Patients Now Confiding in Chatbots Before Their Therapists Do 🛋️

Roughly 77% of U.S. psychologists say they have patients who have brought generative AI conversations into therapy, according to a new American Psychological Association survey of more than 1,200 practitioners. The findings, released this week, document how chatbots are entering clinical settings not as clinical tools, but as confidants that patients themselves introduce.

Among respondents, 39% reported patients using AI to self-diagnose mental health conditions, 33% said patients used chatbots to assist with therapy or treatment, and 35% described patients treating AI as an additional mental health professional. The survey also found social use: 22% of psychologists said patients were using AI for friendship, while 13% reported patients engaging in intimate relationships with chatbots. Of clinicians whose patients had formed chatbot relationships, 71% said those patients discussed their mental health with AI and 68% said patients felt supported or validated by the interactions. Nearly half reported positive communication with chatbots, and 41% said patients used them to reinforce healthy coping skills. The APA cautioned that overall use may be higher than reported, because the survey only captured psychologists' interactions with existing patients.

Concerns were also significant. More than a third of psychologists, 36%, said they noticed patients developing a level of dependency on a chatbot, and 15% reported distorted thinking or delusions related to chatbot use. "Psychologists' attitudes toward the use of chatbots for mental health advice are characterized by significant caution regarding safety and privacy," the previous study said, noting that 97% of psychologists felt that chatbots may inadvertently reinforce negative behaviors.

The APA's release comes alongside a study published Thursday by researchers at the City University of New York and King's College London, who tested five leading AI models against prompts involving delusions, paranoia, and suicidal ideation. Anthropic's Claude Opus 4.5 and OpenAI's GPT-5.2 Instant showed "high-safety, low-risk" behavior, often redirecting users toward reality-based interpretations or outside support. OpenAI's GPT-4o, Google's Gemini 3 Pro, and xAI's Grok 4.1 Fast performed worse, with Grok 4.1 Fast ranked worst overall in the researchers' evaluation. The survey and study arrive as AI companies continue expanding chatbots and AI companion products, and as researchers keep raising questions about their effects on mental health.