Community members (CMs) and MHPs were invited to complete one of two web-based surveys. The CM survey was advertised to the general population of people aged 16 years and older who reside in Australia. The MHP survey was advertised to MHPs who reside in Australia. The survey was advertised on social media platforms including LinkedIn, Instagram, and Facebook using a snowballing method for 8 weeks between mid-February and mid-April 2024.

The web-based survey was administered using Qualtrics XM (Qualtrics). After accessing the survey link, interested potential participants were screened for eligibility (aged older than 16 years and residing in Australia).

The survey included questions regarding the following topics:

The survey is available from the authors upon request.

Quantitative data were analyzed using descriptive statistics in SPSS (version 22.0; IBM Corp). The sentiment and thematic elements in free-text responses within free-text responses were manually sorted by classifying responses into positive, negative, or neutral categories based on specific keywords and context. Each response was read in its entirety, and sentiments were categorized as positive, negative, or neutral based on the presence of specific keywords and the overall context of the response. Positive sentiments were identified by words such as “hope,” “benefit,” and “optimistic,” among others. Negative sentiments were indicated by terms like “concern,” “risk,” and “fear.” Responses lacking clear sentiment indicators or expressing ambiguous feelings were classified as neutral. Concurrently, thematic analysis [26] was performed to identify and interpret patterns within the data. This involved an iterative process of reading and coding the data, generating initial codes, and collating codes into potential themes. Themes were reviewed and refined through discussions among the research team to ensure they accurately represented the data set.

Ethical approval for the use of the data was obtained from the University of Melbourne Human Research Ethics Committee (reference 2024-27805-48669-4). This study complied with the Declaration of Helsinki.

The final sample consisted of 107 CMs and 86 MHPs. Demographic characteristics of both samples are presented in Table 1. The mean age of both groups was equivalent. The majority of CMs were employed or studying. MHPs tended to have higher levels of formal education than CMs. For MHPs, the majority were either clinical or general psychologists (21/86, 40%). For CMs, the majority had a previous mental health diagnosis or significant difficulties with their mental health or emotional well-being (76/108, 70%), and the majority had also seen a professional for these difficulties (74/108, 69%). The mean K10 score was 22.8 (SD 8.9), indicating mild levels of psychological distress.

In terms of comfort with using digital technology, 79% (68/86) of MHPs and 82% (89/108) of CMs rated themselves as being very comfortable, somewhat comfortable, or comfortable, whereas 22% (19/86) of MHPs and 18% (19/108) of CMs described themselves as being somewhat or very uncomfortable. Table 2 shows responses to the AI attitudes scale. CMs had neutral attitudes and MHPs tended to have more positive attitudes toward AI across all measured dimensions. Tables 3 and 4 show that MHPs also tend to be more interested in using AI and more are more likely to use it in the future for work purposes than CMs are to use AI to manage emotional and mental well-being.

Table 3 shows that CMs tended to rate their likelihood of using AI for a range of tasks associated with managing their emotional and mental well-being midway between unlikely and likely. The use cases that were most to least popular were (1) providing personal recommendations, (2) mood tracking, (3) detecting early warning signs, (4) use of therapeutic chatbots, and (5) crisis or suicide prevention support. MHPs on the other hand rated themselves as more likely to use AI in use cases across the board. The use cases that were most to least popular were (1) administrative tasks support, (2) synthesizing the latest clinical evidence, (3) training and simulation, (4) tracking consumer progress, (5) personalized recommendations for clients, (6) assisting with assessment and diagnosis, and (7) enhancing consumer engagement with treatment.

In total, 30 CMs (30/108, 28%) and 37 MHPs (37/86, 43%) reported use of AI in the previous 6 months. Of those, ChatGPT was the most common AI tool used by both CMs (16/30, 52%) and MHPs (20/37, 54%). Table S1 in Multimedia Appendix 1 outlines the reasons respondents provided for using these tools, as well as their experienced benefits, harms, and concerns. Further, 77% (23/30) of CMs and 92% (34/37) of MHPs reported AI to be very beneficial, somewhat beneficial, or beneficial, whereas 10% (3/30) of CMs and 3% (1/37) of MHPs found AI to be very harmful, somewhat harmful, or harmful. CMs mainly used these tools to obtain quick advice when emotionally distressed (18/30, 60%) or as a personal therapist or coach they could converse with to help manage their emotional and mental health (14/30, 47%). The most reported benefits were their availability (20/30, 67%), their low cost compared to therapy (18/30, 60%), and their privacy (16/30, 53%). About half of CMs (16/30, 53%) reported that they did not experience harms or concerns. The rest reported a range of concerns, such as responses being too general or not personalized (11/30, 37%) being unsure where their data was going (11/30, 37%). MHPs primarily used these tools to research mental health topics (24/37, 65%) and to assist with report and letter writing (20/37, 54%). Most reported it being helpful (25/37, 68%) and time saving (25/37, 68%). No harms or concerns were experienced by 49% (18/37); however, the rest reported concerns such as the outputs being too general (12/37, 32%), outputs being inaccurate (10/37, 27%), and being uncertain about the ethics of using these tools for these professional purposes (9/37, 24%).

Respondents were invited to share any concerns or interests they had regarding the use of AI for their specific purposes. A total of 66 responses were received from CMs, and 50 responses were received from MHPs. Among CMs, sentiment was rated as positive in 13 (20%) comments, negative in 17 (26%) comments, and neutral in 38 (58%) comments. Of those with positive sentiment, most were excited about AI making mental health care more accessible and efficient, more personalized, and better integrated with other technologies. The content of their negative sentiment was the lack of human support, errors and misdiagnosis, and ethical or data privacy concerns.

For MHPs of the 50 responses sentiment was rated as positive in 12 (24%) comments, negative in 13 (26%) comments, and neutral in 25 (50%) comments. Positive sentiment themes involved increase in efficiency and therefore increased accessibility of mental health care, as well as advanced diagnostics and treatment outcomes. Negative sentiment comments involved concerns about data governance and security, misuse by clinicians, and regulatory challenges (Tables 5 and 6).

This study is to our knowledge the first to survey both CMs and MHPs on their patterns of use, experiences, and perceived benefits and harms associated with the application of AI technologies in mental health care. This analysis provides a critical insight into how AI is currently being used to support mental health care from the perspective of CMs and MHPs, which may inform technological development and guide ethical, professional, and policy initiatives.

Attitudes to AI between the groups varied. CMs scored similarly to published community norms on the AI Attitudes Scale [25] (full scale average score 5.63, SD 2.5 vs 5.54, SD 1.78), while MHPs scored significantly higher (full scale average score 6.74, SD 2.3 vs 5.54, SD 1.78). AI use cases for CMs also had lower levels of endorsement than AI use cases for MHPs. Of note, all CM use cases involved scenarios where AI would be used directly to support personal mental health, whereas MHP use cases were split between indirect or administrative professional tasks and direct client mental health support tasks, the former being more likely to be endorsed. Potentially this is due to direct client use cases conceivably carrying more risk, making CMs and MHPs alike wary of using AI in this way. The intended purpose of commercial AI tools is also more aligned with professional support functionalities than direct mental health care applications. The difference highlights a significant area for future development and the necessity of balancing technological advancement with the training and education of both MHPs and CMs in safe use. The difference in use case endorsements also demonstrates that MHPs and CMs experience different pain points in their day-to-day lives. In the challenging context of embedding new technologies into mental health practice, equal consideration should be given to how AI technology can address these pain points for both groups.

Regarding actual use, we found that AI tools, most commonly ChatGPT, were used by around a third of CMs and 40% of MHPs. CMs tended to use these tools to obtain quick mental health advice or to receive emotional support, and nearly half used them as a personal coach or therapist, reporting the benefits being accessibility, privacy, and low-cost. These tools and associated AI techniques have been recognized for their potential to make mental health care more accessible, accurate, and efficient [14,27]. It is important to note, however, that these commercially available AI tools are not intended for such purposes, and as such, may present predictable and unpredictable risks [27]. About half reported experiencing harms or concerns as a result of use, noting that responses were too general, nonpersonalized, inaccurate, or unhelpful. Further, a lack of clarity regarding data security and the ethics of using AI tools in this way was also reported. These kinds of issues can create harm in some cases. For example, Tessa, an integrated rule-based and large language model AI chatbot designed to support patients with eating disorders [28] had to be withdrawn when it started to provide weight loss advice that ran counter to eating disorder clinical guidelines, sparking calls for greater regulatory measures for the safety of these tools in those and other contexts [29]. Survey respondents also expressed concerns regarding the lack of human support and potential for misdiagnosis, which reflects a need for cautious and informed integration of AI into mental health practices, particularly where AI is used to predict illness or risk states [14].

MHPs report using AI tools to support research, administrative, and training tasks, with a substantial number reporting time-saving benefits. Nonetheless, approximately a third of MHPs indicated concerns about the generality and potential inaccuracy of AI outputs, which emphasizes the need for ongoing scrutiny of the quality and application of AI in diverse settings for a wide range of purposes [27,30]. These findings align with the broader discourse in general health care delivery on the integration of AI into care, where efficiency and productivity gains must be balanced with accuracy, reliability, and ethical considerations [20,31].

The expressed concerns by all respondents regarding data governance, security, and the ethical implications of using AI tools in mental health care were notable. As AI technologies continue to advance, it is paramount that data security and ethical use are prioritized to protect both consumers and professionals, and to maintain trust in these tools [20]. As outlined by Luxton [32] a decade ago, psychologists and other mental health care professionals have an essential part to play in the development, evaluation, and ethical use of AI technologies. In a field still grappling to retrofit regulation for non-AI digital health tools, the rapid development of AI health technology and the associated avalanche of personal health data, “blackbox” processing, and data sharing requires swift action to put the necessary safeguard structures in place.

This study has some limitations. First, the online recruitment strategy may have attracted more respondents familiar with technology, although approximately 1 in 5 reported some discomfort with technology use. Second, the relatively small sample size and recruitment method means that the results may not be fully representative of the broader population and therefore limit the generalizability of the findings. This limitation is expected to be addressed in subsequent similar surveys, which will track the acceptability and perceived concerns and issues of using AI in mental health care, over time. Third, the reporting of benefits and harms may be an underestimate as use may have hidden or time-delayed effects. Nevertheless, the findings provide a useful insight into how AI is both currently perceived and experienced by users. Future research could build on these preliminary findings with larger and more diverse samples, potentially through cross-jurisdictional studies that can provide a more comprehensive view of the impact of AI on mental health care.

Our study underscores the promise and challenges of AI in mental health care. As AI tools evolve, it is essential that they are developed with ethics, inclusivity, accuracy, safety and the genuine needs of end users in mind. This will not only guide technological advancement but also ensure that AI serves as a valuable complement to overwhelmed traditional mental health services, ultimately improving outcomes and efficiencies for all stakeholders involved.