We conducted a scoping review to identify the available evidence regarding AI chatbot interventions for health professionals’ psychological health. A scoping review maps the relevant literature in the field and is useful for identifying gaps in knowledge [15]. This review was conducted in accordance with the scoping review framework of Arksey and O’Malley. The study adopted the following five steps: (1) identifying the research question; (2) identifying relevant studies; (3) study selection; (4) charting the data; and (5) collating, summarizing, and reporting the results. The five stages provided a structured approach to assist in understanding the current evidence by examining the literature. There was no rating of quality performed on the included studies, as scoping studies do not require a quality evaluation [16]. Therefore, a scoping review is suitable for providing a broader perspective on the current research and mapping the relevant literature on AI chatbot interventions for health professionals’ psychological health.

The research question was, “What is the status of the current body of knowledge regarding AI chatbots for the psychological health of health professionals?” According to Arksey and O’Malley, the search strategy involves identifying relevant studies, including published and unpublished works, with an emphasis on being as comprehensive as possible [15]. A literature search was performed between May and August 2024 from the following eight databases: seven global search engines in English (Cochrane Library, CINAHL, Embase, ProQuest Dissertations & Theses Global, PsyclNFO, PubMed, and Web of Science) and one Korean search engine in Korean (Research Information Sharing Service). Various combinations of search terms (Textbox 1) were used to either broaden or narrow the search, depending on the results in a specific database. The reference lists of the retrieved studies were searched manually. Finally, gray literature, relevant letters, and editorials were searched.

The data are presented in tables, summarized, and synthesized qualitatively according to the purpose and research questions of the review.

The final stage of a scoping review involves collating, summarizing, and reporting the results [15]. Data extraction was performed by two researchers (GB and JH) and checked for accuracy by all the researchers. Uncertainties and disagreements were resolved through discussions. The data were extracted using a common data extraction format.

Table 1 presents the characteristics of the analyzed studies. Ten studies published between 2018 and 2024 were retrieved. The studies were published worldwide: in Canada [7,18], Japan [19], Malawi [20], Netherlands [21], Singapore [13,22], Spain [23], and the United States [24,25]. Four studies examined diverse health professionals, including doctors, nurses, nurse assistants, clinical officers, pharmacists, lab technicians, and other allied health professionals [20,21,23,26]. Two studies specifically targeted doctors and nursing staff [13,27] such as nurse practitioners, registered nurses, licensed practical nurses, and nursing assistants, while one study exclusively covered chatbots for nurses [19]. Three studies identified their participants as health professionals, without specifying the specific types [7,18,22]. Of the ten studies, eight studies focused solely on health professionals [13,19-23,26,27]. Two studies included both health professionals and their families [7,18].

The study designs were diverse. Of the 10 studies, three were development studies [18,23,27], one was a design verification study [26], one was a pilot study [22], one was a randomized controlled trial [20], one was a mixed-method feasibility study [21], one was an intervention study [19], one was a cross-sectional study protocol study [7], and one was a qualitative study [13]. Three studies evaluated the development and feasibility of AI chatbots for psychological health support [23,26,27].

Table 2 shows the characteristics of the AI chatbots developed or implemented in the retrieved studies. The types of AI chatbot algorithms were diverse; NLP-based AI was the most commonly used algorithm, appearing in six studies [7,13,18,19,21,23]; rule-based AI, an algorithmic method that does not learn from data or create rules on its own, was used in two studies [20,22]; and learning-based AI that learns from data was adopted in two studies [26,27]. AI chatbots were delivered through mobile apps in six studies [13,19,20,22,26,27], and web-based platforms via web links in four studies [7,18,21,23]. The interaction type in all studies was 1:1 interaction: Among them, one study included 1:1 interaction alongside group chat functionality [26], and another study included 1:1 interaction with optional human coach support [22]. In seven studies, the chatbot actively initiated the conversation [7,13,18-20,23,27], while in three studies, the conversation was passively initiated [21,22,26].

Participant completion of the AI chatbot interventions was reported in four studies [20,21,23,27]; among these, three studies provided the percentage of participants who completed the intervention: 14.3% [21], 67.7% [20], and 73% [27] while one study reported the employee access rate of the intervention as 10.88% [27]. Of the 10 studies, six evaluated the AI chatbot’s usability [13,20,22,23,26,27]. Four studies relied on participant feedback: 54.6% rated it as useful [26], the average satisfaction score was 4.07 out of five [22], more than 80% reported it as helpful [23], and 91% of the users found the chatbot easy to use [20]. One study used user engagement indicators to identify high subjective usability (acceptability, usability, and satisfaction) but low objective engagement (completion, adherence, compliance, and engagement) [27]. Another study assessed chatbot performance metrics, reporting an intent detection accuracy of 99.1% and an entity extraction accuracy of 95.4% for the Mira Chatbot [18]. Chatbot usability was not evaluated in the remaining four studies [7,13,19,21].

Table 3 lists the contents and outcomes of the AI chatbot interventions. The primary component is the screening function for mental health conditions such as anxiety, depression, and stress, which was included in seven studies [19-23,26,27]. Four studies included cognitive behavioral therapy techniques [13,20,22,27]. Additionally, four studies implemented stress management techniques, such as breathing exercises and meditation [13,19,20,22]. Three studies incorporated problem-solving strategies [7,18,22] and psychoeducation [19,22,27]. Two studies used behavioral activation [20,22], cognitive restructuring [21,22], and gratitude training [20,22]. One study featured storytelling [21]. In addition, there was a function to support portal sites connected to community resources in three studies [7,18,23].

Across the studies, a variety of additional features were incorporated. Seven studies included AI chatbot functionalities for crisis management and support [7,13,18,20,22,27]. Specific features encompassed emergency contact information for crises [7,18,20,27], peer support groups and in-app forums for social support [13], expert support through in-app counseling, and options for connecting with human coaches [13,22,23]. Six studies incorporated reminders to encourage chatbot use [13,18,19,22,26,27]. Gamification elements, such as point-based rewards to encourage participation, were featured in four studies [13,19-21]. Peer support functionalities were available in three studies [13,23,26]. Two studies included mood-tracking features for health and lifestyle management [13,20], while another two studies provided an audio recording option [22,27]. Additionally, one study integrated devices for tracking vital signs [26] and provided an immersive virtual reality experience [19]. A total of 4 out of 10 studies explicitly stated the duration of chatbot use or intervention dose ([19]: 4 wk; [20]: 8 wk; [21]: 2 wk; [23]: 1 mo) while it was not specified in six studies. The dose varied among users in two studies [22,23].

The research in four studies evaluated psychological health outcomes [19-21,27], whereas, in six studies, the focus was primarily on the development and description of user perceptions of the AI chatbot without providing specific outcome measures [7,13,18,22,23,26]. The most commonly measured psychological health outcomes were anxiety and depression, which were examined in three studies [19,20,27]. Burnout was investigated in two studies [20,27]. Compassion fatigue [21], loneliness [20], resilience [20], and self-compassion [21] were also evaluated. Anxiety was measured in three studies: one reported a significant reduction of scores [20] while the other two reported no significant differences in anxiety scores [19,27]. Burnout was assessed in two studies, and each reported a significant decrease in its scores [20,27]. Depressive symptoms were measured in two studies: one reported a significant reduction of scores [20] while the other did not [27]. Other psychological variables were measured in one study each; of those, the variables that showed significant change in scores were health counseling needs [19] and resilience [20].

The studies retrieved for this scoping review were published during the past seven years, between 2018 and 2024, and reflect the rapidly evolving trend of AI chatbot technology for supporting psychological health. The global distribution of studies across countries indicates a widespread interest in using AI for psychological health support and health professionals’ psychological health.

Our review revealed that chatbots were used by a broad spectrum of health professionals, highlighting their diverse applications. This diversity underscores the potential for universally designed chatbots to effectively address psychological health needs across different professional settings. However, prior research has shown that nurses, who have direct and continuous patient contact, experience greater mental health distress compared to other health professionals [24]. Therefore, future studies should consider incorporating role-specific options into chatbot designs to create tailored interventions for diverse professional groups.

The diverse research methods used in the retrieved studies signify the nascent stage of research on AI chatbots for psychological health support, emphasizing technology refinement and user experience comprehension. Such a variety of research approaches is crucial for validating the efficacy and safety of AI chatbot technology, enabling iterative improvements and a deeper understanding of how health professionals interact with AI chatbots [25]. However, it is noteworthy that only 4 out of 10 studies measured psychological health outcomes. This limitation underscores the significant gap between technological developments and clinical validation. As the field matures, it becomes imperative to quantitatively assess the impact of AI chatbots on psychological health indicators. Further studies measuring outcome variables to demonstrate the clinical effectiveness of AI chatbots are needed.

The studies reviewed here demonstrate the variety of AI algorithms used in chatbot design, including NLP-based-, rule-, and learning-based approaches. NLP-based AI chatbots which provide personalized responses by analyzing user inputs [28] are the most commonly used AI algorithms. Evidence has shown that an NLP-based AI chatbot, which simulates natural conversations, is successful at delivering tailored interventions [29]. Rule-based AI algorithms, adopted in 2 studies out of the 10 retrieved studies, are effective in providing fixed and predetermined responses; however, they lack the adaptability required to provide more dynamic and engaging psychological interventions [30]. Previous research has shown that NLP technology, which interprets psychological health cues from user-generated texts, significantly improves the effectiveness of interventions [31,32], a finding consistent with our review [7,13,18,19,21,23]. Moreover, the development of deep learning techniques has led to the widespread use of general-purpose pretrained language models in various NLP applications [33]. This suggests that NLP-based algorithms are particularly suitable for delivering personalized psychological support in real-time interactions. Machine learning and deep learning algorithms show promise in enhancing AI chatbots’ effectiveness by continuously learning through user interactions. These algorithms can provide accurate and personalized interventions over time [34]. Future studies should investigate how these advanced algorithms can be used to offer more precise psychological support tailored to individual user experiences.

User engagement is a critical factor in the success of mental health interventions; however, several studies did not report intervention completion rates [20,21,23,27]. These issues might become more pronounced in real-world settings, where factors such as interface complexity, the chatbot’s ability to provide meaningful and personalized feedback, and technical issues impact the overall user satisfaction and perceived effectiveness of the intervention [35]. This underscores the need for further exploration of how chatbot interactions can be designed to maintain user engagement, possibly through improved user experiences and personalization techniques [36,37]. From a different perspective, unlike traditional interventions, which are typically designed to be completed from start to finish, chatbot interactions often do not follow this linear structure. Users may terminate interactions once their immediate concerns are addressed, reflecting the unique flexibility of chatbots [38]. For instance, if a chatbot provides critical information for crisis management and the user responds affirmatively, the user is likely to exit the chat after their need is met. This immediate resolution highlights the efficiency of chatbot interventions in addressing user concerns promptly, even if the interaction appears incomplete. Recognizing this distinction is essential for accurately interpreting dropout rates and evaluating chatbot effectiveness.

The AI chatbots reviewed here comprised various therapeutic approaches. The incorporation of established psychological treatment techniques such as cognitive behavioral therapy, mindfulness, and dialectical behavior therapy into AI chatbots represents a novel approach to delivering psychological health interventions. This approach is considered novel because it uses AI to deliver evidence-based psychological techniques 24/7, serve multiple users simultaneously, tailor interventions based on individual responses, and combine various therapeutic modalities on a single platform.

This multidisciplinary therapeutic approach has the potential to address health professionals’ diverse psychological needs [31]. AI chatbot interventions are characterized by a digital and responsive design, which allows for integrating practical tools such as stress management techniques, mood tracking, and personalized goal setting [39]. These features align with the immediate and personalized support often required by health professionals. However, while personalization is a key advantage of AI chatbots, a need also exists for some level of standardization to ensure the interventions’ quality and effectiveness [40]. This standardization should focus on establishing evidence-based core components and assessment methods while still allowing for personalized delivery [31]. Future research should focus on identifying the most effective components of these chatbots and tailoring them to the specific psychological health needs of health professionals while maintaining a balance between standardization and personalization.

In this study, a temporary increase in depression scores was observed in one study [20], whereas previous research has reported AI chatbots’ positive effects in reducing anxiety, depression, and work-related burnout among health professionals [41]. These contrasting findings may reflect the challenges faced by such professionals, who frequently experience high levels of stress and emotional demands [42]. This suggests the complexity of psychological interventions and highlights the importance of careful monitoring to determine the appropriate duration and intensity of chatbot use.

Some studies in our review demonstrated the potential of AI chatbots to act as first-line responders in identifying and managing mental health crises. Prior research has also shown the potential benefits of AI chatbots not only in improving mental and emotional well-being but also in promoting behavioral changes [43]. While these features may only benefit a subset of users, their impact on preventing escalation to severe mental health conditions or emergencies is invaluable. Future research should prioritize the integration and evaluation of these features to enhance the safety and efficacy of chatbot interventions.

Although AI chatbots address the unique time constraints and stress among health professionals, balancing automated responses with genuine therapeutic interactions remains a significant challenge. Personalization technologies that provide tailored features to AI chatbots have emerged as key factors in their effectiveness, with studies reporting increased user engagement and satisfaction through customized responses and immediate support [29]. Consequently, the development of tailored intervention strategies that consider the individual needs and preferences of health professionals is essential to maximize AI chatbots’ effectiveness. These strategies should integrate ongoing user feedback and dynamically adapt interventions to maintain long-term user engagement and effectiveness. Future research should focus on larger sample sizes, extended follow-up periods, and diverse health care professional groups of health professionals. Such comprehensive studies will play a crucial role in developing evidence-based guidelines for implementing AI chatbots in psychological health and in advancing the sophistication of personalization algorithms.

This scoping review has several limitations. First, the limited number of studies adopting diverse study designs and using various AI algorithms complicated making direct comparisons among the retrieved studies. Second, the findings’ generalizability is restricted because many studies were published in specific regions, particularly high-income countries. Additionally, the short duration of many interventions and limited follow-up periods make it difficult to evaluate the long-term effectiveness and sustainability of AI chatbots. Future research should address these limitations by conducting more rigorous trials, standardizing the outcome measures, and exploring the long-term applications of AI chatbots across diverse populations and settings.

This scoping review explored the current state of AI chatbots aimed at supporting the psychological health of health professionals. Although the reviewed studies demonstrated AI chatbots’ potential to reduce stress, anxiety, depression, and burnout, research in this area remains in its early stages. The diversity of study design, AI algorithms, therapeutic approaches, and outcome measures highlights this field’s innovative but fragmented nature.

Despite promising results, particularly with NLP-based chatbots, a significant need exists for more rigorous and standardized studies to fully evaluate their clinical efficacy. Challenges such as usability issues and limited generalizability of the findings must be addressed to enhance the long-term application and effectiveness of AI chatbots in real-world settings. Future research should focus on refining chatbot designs, expanding the research to diverse populations, and conducting long-term studies to clarify the role of AI chatbots in supporting the psychological health of health professionals. Thus, AI chatbots could become a valuable solution for addressing the psychological health challenges faced by health professionals worldwide.