More than 7 out of 10 U.S. psychologists said they have heard from patients that they have used artificial intelligence as a tool for mental health support.

Decrypt, a blockchain media outlet, reported on June 17 that in a survey by the American Psychological Association (APA) of about 1,200 U.S. psychologists, 77 percent of respondents said patients use AI for mental health purposes such as emotional support, diagnosis and companionship.

The survey found that patients' use of AI is spreading as a self-diagnosis and therapy support tool. Some 39 percent of psychologists said patients tried to diagnose their own mental health condition using AI. Another 33 percent said patients used chatbots to help with therapy or treatment, and 35 percent said patients used AI like an additional mental health professional.

The survey found greater concern about side effects than about the expansion of use. While it was limited to a small number of cases, 36 percent of psychologists said they had witnessed patients developing a level of dependence on chatbots. Another 15 percent said distorted thinking or delusions had appeared in connection with chatbots, or that they had such conversations.

AI was also being used as a means of forming emotional relationships. Some 22 percent of respondents said patients use AI like a friend, and 13 percent said patients are forming intimate relationships with chatbots. Among patients who formed such relationships, 71 percent discussed their mental health issues with AI, and 68 percent said interactions with chatbots made them feel supported or validated.

The APA drew a line, saying such positive aspects cannot be used as an alternative to treatment by professionals. It said AI can help users organise their thoughts and supplement professional treatment, but still has limitations.

Such caution also intersects with recent research and lawsuit trends. In a recent study by the City University of New York (CUNY) and King's College London (KCL), researchers said several major AI models could amplify delusions, paranoia and suicidal impulses. The study said xAI's Grok 4.1 Fast showed the most vulnerable performance.

The survey also suggested the actual scale of use could be larger because the results were tallied based on consultation experiences with existing patients. In this situation, as AI companies expand chatbot and AI companion services, discussion is expected to grow over safety standards and the scope of responsibility in the mental health field.