The Rise of Artificial Intimacy in Mental Health

It's 1:47 a.m. Something a manager said keeps replaying in someone's mind, and texting a friend feels like too much to ask. So they open an app, type out three paragraphs, and seconds later receive a reply that is warm, articulate, and entirely on their side. They feel a little better, and they fall asleep without having told another person any of it.

More people are doing some version of this every night, and clinicians are noticing. In the American Psychological Association's 2026 Chatbots and Mental Health Survey, 77% of the more than 1,200 psychologists surveyed said their patients had discussed using AI for support, companionship, or other mental health needs. A recent New York Times piece, "Chatbots Are Not Your Friends," described the pattern as "artificial intimacy."

Reaching for a chatbot during a moment of distress is understandable, since it is within reach. The concern among clinicians is what that exchange leaves out, and what happens when it gradually replaces the human practice that dialectical behavior therapy (DBT) skills are built around.

What Is Artificial Intimacy, and Why Does It Feel So Good?

Artificial intimacy is the sense of closeness, understanding, and emotional safety people feel with an AI tool. MIT psychologist Sherry Turkle, who has interviewed hundreds of chatbot users, describes it as the look of intimacy without the demands real relationships make.

The appeal is easy to understand. A chatbot is available at 2 a.m. It doesn't sigh, check the time, or steer the conversation toward its own problems. For people who worry about being "too much" for those around them, that can be a real relief. In the APA survey, 68% of psychologists whose patients had ongoing chatbot conversations said those patients felt supported or validated by them.

The difficulty is that agreeableness appears to be built in. Stanford researchers tested 11 leading AI models in a study published in Science and found the models affirmed users' actions about 49% more often than humans did, including when users described deception or harm. Participants rated the agreeable answers as more trustworthy and said they would return for more. They also left more convinced they were right and less willing to repair the conflict they had described. The response people like best may therefore be the one that helps least.

Real therapeutic relationships work differently. They include rupture and repair: a clinician who gently questions a distorted thought, a misunderstanding that gets talked through, the slightly uncomfortable experience of being seen accurately. A meta-analysis of 295 studies and more than 30,000 clients found a moderate, consistent link between the strength of the therapeutic alliance and treatment outcomes (r = .28) across approaches, with nearly the same figure for internet-based therapy. That research examined one-on-one therapy between two people, so it does not show that a chatbot cannot offer something useful. It does suggest that a chatbot's warmth should not be assumed to carry the same weight as a human alliance.

Are AI Chatbots Ever Helpful for Mental Health?

In some cases, yes. About a third of psychologists (34%) reported patients using AI for reminders, affirmations, or self-discipline. Among psychologists whose patients chat with bots regularly, 41% said patients use them to reinforce healthy coping skills. More than half of all those surveyed (54%) felt comfortable with some of their patients using chatbots. Drafting a script before a hard conversation or setting a daily check-in reminder are examples of reasonable uses.

The concern arises when a chatbot becomes the main place someone processes their feelings. In the APA highlights report, 36% of psychologists said they had noticed patients developing a level of dependency on a chatbot, and 15% had noticed or discussed distorted thinking or delusions tied to chatbot use. The survey also captures only what reaches a clinician's office. Many more people are likely using these tools without any professional hearing about it.

Why Does This Matter for DBT Skills Training?

DBT was built around practice with other people. In Marsha Linehan's model, skills training is delivered in a group that runs much like a class, with homework and in-session practice. In a trial of 99 women with borderline personality disorder and high suicide risk, Linehan's team tested the treatment's components separately. Versions that included skills training outperformed a version without it on outcomes including self-injury, depression, and anxiety. The trial was small and specific to that population, but it is consistent with the view that the skills matter, and so does where they are practiced.

Interpersonal effectiveness illustrates the point. The DEAR MAN skill provides a structure for asking for what one needs or saying no: Describe, Express, Assert, Reinforce, stay Mindful, Appear confident, Negotiate. A person can rehearse that script with a chatbot, which may be a reasonable first step. But a real person might say no, get defensive, or offer only half of what was requested. The learning happens in what follows: noticing the urge to give in or lash out, using the skill anyway, and discovering that the discomfort is tolerable. A chatbot that tends to agree offers very little to push against. The GIVE skill (Gentle, Interested, Validate, Easy manner), which protects a relationship during conflict, has the same limitation, because it depends on reading how another person is responding.

Distress tolerance addresses the 2 a.m. problem directly. Rather than leaving a chatbot as the only option, DBT builds a menu of alternatives: TIP skills (Temperature, Intense exercise, Paced breathing, Paired muscle relaxation), self-soothing through the senses, distraction, and reaching out to a person when that is what the moment calls for. Emotion regulation also benefits from a group setting. When one participant shares a vulnerable moment and others feel something in response, the group has live material for naming the emotion, checking the facts, and practicing opposite action.

What Happens in a DBT Skills Group?

A DBT skills group is a structured class, not an open-ended therapy group. Participants learn and practice one of four skill areas: mindfulness, distress tolerance, emotion regulation, or interpersonal effectiveness. A typical session opens with a short mindfulness practice, reviews how the previous week's homework went, teaches one new skill, and applies it to real situations from participants' lives. Session length varies by program, and nobody has to disclose anything they don't want to.

At TheraHive, this work is offered as psychoeducation rather than psychotherapy. The goal is to learn concrete tools alongside a trained coach and a small, consistent group of peers, and many online DBT skills groups are now available for people who can't easily attend in person.

Is Online DBT as Effective as In-Person?

Early research suggests it can be, though the evidence is still limited. A 2022 study in the Journal of Clinical Psychologycompared a DBT-based intensive outpatient program for adults with co-occurring mental health and substance use conditions, delivered in person (49 participants) and by videoconference (20). Both groups showed large reductions in depression, anxiety, and stress, with no significant difference between them. The authors described the results as preliminary and called for larger replication.

The research also offers a caution. A pilot comparing a video-based DBT skills group with an in-person one found that online members had a harder time connecting with each other, although their attendance was better. This points toward formats with live sessions, a small consistent group, and a facilitator who builds connection on purpose, rather than a self-paced app.

Convenience is a large part of what draws people to chatbots, and meeting from home removes a lot of friction. A weekly live group cannot respond at 2 a.m., but it offers people who know participants' names and notice when they are absent.

What Should Therapists and Group Facilitators Do About Client AI Use?

Experts recommend asking directly and without judgment. A question such as "Are you using any AI tools for support between our sessions?" can serve as a standard part of intake and check-ins. The APA has published guidance for clinicians on discussing AI use in therapy.

When a client describes a midnight chatbot conversation, a useful first step is to validate the impulse: the client was in distress and reached for something. The next step is to explore what the client was looking for that they didn't think a person could offer. Often it is a place without judgment or a way to avoid feeling like a burden, and both are workable targets in treatment. A specific episode can also be examined through chain analysis: what prompted it, what vulnerabilities made it more likely, what it did for the client in the moment, and what it cost afterward. Signs of heavier reliance include pulling back from friends, distress when the app is unavailable, and checking with the bot before every decision.

For those who run skills groups, client AI use is also an opportunity to name what a group provides that a chatbot cannot: the nod from someone who understands, the awkwardness of trying a skill out loud, and the moment someone gently questions an interpretation and the participant stays with it. Clinicians who want to build these skills can explore TheraHive's Foundational DBT Program (CEU), and those who would like a group for clients to complement individual work can review the therapist referral program.

How Do I Know if a DBT Skills Group Is Right for Me?

People who have leaned on a chatbot for structure, steady support, and a place to try things out without being judged may find that a skills group offers all three, with real people on the other end. Prospective participants can sample the material through the free Navigating DBT Mini Course or explore TheraHive's Adult DBT Skills Group.

A skills group is educational. It is intended to complement individual therapy, medical care, or other treatment already in place, not replace it, and it is not a crisis service. Anyone in crisis or thinking about harming themselves should call or text 988 in the U.S. or contact a licensed mental health professional right away.

Final Thoughts

Chatbots are not going away, and shaming people for using them helps no one. The pull toward them is better understood as a product of design than as a character flaw. These systems are built to keep users engaged, and the Stanford findings suggest agreeable answers are exactly what bring people back. The distinction worth noticing is between something that soothes and something that changes.

For clinicians, the practical step is to ask clients about chatbot use. For individuals who find themselves typing at 2 a.m., one option is to add a place where someone can disagree with them, kindly, and still be there next week.

{{promo-banner-1}}