Exploring Nurses’ Perspectives on the Use of Artificial Intelligence Chatbots for Mental Health Support: A Cross-Sectional Study in Greece

1. Introduction

According to WHO, in 2019, 970 million people globally struggle with a mental disorder, with anxiety and depression the most common [1]. Throughout the complex content of mental disturbances, many aspects of life are impacted, including relationships, family, friends and community. They can result to problems at school and work. People who suffer from a mental disorder, are in a high risk of a suicide, lose their productivity and as a consequence the economic impact in health care is tremendous. Approximately it is estimated that 70% of people with mental illness do not receive a formal treatment [2], while others do not seek for treatment due to low perceived need, attitudinal barriers like stigma or low income [3]. Stigma associated with mental illness is a significant barrier to seek for help, leading to negative attitudes and intentions toward mental health services among individuals [3,4].

On the other hand, there are critical concerns about the global shortage of mental health professionals [5]. There is reported worldwide poor distribution of behavioral health professionals across the health care domains, which considers several issues about the delivery of mental health services to individuals. Findings support that there is, also, shortage of mental health nurses [6,7]. Nursing workforce shortages have long been addressed with exacerbating attrition and projected workforce shortfalls [8]. Due to the relational nature of interpersonal work, the large part of stressors in the everyday workplace attrition, nurses require strong cognitive, emotional and relational skills and the ability to self-regulate [9]. They are often exposed to challenges related to the provided interpersonal care of people in mental, emotional distress, with self-harm or suicidal behaviors, as well as, they have to handle clinical aggression and interpersonal conflicts [10]. These can impact nurses’ own mental health, resulting lower mental health than population norms [10].

During COVID-19 pandemic, issues about the accessibility in mental health care were raised, due to the fact that access to care became more difficult. Immediate interventions to target in mental health care were needed to overcome the barriers and to engage adult and young population in psychosocial mental health treatment [11]. In this direction, the rapid expansion of the data ecosystem leading to the development of artificial intelligence innovations and the development of machine learning models. Among this, the integration of large language models (LLMs) into health care gathered attention due to their potential to enhance diagnostic accuracy, streamline clinical workflows, and address health care disparities [12]. LLMs typically refer to pretrained models (PLMs) that have a large number of parameters and are trained on massive amount of data. The innovation that has emerged the LLMs as the most prominent area of research in artificial intelligence is the ability to solve complex tasks. LLMs are the core element of generative AI applications which have the ability to understand natural language but also generate corresponding text, images, and videos. This human-machine interaction has a great potential in medical diagnostics and a medical decision-making [12]. The field of LLMs is rapidly expanding to disease classification, medical question answering and diagnostic content generation. Previous studies demonstrated high accuracy of LLMs in radiology, psychiatry and neurology [13,14].

In the era of LLMs, conversational AI, driven by these models, rapidly expands. AI-driven conversational technologies has the potential to provide instant, interactive and context-aware access to clinical recommendations [15]. AI-powered chatbots are referred as conversational agents or relational agents, that are able to hold a conversation with the human user, enhancing the engagement of healthcare professionals with clinical guidance [16,17]. Digital mental health care industry bridges the gaps between mental health services and individuals through the incorporation of artificial intelligence (AI) creating AI-guided products. AI technologies that perform human-like physical and cognitive tasks, enable the automation, engagement and decision empowered by AI commonly reviewed by domain experts before they implement in a patient’s treatment plan [18]. Moreover, the capability of AI systems support both clinicians and patients, impact the quality of population health by transforming mental healthcare through enhancing patient retention, work life of healthcare professionals and reducing costs [19]. LLM-based chatbots or conversational agents are digital assistants that exist either in hardware or software that use machine learning and artificial intelligence methods to mimic human behaviors and evolve dialogue able to participate in conversation [17].

Previous studies have explored the potential of AI-powered chatbots in mental health [11,16,20,21]. For nurses, intelligent conversational agents highlight opportunities to augment nursing practice, care planning, support patient education, empower individuals and manage their health [22,23]. While AI chatbots show promising signs in diagnostic accuracy and clinical-decision making, currently human clinicians exhibit benefits in holistic clinical abilities, a skill requiring experience, contextual knowledge and ability to bring concise responses [22]. Prior research ensures that conversational agent-based interventions are feasible, acceptable and have positive effects on physical function, healthy lifestyle, mental health and psychosocial outcomes [24]. Others suggest the adoption of Retrieval Augmented Generation (RAG)-enabled AI conversational systems to mitigate misleading information due to biases, outdated training data or misinformation embedded in public sources [15]. There is a further need of ongoing research to examine if AI conversational technology can adequately substitute nursing expertise, while prioritizing patient safety, privacy and equitable assess to care.

Since LLMs and AI-driven chatbots have been increasingly explored as problem-solving aids across various healthcare domains, the access in medical information is providing quickly. Scientific community has continued working to expand the capabilities of these models and as a consequence many ethical concerns have raised, including privacy risks, model hallucination, alongside regulatory fragmentation. As technology evolves, confidence among healthcare professionals and patients on medical technology is affected by ensuring that AI is evidence-based, free from bias and promotes equity in care [19]. More precisely, all the above determine that the adoption of AI chatbots in clinical settings remains undoubtedly [19,22]. Therefore, giving the growing presence of artificial intelligence in healthcare, it is essential to assess whether nurses trust and adopt AI-powered conversational technologies in nursing decision-making, under what conditions and circumstances or reject and dispute them.

Thus, this study explores nurses’ perceptions and concerns regarding the use of artificial intelligence (AI) chatbots as a tool for mental health support. Additionally, we evaluate the levels of acceptance and fear toward AI, while we focus to examine the influence of demographic variables on these attitudes.

2. Materials and Methods

2.1. Study Design

A cross-sectional study was conducted in Greece, with data collection taking place in October 2025 via an online survey. The questionnaire was digitized using Google Forms and disseminated through nursing-focused groups on social media platforms such as Facebook and LinkedIn, resulting in a convenience sample. Eligibility criteria for participation included: (a) being a registered nurse, (b) having an active account on at least one social media platform, (c) engaging with social media and/or the internet for a minimum of 30 minutes daily to ensure sufficient exposure to digital technologies, and (d) providing informed consent prior to completing the survey. The study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [26].

The sample size was calculated using G*Power software (version 3.1.9.2). The following parameters were applied in the computation: (a) a Type I error probability of 5%, (b) statistical power of 95%, (c) five independent variables, and (d) a small effect size (f² = 0.05) for the association between each independent and dependent variable. Based on these criteria, the required sample size was determined to be 262 nurses.

The sample size was calculated using G*Power software (version 3.1.9.2). The following parameters were applied in the computation: (a) a Type I error probability of 5%, (b) statistical power of 95%, (c) five independent variables, and (d) a small effect size (f² = 0.05) for the association between each independent and dependent variable. Based on these criteria, the required sample size was determined to be 262 nurses.

2.2. Measurements

The demographic variables examined were: (a) gender (male or female), (b) age, treated as a continuous variable, (c) self-assessed financial status on a scale from 0 (very poor) to 10 (very good), (d) daily time spent on social media and/or websites (continuous variable), and (e) self-reported proficiency in digital technologies - including smartphones, personal computers, AI applications, and tablets - measured on a scale from 0 (very poor) to 10 (very good).

Nurses’ attitudes toward mental health chatbots powered by artificial intelligence were evaluated using the Artificial Intelligence in Mental Health Scale (AIMHS) [27]. This instrument comprises five items distributed across two dimensions: technical advantages (2 items) and personal advantages (3 items). Participants responded using a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Factor scores were computed by averaging the item responses within each dimension, yielding scores between 1 and 5, with higher values reflecting more favorable attitudes toward AI-based mental health chatbots. Two negatively worded items were reverse-coded prior to analysis. The validated Greek version of the AIMHS [27] was employed in this study. Internal consistency was deemed very good, with Cronbach’s alpha coefficients of 0.788 for the technical advantages subscale and 0.896 for the personal advantages subscale.

To assess nurses’ acceptance and fear of artificial intelligence, we employed the Attitudes Towards Artificial Intelligence Scale (ATAI) [28]. This instrument comprises five items: two items evaluate acceptance of AI, while three items assess fear related to AI. Responses are recorded on an 11-point Likert scale ranging from 0 (strongly disagree) to 10 (strongly agree). The overall score is calculated as the mean of all item responses, yielding a range from 0 to 10. Higher scores on the acceptance subscale reflect greater acceptance and more favorable attitudes toward AI, whereas higher scores on the fear subscale indicate increased fear and more negative attitudes. For this study, we utilized the validated Greek version of the ATAI [29]. Internal consistency was very good, with Cronbach’s alpha coefficients of 0.857 for the acceptance subscale and 0.715 for the fear subscale.

2.3. Ethical Issues

Participation was entirely voluntary and conducted anonymously. Prior to enrollment, all participating nurses received comprehensive information regarding the study’s aims and procedures and provided written informed consent. The research was carried out in full compliance with the ethical principles outlined in the Declaration of Helsinki [30]. Ethical approval for the study was granted by the Ethics Committee of the Faculty of Nursing at the National and Kapodistrian University of Athens (Protocol No. 01, dated September 14, 2025).

2.4. Statistical Analysis

Categorical variables were described using frequencies and percentages, whereas continuous variables were summarized through means, standard deviations (SD), medians, and interquartile ranges. The normality of continuous variable distributions was evaluated using the Kolmogorov–Smirnov test and Q–Q plots, both of which supported the assumption of normality. Demographic characteristics were treated as independent variables, while attitudes toward AI-based mental health chatbots, AI acceptance, and AI-related fear were considered as dependent variables. Given the continuous and normally distributed nature of the dependent variables, both simple and multivariable linear regression analyses were conducted. Initially, bivariate linear regressions were performed, followed by the development of final multivariable models. Multicollinearity was assessed using variance inflation factors (VIFs), with values exceeding 5 indicating potential multicollinearity concerns. Model assumptions were verified by inspecting residual histograms for normality, and residual-versus-predicted scatterplots for homoscedasticity and linearity [31]. The findings are presented as unadjusted and adjusted beta coefficients, along with 95% confidence intervals (CIs) and p-values. Furthermore, Pearson’s correlation coefficient was employed to examine the association between AIMHS and ATAI scores. P-values less than 0.05 were considered statistically significant. We used the IBM SPSS 28.0 (IBM Corp. Released 2021. IBM SPSS Statistics for Windows, Version 28.0. Armonk, NY: IBM Corp) for the analysis.

3. Results

3.1. Demographic Characteristics

The final sample comprised 276 nurses, with the majority being female (76.2%). The mean age of participants was 42.94 years (SD = 9.61), with a median age of 45 years and an interquartile range of 13. The average self-perceived financial status was 5.97 (SD = 1.56), with a median score of 6 and an interquartile range of 2. Participants reported a mean daily duration of social media and/or website use of 3.07 hours (SD = 2.40), with a median of 2 hours and an interquartile range of 3.5. The mean score for self-assessed digital competence—including the use of smartphones, computers, AI applications, and tablets—was 7.63 (SD = 1.67), with a median of 8 and an interquartile range of 2. A detailed overview of the demographic characteristics is provided in Table 1.

3.2. Study Scales

The mean score on technical advantages factor was 1.76 (SD; 0.69), while on personal advantages factor was 3.26 (SD; 0.88). The mean acceptance score was 5.15 (SD; 1.99), while the mean fear score was 5.72 (SD; 1.80). Detailed descriptive statistics for the study scales are presented in Table 2.

We found a statistically significant positive correlation between technical advantages factor and acceptance of AI (r = 0.533, p-value < 0.001). Moreover, we found a statistically significant negative correlation between technical advantages factor and fear of AI (r = -0.148, p-value = 0.014). A positive correlation between personal advantages factor and acceptance of AI was identified (r = 0.537, p-value < 0.001). Detailed correlation results are presented in Table 3.

3.3. Dependent Variable: Artificial Intelligence in Mental Health Scale

Table 4 displays the findings from the linear regression analysis, wherein the "technical advantages" dimension of the AIMHS served as the dependent variable. In the final multivariable model, both older age (b = 0.011, 95% CI: 0.002–0.020, p-value = 0.018) and increased daily engagement with social media or websites (b = 0.058, 95% CI: 0.021–0.095, p-value = 0.002) were significantly associated with more favorable perceptions of AI-based mental health chatbots. Diagnostic checks confirmed that the model met the necessary assumptions: residuals were normally distributed (see Supplementary Figure 1), homoscedasticity and linearity were evident (see Supplementary Figure 2), and variance inflation factors (VIFs) revealed no multicollinearity among the independent variables.

Table 5 presents the results of the linear regression analysis, with the personal advantages dimension of the AIMHS serving as the dependent variable. In the final multivariable model, male nurses exhibited significantly more favorable attitudes toward AI-based mental health chatbots in terms of perceived personal benefits (b = 0.548, 95% CI: 0.277–0.820, p-value < 0.001). Diagnostic assessments confirmed that the assumptions underlying linear regression were satisfied: residuals were normally distributed (see Supplementary Figure 3), homoscedasticity and linearity were observed (see Supplementary Figure 4), and VIFs suggested the absence of multicollinearity among the independent variables.

3.4. Dependent Variable: Attitudes Toward Artificial Intelligence Scale

Table 6 displays the results of the linear regression analysis, with the acceptance dimension of the ATAI serving as the dependent variable. In the final multivariable model, higher levels of digital technology competence were significantly associated with greater acceptance of artificial intelligence (b = 0.164, 95% CI: 0.014–0.313, p = 0.032). Additionally, male nurses reported significantly higher acceptance of AI compared to their female counterparts (b = 1.587, 95% CI: 0.993–2.181, p < 0.001). Model diagnostics indicated that the assumptions of linear regression were adequately met: residuals were normally distributed (see Supplementary Figure 5), homoscedasticity and linearity were observed (see Supplementary Figure 6), and VIFs confirmed the absence of multicollinearity among the independent variables.

Table 7 outlines the findings of the linear regression analysis, with the fear dimension of the ATAI utilized as the dependent variable. In the final multivariable model, lower financial status was significantly associated with heightened fear of artificial intelligence (b = -0.329, 95% CI: -0.467 to -0.190, p < 0.001). Furthermore, male nurses reported significantly greater levels of fear toward AI compared to their female counterparts (b = 0.772, 95% CI: 0.231–1.312, p = 0.005). Diagnostic evaluations confirmed that the assumptions of linear regression were satisfactorily met: residuals exhibited a normal distribution (see Supplementary Figure 7), homoscedasticity and linearity were observed (see Supplementary Figure 8), and VIFs demonstrated no evidence of multicollinearity among the independent variables.

4. Discussion

This study aimed to identify the nurses’ perceptions regarding the use of artificial intelligence driven mental health chatbots. Also, the findings from this study highlight the levels of acceptance and the fear toward AI integration in clinical practice. Also, our findings offer several insights into the relationships among nurses’ familiarity with AI, their beliefs, their affective attitudes and their behavioral intention among mental health support.

A key implication of this research is that the perceptions of AI-driven mental health chatbots varied widely across participants, but overall nurses present positive attitudes among the utilization of AI-driven chatbots. Generally, nurses did not recognize strong technical advantages of AI technology, but they reported more personal benefits among the use of AI mental health chatbots, although they pointed out a considerable amount of fear. As illustrated by results, nurses who perceive more technical advantages of artificial intelligence, feel more comfortable and less fearful to accept AI applications. Also, nurses who see personal benefits from the use of AI chatbots, tend to experience higher fear which means that although they recognize personal gains, they still feel anxious about AI. These findings align with previous research which suggest that experience with technology influence the adoption and acceptance of AI in healthcare [32,33]. Due to the fact that, nurses perceive many advantages by the use of artificial intelligence, they remain wary of their shortcoming [34]. AI-conversational agents have the potential to become a useful tool in nursing tasks, but many practical factors such as lack of knowledge, explainability, ethical issues related to nursing profession are still pending. Previous inquiry highlights the ability of large language models to assess the prognosis of schizophrenia and found that they aligned closely with the predictions of mental health professionals [35]. Overall, nurses need to have an attitude of openness toward learning new technologies and integrating appropriate culture into artificial intelligence use.

Another important result of our survey was that participants who receive more technical advantages of AI tend to have higher acceptance of AI. In other words, as belief in technical benefits increases, the acceptance increases. Nurses who admitted more technical advantages in AI-driven mental health technology, they tend to experience slightly less fear of AI. In other words, nurses who recognized the meaningful technical contribution of AI to the nursing profession, were more inclined to trust AI tools. These results are consistent with prior research, which emphasize the role of beliefs to behavioral intention [36,37,38], leading to a common conclusion with those of previous studies. The existing literature underscores AI’s expanding role in nursing to ensure that intelligent nursing evolves in alignment with clinical routines [39,40]. AI’s strong prospects as a nursing support tool in assisting decision-making, writing nursing documents, nursing operations with exposure risks and physical exertion, carrying out many nursing activities [39]. Indeed, health care professionals are optimistic about AI’s potential to improve decision-making safety and quality, literature emphasizes that the human touch remains essential for patients with complex needs [41].

Results from linear regression analysis support the notion that the older participants, the male gender and more time daily engagement on social media or websites tend to have more positive attitudes toward AI mental health chatbots. Furthermore, higher levels of digital technology competence were significantly associated with greater acceptance of artificial intelligence. Additionally, male nurses reported significantly higher acceptance of AI compared to their female counterparts, while they reported significantly greater levels of fear. Consistent with previous literature [36], in which female nurses reported lower levels of familiarity with AI, stronger beliefs about AI’ role in their job and higher levels of anxiety toward AI compared to their male counterparts. Additionally, previous literacy was significantly associated with this study dealing with the acceptance to use AI [32,45]. As indicated in literature, neonatal nurses can utilize generative artificial intelligence in clinical practice, but its effectiveness depends on structured training, reliable infrastructure and culturally sensitive implementation [45]. Notably, nurses’ perceptions of AI’s relevance to their role and their trust in AI-driven technologies were significant predictors of their intention to integrate AI into clinical practice. Nurses who recognized the meaningful contribution of AI to the nursing profession, trust its outputs and are more inclined to adopt AI tools [37]. These are consistent to prior research which highlight the role of beliefs as a cognitive antecedents to behavioral intention [42,43,44]. According to Schaivo et al. [46], literacy fosters a positive attitude toward acceptance, on the other hand anxiety has a significant, direct negative effect. Moreover, they found that learning and sociotechnical dimensions of AI anxiety serve as a complementary partial mediator between AI literacy and acceptance. Moreover, previous study illustrates the emergency of AI technologies in mental health care. Mental health professionals have generally been slower in the adoption of AI [32].

AI-driven conversational technology can provide to clinicians advanced tools for understanding and addressing complex behavioral health patterns. Moreover, through realistic stimulations and adaptive technologies, facilitate data driven monitoring for patient progress, collaborative treatments, planning and timely adjustment to interventions [48].

Our study underscores the perceptions and readiness of health care professionals for the transformative role of AI-driven conversational technology in health care, particularly in mental heath care. Previous literature emphasis the potential of mental health chatbots to improve patient care accessibility and patient management [49]. Moreover, the administrative role of AI chatbots aligns with the growing need for patient engagement in health care and resource optimization in health care settings. AI chatbots are capable to revolutionize mental health care though enhanced accessibility and personalized interventions [50]. This approach allows the individualization of care, enhances cultural sensitivity, optimize outcomes contributing to a more responsive and effective mental health care [48]. However, careful consideration of the ethical implications and methodological concept under cultural adaptions are essential to ensure the responsible deployment of AI-driven technology [51,52]. A conscientious and ethical integration of generative AI techniques is an obligatory, ensuring a balanced approach that maximizes the benefits in mental health practices [47,53].

To the best of our knowledge, this study is the first study that examines nurses’ perceptions regarding the use of AI chatbots as a tool for mental health support. Nonetheless, several limitations should be addressed. First, the cross-sectional design of our study limits the ability to draw causal inferences. Thus, we cannot determine the temporal sequence of the observed relationships. Future longitudinal research designs must be adherend to track nurses’ changes in beliefs, attitudes and behavioral intentions. Second, this study was conducted in one country with a hybrid health care system (both public and private), so the generalizability of the findings is limited. Future studies could replicate this research in different countries and health care systems. Third, further research is needed to explore other potential factors influencing AI adoption on organizational and environmental basis to dive into the core barriers of AI integration into nursing practice.

5. Conclusions

This study leverages insights for policymakers and healthcare leaders to ensure artificial intelligence integration into nursing practice, maintaining patient safety and patient-centered. There is an indisputable need to strengthen policy, education and resource allocation to support the sustainable integration of AI into clinical care. Notably, strictly requirements must be considered for AI adoption, including patients’ perceptions, time efficiency, preservation of clinician and patient autonomy. Many studies promote the design of relevant and sustainable education programs to support the adoption of AI within mental health framework [48,54,55]. However, their effectiveness depends on many key issues, including ethical AI practices, security of patient data, and smooth integration into existing health care infrastructures [49]. Future initiatives should prioritize the responsible use of AI and the development of inclusive and accessible mental health chatbots supporting fair access of all diverse populations to health care.