Mental well-being is a complex construct that is commonly defined to include 2 perspectives—the hedonic and the eudaimonic perspective [1]. The hedonic perspective refers to a subjective experience (often summarized as happiness), which includes satisfaction with life, the presence of a positive mood, and the absence of a negative mood [2]. The eudaimonic perspective is defined as a process of self-realization that contains six dimensions, including (1) autonomy, (2) purpose in life, (3) personal growth, (4) positive relations with others, (5) environmental mastery, and (6) self-acceptance [3]. Mental well-being can be promoted through activities that aim to enhance, build, or strengthen individuals’ strengths, resourcefulness, or resiliency [4]. The main aim of mental well-being promotion is to enhance or maintain good levels of mental well-being. However, mental well-being and mental health are closely related, and therefore, mental well-being promotion often simultaneously improves mental health outcomes (such as anxiety and depression) [5].

A large proportion of nonclinical university students experience suboptimal levels of mental well-being [6-8]. This is likely due to the significant lifestyle changes university students experience, such as moving away from home, academic and financial pressures, and a loss of social and emotional support networks [9]. Additionally, research has suggested that approximately 25% of university students experience depression and anxiety [10], and this prevalence of mental health conditions in university students continues to rise. While many universities identify a duty of care for students’ mental health and well-being, support services at universities are often overstretched [9], thereby limiting students’ access to mental health and well-being support. This highlights the importance of accessible care in a nonclinical student population to prevent the onset of mental health conditions and improve their mental well-being.

A particularly promising way to deliver accessible mental health and well-being support to nonclinical university students is through fully automated digital technologies—digital technologies that do not require any form of ongoing human input to be delivered [11]. Automated digital technologies include apps that can be accessed through technological devices, such as smartphones and tablets [11]. Apps are designed to perform specific tasks, such as the delivery of activities for mental health and well-being. However, apps can also include generative artificial intelligence (GenAI), which allows for greater personalization of content, or they can rely entirely upon artificial intelligence (AI; eg, large language models) [12]. The automated delivery through apps and AI allows for mental health and well-being support to be highly scalable and cost-effective, thereby providing an easily accessible way to access support [13]. Unsurprisingly, mental health and well-being apps are becoming increasingly prevalent. The apps market size of US $6.25 billion in 2023 is expected to continue growing at 15.2% annually, with young adults (including university students) being its primary user [7,14].

A recent systematic review and meta-analysis indicates that automated apps and AI can effectively improve mental well-being in the nonclinical population [15]. Systematic reviews and meta-analyses also indicate that apps can effectively reduce mental health symptoms alongside improving mental well-being [16]. Additionally, AI has been shown to effectively support mental health [17]. Despite this great potential, studies have shown that these technologies suffer from suboptimal levels of engagement and high levels of dropout, thereby not reaching their full potential [15,18].

Some research has been conducted to understand engagement with specific mental health and well-being apps. Specifically, research investigated student engagement after providing them with access to a mental health app “Thought Spot” (Thought Spot was created through participatory design research methods and usability testing with college and university students [19]) for 6 months [20]. They found important app-specific determinants of continued engagement, such as detailed and inclusive content enhancing engagement, and technical errors and a lack of integration with other apps reducing engagement. Further research aimed to understand participants’ engagement with mood-tracking apps [21]. They found further context-specific determinants of engagement, such as enjoying keeping track of trends in their moods through calendars. Similarly, research found app-specific determinants of engagement with a mental health promotion intervention, such as engaging and accessible app content [22].

Borghouts et al [23] conducted a review on barriers to engagement across different mental health and well-being apps, investigating studies from 2010 to 2019. However, this review did not adopt a nonclinical university student sample, which is particularly important since the review found mental health status to be an important determinant of engagement [23]. Furthermore, tremendous digital developments have occurred since then, and the review does not include perspectives regarding engagement with more advanced technologies, such as AI. Melcher et al [24] further conducted an empirical study to understand engagement across different mental health apps in college students, but did not conduct an in-depth qualitative analysis and may therefore provide less depth toward the contextual understanding of engagement [25]. Jardine et al [26] conducted an exploratory analysis of questionnaire and interview responses regarding engagement with mental health apps. However, this study focused on nonengagers. Overall, there is a need for research into engagement with AI and apps for mental health and well-being in a nonclinical student population, adopting in-depth qualitative analyses.

Therefore, this study aimed to understand university students’ experiences of engaging with AI and apps for their mental health and well-being.

This study adopted semistructured qualitative interviews.

A constructivist epistemological stance was taken in this study. This approach was adopted as it identifies that the researcher, interviewing and analyzing the data, plays a role in the construction of truth and meaning [27]. The approach identifies that people can construct the truth or meaning differently in similar situations. Therefore, it was considered important to reflect on the researchers’ subjectivity and its impact on the data and findings by adopting a reflexive thematic analysis (TA) [28].

Participants were recruited via convenience sampling at the researchers’ institution. A study advert (with relevant information regarding the study and a link to sign up) was promoted at sites at the researchers’ institution, and via online university channels. University channels used to advertise the study targeted students in the Department of Psychology and Computer Science.

To ensure the confidentiality of participants, no demographic information was collected. Particularly, since demographic characteristics (such as age, gender, year of study, and department) in combination with qualitative interview data were considered to lead to highly identifiable data. Therefore, the general demographics of students in both departments were obtained instead.

The sample size was considered using the information power principle. This suggests that the more information a sample holds, the smaller the sample is needed [29]. This study adopted a relatively broad study aim and cross-case analysis, but a dense sample specificity, use of an established theory, and strong dialogue. Therefore, the initial aim was to recruit 15‐20 participants.

A convenience student sample was recruited. Eligible participants for this study were (1) university students, (2) aged more than 18 years. Participants were excluded if they (1) were receiving treatment or psychopharmacological drugs for a mental health condition at the time of the study, (2) they scored ≥10 on the Patient Health Questionnaire-8, and (3) they scored ≥8 on the Generalized Anxiety Disorder-7. This was to ensure participants’ mental health could be considered “nonclinical,” as fully automated digital technologies are considered particularly beneficial as universal interventions targeted at the general nonclinical population [30].

There were no specific inclusion or exclusion criteria regarding students’ experiences with mental health and well-being apps. Therefore, some students were currently using or had used several mental health and well-being apps, while other students had never used them before. Furthermore, the apps that students had used included a range of different ones, and the results are therefore not specific to an app. This was to ensure general perspectives and experiences regarding engagement across mental health and well-being apps were considered. Nevertheless, inductive coding was adopted to identify the mental health and well-being apps (including AI) that students referred to during the interviews.

A semistructured interview schedule was developed for the purpose of this study (Part S1 in Multimedia Appendix 1). This interview schedule adopted several open-ended questions aimed at understanding students’ experiences of using AI and apps for their mental health and well-being. Since the initial focus of the interview was entirely on mental well-being promotion, participants were asked about their understanding of “mental well-being promotion” and provided a definition by the researcher (refer to Introduction section) to ensure participants had a consistent awareness of the meaning of this construct.

Students expressed their interest in participating in the study by signing up via an online questionnaire link on the study adverts. This link provided participants with the information sheet (Multimedia Appendix 2), after which they were taken to the screening questionnaires (Patient Health Questionnaire-8 and Generalized Anxiety Disorder-7). After filling out the screening questionnaires, participants who were not eligible were provided with relevant information and resources for their mental health and well-being. Participants who were eligible were contacted by the researcher (JG) via email to schedule a convenient date and time for the interview to take place. The interviews took place in the laboratories at the researchers’ institution between April 19 and May 31, 2023. Upon arrival in the laboratory, informed consent was obtained (Multimedia Appendix 3). Following this, a brief explanation was provided, and the interview was conducted. The interview consisted of 2 separate parts, lasting a maximum of 1 hour in total. The first part included a semistructured interview regarding participants’ views of apps and AI for mental health and well-being support; this part of the interview was analyzed and is presented in this study. The second part of the interview involved participants looking at app content and sharing their thoughts out loud; this part of the interview was analyzed in a separate study. Refer to Multimedia Appendix 1 for the interview schedule, including both parts of the interview. Once the interviews finished, participants were provided with a debrief form (Multimedia Appendix 4) and student credits or a £10 (US $13.52) gift voucher to thank them for their time.

Ethical approval was obtained from the Psychology Research Ethics Committee at the University of Bath (reference 23 042). Participants were first presented with an information sheet when they accessed the online expression of interest questionnaire. They were presented with the information sheet again at the start of their interview in the laboratory. Participants were provided with an opportunity to ask questions, and informed consent was obtained in writing before data collection. Participants had the opportunity to withdraw during the interview by informing the interviewer. After the interview was completed, participants could withdraw their data until May 31, 2023. Participants were informed about this on the information sheet. To ensure confidentiality, any identifiable information was anonymized during transcription, and participants were assigned pseudonyms, which are used in this publication. Furthermore, to mitigate participant burden, combined duration of Part 1 and Part 2 of the interview (with Part 2 reported elsewhere) remained within a 1-hour time limit. Aftercare was provided if necessary. Specifically, if sensitive topics were discussed and/or concerns regarding mental health and well-being were raised during the interview, a discussion between the researcher and the supervisory team was conducted to identify the most appropriate form of aftercare—such as the university well-being apps and/or other mental health support services.

Data was analyzed using a reflexive TA [28]. This was considered the most appropriate method of analysis to identify different themes and patterns of meaning regarding students’ engagement with AI and apps for their mental health and well-being. An inductive approach was adopted to analyze the data. The interviews were audio-recorded using an Olympus VN-541 PC high-quality digital voice recorder. The part of the semistructured interviews reported in this study lasted 30 minutes 2 seconds (SD 9 min 49 s) on average, ranging from 16 minutes 20 seconds to 49 minutes 37 seconds. The recordings were transcribed using an orthographic (verbatim) transcription method. This method was adopted as it was decided that only information relevant to understanding and conceptualizing the meaning would be necessary to answer the research question and would therefore be transcribed. Interview transcripts were fully anonymized, and each participant was assigned a pseudonym.

The six phases of TA proposed by Braun and Clarke [28] were used to analyze the primary research question: (1) familiarization with the data took place, (2) transcripts were coded, (3) themes were generated, (4) themes were reviewed, (5) final defining and naming of themes, and (6) write-up took place. Specifically, a reflexive journal was kept throughout the analytical process, and continuing discussions took place between author JG and the research team to further conceptualize meaning. The transcripts were coded by author JG using NVivo 14 (Lumivero), adopting an inductive interpretive approach, which at times was more descriptive in nature. Codes were regularly revisited by JG to conceptualize and construct meaning. Initial themes were constructed by JG, which were further developed and reviewed in discussion with coauthors.

A total of 21 participants took part in this study. The initial estimation of sample size (guided by information power) was continually appraised during the analytic process [29]. Specifically, the authors identified that participants frequently conflated mental well-being promotion and mental health support during interviews. While there is considerable overlap between mental well-being promotion and mental health support, important differences exist. For example, mental well-being promotion primarily aims to maintain or improve mental well-being outcomes and is often delivered through universal interventions targeting the entire nonclinical population, while mental health support primarily aims to support mental health (such as anxiety and depression) and is often delivered to targeted or at-risk populations [30]. Despite this, conflation highlighted the need to revise the initial study aim to include mental health support alongside mental well-being promotion. This revision expanded the scope of the study aim, and therefore, it was agreed that additional participants were required to capture the thematic complexity. The final sample size was then determined using code saturation [32]. Author JG identified code saturation (whereby no new codes were identified) after 21 interviews. Code saturation was then verified through discussion with the research team.

Demographic characteristics of the population from which participants were recruited are presented in Table 1.

Inductive coding identified a range of mental health and well-being apps and AI technologies referred to by participants during the interviews. These include habit and mood trackers, journaling, exercise, sleep, mindfulness, productivity, university well-being apps, mental health support via rule-based and AI-based chatbots, as well as AI therapists, AI to tailor or personalize content within apps, and general large language models, such as ChatGPT (OpenAI). For a full overview of the apps and AI participants (explicitly) referred to during interviews, refer to Multimedia Appendix 5. Some of the below findings referencing these apps and AI tools for mental well-being promotion and mental health support are discussed hypothetically, while others may reflect participants’ actual experiences.

A total of 4 main themes and 4 subthemes were constructed by the authors to provide an understanding of students’ experiences of engaging with AI and apps to promote and support their mental health and well-being (Table 2).

Most students identified that they did not engage with AI and apps for their mental health and well-being, as they did not currently experience a mental health or well-being “need.” Students mentioned that this lack of “need” was due to their mental health and well-being being “good” at that moment.

Jack’s account was interpreted as positioning engagement with AI and apps as conditional on experiencing lower well-being; since he currently feels his well-being is “quite high,” Jack does not perceive a “need” to engage at this time. Interestingly, this contradicts theory that considers self-actualization a “growth need”—a need that occurs out of a desire to grow rather than a deficiency [33]. However, perhaps students perceive mental health and well-being as a deficiency need instead of a growth need. Deficiency needs motivate people when they are unmet, and once a deficiency need has been satisfied, it goes away [34]. Indeed, students identified that they used AI and apps in times of “need,” but discontinued their use once this “need” was satisfied and went away.

Brenda identified that once the “need” to engage disappeared as her mental health improved, she would discontinue using the intervention. This may suggest students are not motivated to self-actualize through AI and apps and perceive these as relevant tools to satisfy a deficiency need instead.

Nevertheless, it may be consistent with the technology acceptance model (TAM). This model suggests that engagement with AI and apps occurs when this is considered “useful” [35]. Perhaps engaging when there is no mental health “need” might be considered less useful. Instead, most students mentioned that they would rather engage in other activities to promote their mental well-being and self-actualize.

Holly and Elsa identified that when there is no immediate “need” to work on their mental health and well-being, they prefer engaging in other (nondigital) activities for their mental health and well-being. This might be because most students identified that they experience more joy and further benefits, such as connectedness and improvements to their physical health, when pursuing other activities. This may suggest that automated digital activities do not lead to the same satisfaction of basic psychological needs. According to the basic psychological needs theory (BPNT), the satisfaction of (1) autonomy (volition and willingness which lead to a sense of authenticity), (2) relatedness (connection, experiencing warmth, bonding, and care), and (3) competence (the experience of mastery to use and extend skills and expertise) are essential for optimal mental health and well-being [36]. Hence, perhaps students pursue other well-being activities as they lead to greater satisfaction of basic psychological needs.

Since students had already implemented these other well-being activities in their daily lives, they would also often prefer to stick to these activities instead of implementing a novel app or AI activity in their daily lives. They mentioned that only when these activities did not provide the necessary support or effect for their mental health and well-being, would they consider adopting novel mental well-being strategies delivered through automated digital technologies.

Molly highlights that her friend group would be her mental health and well-being support strategy. Whenever she is unable to access this support strategy, she would consider looking at novel support strategies such as apps and AI. Molly’s account was interpreted to position apps as a secondary option—useful when her primary support system (friends) is unavailable—supporting the construction of the theme of need-dependent engagement.

Although students mention they are interested and open to using automated digital technologies for their mental health and well-being in times of “need,” they also identified that adopting these self-help strategies when in “need” could be challenging. Primarily, these interventions require individuals to self-manage their condition, rather than receiving more intensive support and guidance, such as through traditional face-to-face support methods. Adopting self-help in times of “need” could be particularly challenging when experiencing reduced mental capacity.

Henry identifies that engaging with automated digital technologies to perform a self-guided activity could feel “unreachable” and “unrealistic” when in times of “need.” Similarly, research has found that the majority of students are interested in using an app for their mental health, but that only a small proportion of students (16/149, 10.7%) have actually engaged with apps during times of need [37]. This might indicate the challenge to engage with automated digital technologies when in times of “need.”

Another important factor impacting student engagement is public and self-stigma. While it has been suggested that AI and apps are associated with less stigma compared with traditional (nondigital) mental health support services [13], it is important to recognize that stigma still plays an important role, thereby reducing engagement. Several students identified that accessing any psychological intervention—including ones delivered digitally—is stigmatized. Perhaps this is unsurprising as stigma is considered a prominent barrier to help-seeking generally [38].

Jack highlights that stigma, mental health and well-being apps, and AI not being normalized, are important reasons for him not to engage. This is in line with research, which has found that self-stigma (someone’s internalized attitudes) [39] has been linked to reduced help-seeking behaviors [40,41]. Jack shows self-stigma by mentioning that he personally does not think it is normalized, as he does not see anyone using it. Nevertheless, this contradicts recent findings that mental health and well-being apps are downloaded over a million times monthly [42].

However, stigma is understood as more prominent when apps have been developed for specific mental health purposes (such as an AI-based therapist) as opposed to low-key well-being activities (such as journaling and mindfulness meditation). It was also understood as a more prominent barrier when digital technologies are developed by recognized mental health organizations (such as universities, government, and mental health charities) compared with the private sector.

Elsa and Jack suggest that the barrier “stigma” to engage with AI and apps for mental health and well-being is perhaps not uniform, suggesting that students perceive greater stigma when tools appear more clinical or institutionally linked. This likely arises from the perception that accessing more intensive psychological support, such as AI and apps developed by recognized mental health organizations, is associated with weakness. Yap et al [43] refer to this specific kind of stigma as “weak-not-sick beliefs,” which is further associated with lower intentions to seek support. Indeed, several students mentioned that admitting to a decline in their mental health and well-being to an extent where they “need” additional support is stigmatized. Therefore, students adopt other strategies before admitting to this “need” for support and accessing AI and apps.

Harry mentions he would try another activity first, as he tries to stay away from “professional help.” Harry’s emphasis on independence demonstrates how internalized stigmatizing beliefs about help-seeking shape reluctance to engage with digital tools. Harry mentions that he is an “independent person,” which indicates that he perceives individuals who access support as not being independent and could relate to the “weak-not-sick belief.” This again highlights the stigma associated with accessing mental health support generally, and therefore, apps for mental health and well-being support too. Nevertheless, mental well-being promotion might also suffer from this barrier, as promotion and support were often considered interchangeably by students.

Additionally, stigma was further shown in students hiding that they were accessing support. Students identified that it is easier to hide that they are accessing support when accessing AI and apps, but even then, they expressed concerns that someone else could find out.

Ralph mentions that someone finding a mental health and well-being app on their phone could be considered a barrier to engagement. This highlights the perceived public stigma associated with accessing any psychological intervention or support, and therefore, not wanting others to be aware of accessing this support. However, this could increase engagement relative to traditional face-to-face support methods, as it may be easier to hide that you are accessing support when accessing AI and apps.

Although students highlighted stigma would prevent them from engaging with apps and AI generally, students also identified that they could potentially provide a solution to public and self-stigma.

Anne and Barbara highlight the importance of engaging in something low-key for mental health and well-being first, which may act as a solution to gradually reduce stigma to engage with more intensive mental health and well-being support. Indeed, Ho et al [44] suggest that a stepped-care model could reduce stigma and improve access to those in “need.” They suggest a stepped-care model in which different levels of care are provided, from less intensive care (such as self-guided interventions) to more intensive care (such as specialist treatment). Research further identifies several approaches that can effectively reduce stigma by adopting apps. For example, education-based approaches are one of the most commonly used approaches to combat stigma [45]; furthermore, mental well-being activities (such as the positive psychology activity “expressive writing”) have been found to reduce self-stigma [46]. Both of these activities could be delivered through digital technologies and thereby help reduce stigma. Thus, AI and apps could potentially provide a crucial way of reducing stigma to alter public and self-stigma and enhance engagement with mental health and well-being support accordingly.

This study aimed to understand nonclinical students’ experiences of engaging with AI and apps to promote and support their mental health and well-being. A reflexive TA of 21 interviews led to the construction of 4 main themes and 4 subthemes. The main themes identified how (1) experiencing a “need” to engage, (2) stigma as a barrier and solution, (3) a lack of trust, and (4) usage dependent on unique AI and app characteristics can impact students’ engagement with apps and AI for their mental health and well-being. Furthermore, in the theme “lack of trust,” the subthemes “insufficient capability to support and promote mental health and well-being?” and “an inability to safeguard mental health and well-being?” highlight the lack of trust students currently experience in AI and apps to effectively support their mental health and well-being, thereby reducing their engagement. Finally, the theme “usage dependent on unique AI and app characteristics” consisted of subthemes “no judgment and pressure” and “a lack of humanity, warmth, and care,” showing the unique characteristics of AI and apps, which determine the situations in which students are likely to engage with these digital technologies for their mental health and well-being.

It is important to reflect on the positioning of the findings of this study to consider its transferability. First, this research was conducted at a university in the United Kingdom, situating the findings within a Western cultural context and a highly educated sample. The majority of students at this university were undergraduate, home-based students, although a growing proportion (28,957/103,618, 23%) were international students [64]. At the time of this study, 65% (12,970/19,770) of students at the university identified as White, while 35% (6800/19,770) identified as belonging to Black, Asian, and ethnic minority groups. According to the TAM, such demographic variables influence engagement with digital interventions [58]. Therefore, the positioning of our participants in a Western university context limits transferability of the findings; for example, to other population subgroups with similar demographics (such as university students in a Western context, which includes hundreds of millions of students).

Second, recruitment strategies focused on the departments of Psychology and Computer Science, further positioning the sample as individuals with greater knowledge of mental health and well-being and/or digital technologies. The TAM suggests how a range of factors, such as self-efficacy, perceived ease of use, and technology anxiety, among others, play an important role in determining engagement with digital interventions. The study population’s enhanced knowledge may lead to greater self-efficacy, ease of use, and reduced anxiety of adopting technology and/or mental health and well-being activities. Hence, this limits the transferability of the findings to other population subgroups with greater knowledge of digital and/or mental health and well-being.

Finally, the rapid and ongoing development of GenAI positions this study temporally. Specifically, findings reflect participant perceptions formed during a specific technological and social context during which GenAI was novel while still rapidly developing. The TAM suggests that subjective norms and attitudes are important in determining engagement with digital technologies [58]. The subjective norms and attitudes participants may have held toward AI and apps may be temporally bounded. Nevertheless, the findings of this study hold relevance for similar populations and contexts, and can provide meaningful contributions to relevant theoretical frameworks, conceptual debates, and practical implications, which are explored in the following sections [65].

This study has several limitations that should be considered when interpreting the findings. One study limitation is that students referred to mental well-being promotion and mental health support interchangeably. Initially, the interview focused on mental well-being promotion entirely, and therefore, a definition of mental well-being promotion was provided at the start of the interview (Multimedia Appendix 1). Despite this, participants frequently drifted from the definition to discuss the relevance of apps for mental health support instead. Participants’ interchangeable use of these terms limits the findings, as it remains unclear whether reported experiences of engaging with apps and AI relate to mental well-being promotion, mental health support, or both. Similarly, participants’ interchangeable reference to apps and AI makes it unclear whether these experiences relate specifically to apps, AI, or both.

Furthermore, while the research question was refined to include mental health support (which is considered good practice when adopting a reflexive qualitative approach [70]), the initial interview schedule did not include questions relating to mental health support. Therefore, findings relating to mental health support and crisis situations were entirely emergent and did not derive from systematic questioning across participants. This limits the representativeness of these findings. Therefore, future research should adopt systematic questioning across participants for more representative insights into their perceptions of engagement with apps for mental health support and during periods of crisis.

Another limitation is that complexities and contradictions may have been lost as the dataset was analyzed cross-cases. This is a particular issue in larger samples [28]. When focusing on patterns, the commonalities in the dataset are identified. Nevertheless, it is important to recognize that individual differences exist. For example, the TAM identifies the importance of demographic and personality characteristics impacting engagement with digital technologies [58]. Similarly, we found that while most students would only use AI and apps for mental health and well-being when in “need,” some students would continually engage to promote their mental well-being. This loss of data is called the “fragmentation of accounts” [71], and while this means there is a loss of contradiction within the dataset, the study’s aim was to provide a general overview of the commonly experienced factors to engage with AI and apps for their mental health and well-being. Nevertheless, it is important to consider that differences in perceptions and engagement exist across individual differences, which should be further investigated in future research.

Finally, a limitation of the study was that no demographic data were collected. While demographic data were not collected in order to protect participants’ confidentiality, this limits the contextual understanding of the findings in this study. For example, it remains unclear whether participants were primarily male or female, and research has shown that sex (among other demographic factors) is important in determining the acceptance and engagement with digital technology [58]. Nevertheless, despite this limitation, the National Institute for Health and Care Excellence suggests that clear reporting of the recruitment strategy (eg, recruitment within the department of Computer Science and Psychology) can still contextualize the study findings to some extent [11].

The themes constructed in this inductive reflexive TA have identified many different areas for future research into AI and apps to promote and support mental health and well-being in students. First, since students primarily identified interest in engaging in times of “need,” future research could focus on engagement and the effectiveness of adopting these unguided self-help interventions during times of “need.” Indeed, a recent systematic review aimed to understand engagement with unguided self-help apps in clinical populations but found insufficient evidence [72]. Thus, it remains unclear how effective these interventions are when in “need.”

Second, while this study presents common perceptions and experiences regarding engagement, such as most students primarily being interested in engaging for mental health support, it is important to consider that individual differences exist. Indeed, Borghouts et al [23] found that individual differences impacted engagement with digital technologies. However, it remains unclear which individual differences impact engagement with and the effectiveness of digital mental well-being promotion [15]. Thus, future research should investigate the impact of individual differences in engagement with AI and apps for mental health and well-being promotion.

Third, while stigma has been identified as a barrier to engagement, the current findings propose that AI and apps could help reduce public and self-stigma. Either through a stepped-care model [44], or through effective activities combating stigma being incorporated into apps [45]. Nevertheless, future research is needed to investigate whether a stepped-care model that incorporates apps and engagement with activities within apps that reduce stigma could ease the public and self-stigma in a UK university setting.

Finally, this study identified that building trust in digital health technologies is an important determinant of engagement with AI and apps for mental health and well-being support. However, it was also identified that mistrust is occasionally warranted due to the limitations in AI and app capabilities [50]. Future research should therefore establish when and under what conditions AI and apps can be “trusted.” It should establish when and under what conditions AI and apps are safe to use and effective for mental health and well-being support. This future research would be essential to ensure appropriate recommendations regarding engagement can be made and to foster justified trust in these interventions.

Adopting an inductive thematic approach in this study allowed the generation of unanticipated insights and insights with actionable outcomes that could shape policy development [28]. Several implications for policy and actionable outcomes for AI and app developers are identified.

First, the study found that stigma was perceived as a common barrier to engagement in a UK setting. Therefore, UK policy and practice should consider ways to reduce the stigma associated with AI and apps for mental health and well-being purposes. Ho et al [44] suggest that less intensive support may be considered less stigmatizing. Hence, it may be important for digital mental well-being promotion to be advertised as distinct from more intensive digital mental health support to reduce stigma. Furthermore, other methods could be adopted to reduce public and self-stigma, including adopting education-based programs [73].

Furthermore, developers should consider whether developing AI and apps for the purpose of mental well-being promotion is most beneficial to nonclinical students. The World Health Organization suggests mental well-being promotion can be developed through universal interventions [30] that are aimed at general (nonclinical) population groups who are not specifically “at-risk” due to lower levels of mental health or well-being. Furthermore, developing these interventions digitally provides many benefits. However, this study suggests that mental well-being promotion interventions may be more engaging when delivered nondigitally. It also suggests that students are more willing to engage with AI and apps in times of “need.” Thus, perhaps future AI and app development should primarily focus on supporting mental health conditions instead of promoting mental well-being.

Finally, policy and practice should incorporate ways to safeguard mental health when making AI and apps publicly available. Mental health conditions may present and can be identified differently across cultures [57], making it more challenging to safeguard mental health when these automated digital technologies are publicly available worldwide. This may be particularly challenging, as identifying high-risk individuals may have consequences for data protection and security [74]. Nevertheless, to build trust and enhance engagement with AI and apps, further strategies should be adopted by developers to safeguard mental health and well-being.

In conclusion, this study aimed to understand university students’ experiences and perspectives of engaging with AI and apps to promote and support their mental health and well-being. Several themes were constructed by the research team, identifying that (1) most students have to experience a “need” to engage; (2) engagement may be impeded by stigma in university students in the UK, although AI and apps might also provide a promising solution to stigma through a stepped-care model and educational activities; (3) a lack of trust in the capability of AI and apps for mental health and well-being and its ability to safeguard are currently perceived as barriers to engagement, and policy and practice should consider ways to build trust to enhance engagement; and (4) unique AI and app characteristics may determine in what situations individuals may engage. Future research should investigate for whom these digital technologies may be engaging in nonclinical populations and when trust in AI and apps for mental health and well-being support may be warranted. Finally, future policy and practice should adopt strategies to reduce public and self-stigma and build trust by incorporating safeguarding strategies for mental health and well-being, particularly when scaling these automated digital technologies worldwide. Nevertheless, these accessible and cost-effective interventions provide a promising method to support mental health and well-being in university students.