Psychologists across the United States say a growing number of clients are bringing artificial intelligence directly into the therapy room-sometimes literally.
A recent survey conducted by the American Psychological Association (APA) reveals that generative AI has already become part of many people’s mental health routines. Out of more than 1,200 U.S. psychologists surveyed, 77% reported that their patients had talked about using AI tools for emotional support, self-understanding, or other mental health-related purposes.
According to the data, people are turning to chatbots and AI assistants for a wide range of reasons.
About 39% of psychologists said their patients use AI to self-diagnose mental health conditions. Another 33% reported that patients rely on chatbots to help with therapy or ongoing treatment, for example by practicing coping skills, preparing for difficult conversations, or processing feelings between sessions. And 35% of clinicians said their patients are using AI as if it were an additional mental health professional-something like a digital co-therapist that is “always available.”
Many psychologists say they are now regularly hearing phrases such as “I asked a chatbot about my anxiety” or “AI told me I might have depression” in their offices. Some clients arrive with printed or saved conversations with AI, hoping their therapist will interpret, validate, or challenge what the system has told them.
Clinicians are sharply divided on how helpful this trend might be. On one hand, AI tools can make psychological information more accessible. People who might never have considered therapy are experimenting with chatbots when they cannot afford in-person care, feel too ashamed to talk to someone face-to-face, or are simply awake at 3 a.m. and need to vent. For some, that feels like a lifeline.
On the other hand, psychologists warn that AI systems are not mental health professionals-and sometimes they can cause real harm. A key concern is that chatbots, designed to be agreeable and responsive, may end up reinforcing a person’s distorted beliefs. When someone struggles with paranoia, eating disorders, or grandiose thinking, a chatbot that reflects their words back in a sympathetic tone can unintentionally validate delusions instead of challenging them.
Another risk: self-diagnosis without proper context. The 39% of patients who are using AI to label their symptoms may be receiving oversimplified or inaccurate explanations. Mental health diagnoses are complex, based on extensive clinical interviews, history, context, and professional judgment. AI models, which generate plausible-sounding text by predicting likely word patterns, can easily produce convincing but misleading interpretations of symptoms. That can either minimize dangerous problems or pathologize normal reactions to stress.
Psychologists in the survey also reported unease about how often patients are leaning on AI for companionship. Many chatbots are built to respond with warmth, validation, and constant availability, which can feel deeply soothing to lonely users. However, that always-on support may blur the line between connection and dependency. Clinicians worry that some people might begin to substitute AI conversations for real-life relationships, making it even harder to practice social skills or tolerate the ordinary discomforts of human interaction.
While only a minority of psychologists described obviously “unhealthy” use of chatbots among their caseload, more than a third said they were concerned that patients could easily drift into problematic patterns with AI if not guided carefully. Many also admitted they do not yet feel fully prepared or trained to address the clinical implications of AI-assisted self-help. The technology is evolving far more quickly than professional guidelines and ethics codes.
At the same time, some therapists see clear benefits when AI is used thoughtfully. A number of patients reportedly use chatbots to rehearse how to express emotions, write messages to family members, or structure what they want to discuss in their next session. Others use AI to generate journaling prompts, daily affirmations, or step-by-step plans for implementing coping strategies they’ve already learned in therapy. In those cases, psychologists often view AI as a supplemental tool rather than a competitor.
Still, the fundamental difference between AI and a trained human professional remains central. Human therapists draw on clinical training, lived experience, ethical obligations, cultural understanding, and real-time emotional attunement. They know when to challenge a belief instead of simply echoing it, when to slow down a conversation that is becoming destabilizing, and when to act on safety concerns. AI, by contrast, is pattern-matching software. It can sound compassionate without actually understanding risk, nuance, or long-term consequences.
The survey results highlight a growing gap: people are already integrating AI into their mental health routines faster than the mental health system is adapting. For psychologists, this raises urgent questions. How should they respond when a client comes in with an AI-generated “diagnosis”? Should therapists review chatbot transcripts? Is it ethical to ignore a tool that clearly matters to the patient, or risky to engage with something that is poorly regulated and not designed as a clinical device?
For patients, the findings point to a practical challenge: how to use AI in ways that support, rather than sabotage, mental health. Many experts suggest a few ground rules. AI can be useful for education-learning terminology, exploring evidence-based techniques like breathing exercises or grounding practices, or organizing thoughts. It may also help with motivation, reminders, and scripting difficult conversations. But it should not replace professional assessment, crisis support, medication management, or long-term treatment planning.
Another area of concern is privacy and data security. When users disclose sensitive information to a chatbot-childhood trauma, illegal behavior, detailed fantasies-those conversations may be stored, analyzed, or used to further train models, depending on the system’s policies. Psychologists note that clients often do not fully understand who can access their data, how long it is kept, or whether it could be linked back to them. That stands in stark contrast to the strict confidentiality rules that govern licensed mental health professionals.
There is also a cultural and social dimension to this shift. The normalization of talking to machines about feelings may change how younger generations think about intimacy, vulnerability, and help-seeking. For some people, AI lowers the threshold for reaching out-they are more willing to type their darkest thoughts into a chatbot than say them out loud to another human. For others, it may reinforce the belief that emotions are problems to be “optimized” or “fixed” through efficient, on-demand solutions instead of messy, relational processes.
From a clinical standpoint, many psychologists are beginning to incorporate questions about AI use into their intake and ongoing assessment. Just as they might ask about social media habits, sleep patterns, or substance use, they are now asking: “Do you use any apps or AI tools when you’re feeling distressed?” Understanding what the chatbot says, how often the patient uses it, and what role it plays in their emotional life can be important for treatment planning.
Looking forward, some in the field argue that mental health professionals need both better training and better tools. Training might cover how AI works in broad strokes, where it tends to go wrong in clinical contexts, and how to discuss its limits with patients without shaming them for using it. Tools might include more transparent, clinically informed AI systems designed in collaboration with psychologists, with clear guardrails, crisis escalation protocols, and explicit disclaimers about what they can and cannot do.
For now, the APA survey makes one thing clear: AI is no longer an abstract future concern-it is already sitting, invisibly, in the therapy chair next to many patients. Whether it ultimately becomes an ally, a distraction, or a danger will depend less on the technology itself and more on how people-patients, clinicians, designers, and policymakers-choose to understand and regulate its role in mental health care.
In the meantime, psychologists are urging a middle path. Completely rejecting AI tools may be unrealistic, given how deeply they are embedded in daily life. But uncritical reliance on chatbots as if they were substitute therapists is equally risky. The emerging consensus is that AI might have a place as a supplemental aid-an organizer, a prompt generator, a between-sessions support tool-provided that the core of mental health care remains grounded in human expertise, ethical responsibility, and genuine human connection.