge,” she says. “An AI chatbot can give a logical, evidence-based answer, but it can’t account for the details and nuance involved in the majority of situations like a person with lived experience can.”

The harmful effects of using AI for mental health support aren’t limited to adults. According to Johns Hopkins Medicine, an increasing number of children and teenagers are turning to AI-generated programs for advice, companionship and mental health support. A 2026 study in the Journal of Multidisciplinary Healthcare showed that social-supportive chatbot users reported higher loneliness than non-users, suggesting that chatbots may serve as markers of distress rather than solutions. This is concerning because children are even less likely to be able to distinguish good advice from bad, and there have been cases of AI conversations leading to tragic results, such as young people taking their lives.

Like most healthcare professionals, mental health clinicians are aware that they can’t just adopt an anti-AI stance, as the use of AI tools is only likely to grow as access increases. The APA has even put out a guide to help users navigate AI to organize their thoughts and generate ideas while also recognizing that it’s not a substitute for a diagnosis or for treatment of mental health symptoms.

According to the National Academy of Medicine, the general consensus among providers is that AI tools can often direct users to legitimate mental health resources, but they ultimately cannot and should not replace therapy or crisis intervention services. Licensed mental health professionals are also highly unlikely to recommend communicating with AI as a substitute for true talk therapy, even in cases where access or cost is an issue, especially when someone may be going through an acute and dangerous mental health crisis. In such cases, Estevez points patients toward various other available resources, such as local community mental health centers, university training clinics, nonprofit counseling groups, school-based services or practices that offer sliding-scale pricing options.

“AI can be a great tool for low-risk activities like creating journal prompts, understanding basic psychological concepts and tools, and organizing a list of things to talk about in therapy, but it ultimately should not be a replacement for a trained therapist’s clinical judgment,” she says.

Emilia Benton is a Denver-based health and wellness journalist whose work has appeared in publications such as Runner’s World, SELF, and Women’s Health. She is also a 15-time marathoner, USATF Level 1-certified running coach, and an avid traveler.