Patients Want a Human in the Loop

A common narrative holds that patients are ready to hand their mental health care to chatbots and algorithms. A survey published in JAMA Network Open on October 14, 2026, points the other way. Researchers at Columbia University School of Nursing, led by Natalie Benda, PhD, surveyed 1,185 U.S. adults about AI use in secure medical portals. Sixty-three percent said they would not endorse AI-generated medical responses without a clinician's review, and about two-thirds said they would wait up to 48 hours for a doctor or nurse to review and edit an AI answer to a medical question. Preferences depended on the task: 57% favored a fast AI reply for logistical requests such as scheduling and prescription refills.

The survey also asked about mental health support. Roughly 64% said they were likely to use a licensed human therapist even if the wait was three months. Acceptance fell to 50% for a therapist-monitored AI chatbot available within one month and to 44% for an AI-only chatbot available immediately. These figures describe stated preferences, not clinical outcomes, and the mental health questions offered different options than the medical-response questions, so they do not show that mental health was a more cautious category than general medicine.

The findings match what Dr. Alicia Smart has observed in clinical work. Smart has practiced DBT for more than 15 years and has watched the field move toward telehealth platforms and, more recently, AI integration. Some of those changes widened access; others, in her assessment, arrived faster than the evidence supporting them. Clients and colleagues repeatedly raise the same point with her: people want to know that a real person is paying attention.

The public's caution sits alongside early evidence that some AI tools can help. A randomized trial of a generative AI chatbot called Therabot, published in NEJM AI in 2025, enrolled 210 adults with depression, generalized anxiety or high-risk eating-disorder symptoms and reported significant symptom reductions compared with a waitlist control. The authors noted that future work should include an active control and clarify the chatbot's best role as an adjunct to psychotherapy (Heinz et al., NEJM AI 2025).

Why Mental Health Feels Different

The personal stakes of mental health care help explain why patients in the survey drew a line around AI. A person who is struggling emotionally, learning to sit with distress or repair a relationship, is rarely looking only for an accurate answer. That person is looking for someone who understands.

Research on large language models points to specific risks. A 2025 analysis in Healthcare reported that LLMs tested in simulated therapeutic settings frequently showed stigmatizing attitudes toward mental health conditions and responded inappropriately to acute symptoms such as suicidal ideation, psychosis and delusions. Its authors recommended clinician oversight and described AI as a supportive tool rather than a substitute for therapists.

DBT skills fit this picture. Emotion regulation, distress tolerance and mindfulness are practices learned in context, often at moments of real vulnerability, and they are not facts to memorize from a textbook. In Smart's view, a well-designed AI tool can remind someone to use opposite action or walk through a TIPP skill, but it cannot notice the slight change in a voice that tells an experienced group leader a participant is about to shut down. She frames this as an observation about the weight the therapeutic relationship carries, even in a psychoeducational setting.

What Happens in a DBT Group Session?

A DBT skills group teaches across four modules: mindfulness, distress tolerance, emotion regulation and interpersonal effectiveness (Psychotherapy Research, 2021). In standard DBT, group skills training is one of four treatment modes, alongside individual therapy, phone coaching and a therapist consultation team (iDBT-Pain trial protocol). A published pilot gives a typical session format: weekly two-hour meetings led by two co-leaders who follow the DBT Skills Training manual, with a review of the previous week's skills use and homework, a mindfulness practice, instruction in a new skill and a new homework assignment (Veterans pilot study).

The research suggests the format works for reasons beyond the curriculum. In a randomized trial of 44 adults with anxiety or depression and high emotion dysregulation, 16 weeks of DBT skills training outperformed an activities-based support group in reducing emotion dysregulation (d = 1.86) and increasing skills use (Neacsiu et al., 2014). A study of 136 patients in a rolling skills group found that each module was linked to improvements in emotion regulation (Psychotherapy Research, 2021). A three-level meta-analysis of 57 studies found that the alliance patients form with other group members, and with the group as a whole, predicts outcomes in group therapy (Lo Coco et al., 2022).

Smart points to the feedback loop between facilitator and participant as the working part of this structure. A group leader who sees that a concept is not landing can change the explanation, offer a personal example or acknowledge the frustration of trying something unfamiliar. The connection among participants doing the same work is, in her view, something AI alone cannot produce.

Is Online DBT as Effective as In-Person?

The evidence on online DBT skills groups is encouraging and still thin. Telepsychology in general has shown efficacy comparable to face-to-face care in many populations, but a 2021 scoping review of telepsychology in DBT concluded that there was an urgent need to compare standard DBT with online or blended formats.

Most of the available data come from the first year of the pandemic. A study of 143 clients in Australia and New Zealand over six months in 2020 found no significant difference in attendance between face-to-face and telehealth DBT, for either individual therapy or skills training. The authors did not compare clinical outcomes and called for further research (Walton et al., 2023). A review of online DBT skills groups reported greater attendance online and comparable clinical improvement among clients who stayed in therapy, along with challenges in managing risk, therapist preparedness and technology. Providers in the U.S. Department of Veterans Affairs described the skills group as the hardest DBT mode to deliver by telehealth, although most said the experience was better than they had expected (provider perspectives study).

The access implications are significant. Online programs can reach people in rural areas, people with mobility limits and people whose schedules rule out a weekly in-person group. The Walton team described telehealth DBT as a viable option for increasing access, particularly where face-to-face treatment is not available.

Online delivery and AI-driven delivery are separate categories. An online skills group led by a trained facilitator over video differs from a chatbot trained on DBT content, and the 2026 survey offered those options separately, with respondents' acceptance varying across them.

Where AI Can Help, With Guardrails

Smart does not oppose the technology. She founded a psychoeducational platform because she believes DBT skills should be more accessible than traditional therapy models have allowed. In her assessment, AI tools can reasonably handle support functions: sending skill reminders between sessions, helping people track mood patterns, generating practice prompts and offering coping suggestions during low-risk moments of distress. These functions extend the learning that happens in relationship with a trained clinician and do not replace it.

Research teams are testing similar designs. A registered Danish randomized trial (NCT03191565) was designed to compare a DBT smartphone app, offering skills reminders, skills-use tracking and psychoeducation as an adjunct to treatment, against paper diary cards. A published trial protocol describes a six-session internet-delivered DBT skills program for people with chronic pain.

The policy discussion around the 2026 survey points the same direction. In an accompanying commentary, Felix Busch, MD, and Dennis Hedderich argued that safety should set the minimum level of oversight, while patient preference can guide additional review (Medical Economics).

The survey's central finding works as a call for transparency. Anyone weighing an online DBT skills group can reasonably ask who designed the program, whether a clinician reviews the content and facilitation, and, if AI tools are involved, what role they play and where human judgment takes over. These are ordinary consumer questions, and a program worth joining answers them openly.

How Can Someone Tell Whether DBT Group Therapy Is the Right Fit?

DBT group therapy tends to suit people who want concrete coping skills for emotional intensity, relationship conflict, impulsive behavior or chronic feelings of emptiness, and who want both structure and community. Skills training is increasingly delivered as a stand-alone intervention (trial protocol), and the 2014 randomized trial described above found benefit for anxious or depressed adults with high emotion dysregulation who did not have borderline personality disorder (PubMed).

Group and individual formats serve different functions. In Smart's description, individual DBT works on a person's specific life situations and motivation, while the skills group teaches the tools. Comprehensive programs include both, along with phone coaching and a clinician consultation team, though a well-facilitated stand-alone skills group can still produce meaningful change.

Facilitator training deserves attention. The DBT-Linehan Board of Certification is an independent nonprofit that certifies clinicians and programs on their delivery of DBT as it has been researched. Certification involves comprehensive DBT training, consultation and a board-reviewed application (Linehan Institute), and program certification requires a certified clinician as team lead and completion of a fidelity scale (DBT-LBC). Smart treats these credentials as markers that separate clinicians who practice the model with fidelity from those who borrow its vocabulary.

Final Thoughts

The October 2026 survey adds data to an intuition many clinicians already held: patients want people involved in their mental health care, and they say so most clearly when AI enters the picture. Smart's conclusion is that clinicians should neither resist the technology nor defer to it. Her position is that AI should serve the people using it, with transparency, oversight and the human attunement no algorithm has yet matched. For anyone exploring virtual DBT skills training, she advises looking for programs in which real clinicians do the teaching.

{{promo-banner-1}}