General-Purpose Artificial Intelligence Use in Routine Mental Health Practice Among Australian Clinicians: Mixed Methods Study.

Background: General-purpose AI tools are increasingly accessible to mental health professionals, yet little is known about how these off-the-shelf systems are being integrated into routine therapeutic practice. Objective: This study aimed to examine how Australian mental health professionals are using general-purpose AI tools in routine practice, including patterns of use, perceived usefulness, concerns, workplace governance, and factors associated with frequent AI use. Methods: We conducted a sequential mixed methods study with Australian mental health professionals. Semistructured qualitative interviews were conducted with 12 clinicians recruited through the research team's professional networks between June 1, 2025, and August 10, 2025, to explore current uses, perceived benefits, limitations, and governance issues. Interview data were analyzed using deductive qualitative content analysis, guided by predefined domains from the interview guide. Findings from the qualitative phase informed a national survey of 278 respondents. Survey respondents were recruited using convenience sampling through physical posters, university networks, social media, organizations, Primary Health Networks, newsletters, and direct contact with psychology and counseling clinics. Survey recruitment occurred from September 1, 2025, to April 1, 2026. Survey analyses included descriptive statistics, bivariate tests, and multivariable logistic regression models examining daily AI use across client-facing and administrative or clinician-support tasks, perceived performance, concerns of use, workplace permissions, and demographic and professional factors associated with daily use. Results: Interview participants described AI use across client-facing, administrative, documentation, translation, planning, research, and emotional-support tasks. Qualitative findings indicated that clinicians used AI primarily as a clinician-supervised support tool, particularly for saving time, reducing cognitive load, supporting documentation, and improving in-session focus. Participants also raised concerns about privacy, data security, accuracy, client acceptability, overreliance, loss of human judgment, and uneven workplace governance. Survey findings showed that 121 of 278 respondents (43.5%) reported daily administrative or clinician-support AI use, and 92 of 278 respondents (33.1%) reported daily client-facing AI use. Commonly endorsed concerns included privacy and security, accuracy, insufficient training or support, limited confidence in judging outputs, and a preference for human judgment. In adjusted cross-sectional analyses, speaking a language other than English at home was statistically associated with daily administrative or clinician-support AI use and daily client-facing AI use. Age, years practicing, gender, education, and profession were not independently associated with either outcome. Conclusions: General-purpose AI tools appear to be entering routine mental health practice across a broad range of tasks, particularly as clinician-supervised workflow support tools. However, uptake is occurring alongside unresolved concerns about privacy, accuracy, training, human oversight, and workplace governance. Clearer guidance, practical training, and task-specific evaluation are needed to support safe and appropriate AI integration into mental health care.