A total of 982 full-time undergraduate nursing students from a particular university were selected as research participants. Among them, 117 were male, and 865 were female, with ages ranging from 18 to 24 years (mean age 21.18, SD 0.51 years). The inclusion criteria were (1) full-time undergraduate nursing students enrolled at the university; (2) students who had encountered or used AI-related technologies or tools in their daily studies or lives; and (3) voluntary participation in the experiment with signed informed consent and confidentiality agreements. Exclusion criteria applied to students on formal leave or extended sick leave during data collection. Participants could withdraw freely without affecting their studies or academic performance.

The research procedures have been approved by the Ethics Review Committee of Xijing University (Approval number: XJU-202502). Participants’ privacy has been protected. All participants provided written informed consent. No harm was inflicted upon participants. All nursing students had the right to choose whether to participate in this study. Data were securely stored and encrypted.

Guided by the theoretical framework of NUR.S.E.S., this study establishes conceptual mappings between the 6 core competency dimensions of the framework (navigate, use, recognize, skills support, ethics in action, and shape the future) and their corresponding measurement variables. This mapping aims to structure the AS of nursing students as defined by the framework into measurable psychological and competency indicators, thereby providing a theoretical basis for subsequent network analysis and testing the intrinsic interrelationships among the framework’s components.

Researchers (XF and JW) contacted class administrators at target institutions via social media (WeChat; Tencent Holdings Limited). After explaining the study objectives and procedures and obtaining permission, they joined the corresponding class groups. Study information was disseminated through these class groups. Potential participants voluntarily confirmed their participation online after reading the informed consent form. The research team screened applicants based on predetermined inclusion and exclusion criteria. Students meeting the criteria were invited to participate in the survey. To facilitate organization and enhance response efficiency, researchers distributed electronic questionnaire links uniformly in classrooms during after-school hours. Data collection occurred from September 25 to September 28, 2025, using the online survey platform “WJX” for questionnaire distribution and retrieval. Participants independently completed the questionnaire after clicking the link and could exit at any time during the process. Completing the full questionnaire took an average of 10‐15 minutes. A total of 986 questionnaires were distributed, with 982 returned. Platform settings ensured only fully completed questionnaires were recorded as valid submissions, eliminating incomplete submissions from being counted. Verification confirmed that returned questionnaires were valid. Data from 982 participants were ultimately analyzed, yielding a valid response rate of 99.59%.

Descriptive statistics were calculated using SPSS (version 25.0; IBM Corp), with means and SDs for continuous variables and frequencies (percentages) for categorical variables. All continuous variables underwent normality testing in this study, and the results indicated that their distributions were generally normal [39].

Network analysis was performed using R (version 4.3.0; R Core Team). In 2-tailed tests, P<.05 was considered statistically significant. A Gaussian graph model was selected to construct the network structure using the R package “qgraph” [37]. To reduce spurious correlations and obtain a simplified network, the graph minimum absolute contraction and selection operator was used. The graph minimum absolute contraction and selection operator implements regularization techniques to constrain small partial correlations to zero, effectively eliminating potential false positive edges and yielding sparse, interpretable network structures [40]. To avoid overfitting and determine optimal tuning parameters, an extended Bayesian information criterion model is used to select the best-fitting model. This approach ensures that only the most robust and meaningful connections are retained in the final network.

In a network, nodes represent variables, and edges between 2 nodes represent the unique relationship between them, controlling for the influence of other nodes. Red and blue edges denote negative and positive correlations, respectively. The thickness of the edges corresponds to the strength of the association; thicker edges indicate stronger correlations [41]. The R package “qgraph” is used to calculate the strength, proximity, and betweenness centrality of nodes in a network [37]. Strength is defined as the sum of the absolute values of the edge weights connecting a node to all other nodes. Previous studies have shown that strength is the most stable and interpretable centrality measure in networks [42]. Compared to other centrality measures such as closeness and betweenness centrality, strength is more suitable for evaluating the importance of nodes [43]. Therefore, this study adopts strength as the centrality metric, with nodes exhibiting the highest strength values being regarded as the most influential nodes within the network.

Additionally, the R package “networktools” was used to calculate bridging strength and identify potential bridging nodes [44]. Bridging strength refers to the sum of the absolute values of the edge weights connecting a node to other community nodes, indicating the node’s importance to other communities. Nodes with the highest bridging strength values are considered bridging nodes within the network. Bridging nodes connect multiple domains, meaning interventions targeting these nodes can simultaneously impact multiple domains, which maximizes the effectiveness of the intervention.

Use the R package “mgm” to calculate the predictability for each node [45]. Predictability is a metric that reflects the extent to which the variance of a specific node can be explained by all its neighbors in the network. Nodes with high predictability are susceptible to the influence of their neighboring nodes.

The R package “bootnet” assesses the accuracy and stability of networks [46]. First, the accuracy of edge weights (the number of bootstrap samples is 1000) was examined by calculating their 95% CIs using nonparametric bootstrapping. Second, the stability of the network model is assessed using the case deletion procedure and the correlation stability (CS) coefficient, which indicates the proportion of samples that can be excluded from the analysis while maintaining at least 0.7 correlation with the original sample. According to network analysis guidelines, the CS coefficient should not fall below 0.25, and a value above 0.50 indicates sufficient stability [47]. Finally, a self-assessment method was used to conduct a difference test, examining variations in edge weights, node strengths, and node bridge strengths.

A total of 982 graduate nursing students were included in the statistical analysis, with their characteristics listed in Table 2. The mean age of participants was 21.7 years, and the vast majority (865/982, 88.1%) of participants were female. Among participants, 511 out of 982 (52.0%) resided in rural areas. The number of students in the first, second, third, and fourth years was 246 out of 982 (25.1%) participants, 242 out of 982 (24.6%) participants, 251 out of 982 (25.6%) participants, and 243 out of 982 (24.7%) participants, respectively. Among the 982 undergraduate nursing students, daily active use of GenAI tools was reported as follows: 22 out of 982 (2.2%) participants rarely used them, 608 out of 982 (61.9%) participants used them several times weekly, 297 out of 982 (30.2%) participants used them 3‐5 times daily, and 55 out of 982 (5.6%) participants used them more than 5 times daily.

Descriptive statistics for nursing undergraduates’ EI, AILS, and AS are presented in Table 3. Normality tests confirmed that all variables were approximately normally distributed. Data were described using mean (SD). SDs fell within normal ranges, with no outliers. Notably, “Emotional Perception (EI1)” demonstrated the highest predictive power and served as the critical component of the entire network.

Figure 1 illustrates the network structure of EI, AILS, and AS among undergraduate nursing students. Node colors represent theoretical constructs: blue=AILS, green=EI, and red=AS. Letters on nodes denote dimension codes. Red indicates positive correlations; blue indicates negative correlations. Thicker lines denote stronger connections. Of the 66 possible connections, 54 (81.82%) were nonzero, and the network density was 0.82, indicating that the variables were generally closely interconnected within this sample. Among these, 41 were positive correlations and 13 were negative correlations. From the perspective of community structure, the nodes are primarily clustered into 3 distinct communities based on theoretical constructs, including emotional intelligence (EI1-EI4), AI self-efficacy (AILS1-AILS4), and AI competence (AS1-AS4). Connections within each community are relatively dense, while cross-community connections are relatively sparse, which is largely consistent with the dimensional divisions of the NURSES framework. Stronger connections primarily occurred within dimensions rather than between them. The 3 strongest intradimensional connections were EI4 (emotional use)-EI3 (managing others’ emotions), AS3 (assessment)-AS1 (awareness), and AILS2 (anthropomorphic interaction)-AILS1 (assistance). The strongest interdimensional connection was between EI1 (emotional perception) and AILS1 (assistance). All edge weights (association strengths) are listed in Table 4.

The standardized estimates of centrality indicators for factors influencing EI, AILS, and AS among nursing undergraduates are shown in Figure 2. Emotional use (EI4) and assessment (AS3) exhibit the highest strength centrality, indicating that these 2 variables showed the highest strength centrality, indicating they are highly connected within the network. Emotional perception (EI1) and ethics (AS4) exhibited the strongest mediating centrality, signifying their pivotal bridging role in network information transmission with potent moderating capacity; emotional perception (EI1) and awareness (AS1) also demonstrated the most prominent proximity centrality, reflecting their highest information transmission efficiency and accessibility within the network. Stability tests for network centrality metrics revealed correlation stability coefficients of 0.749, 0.749, and 0.361 for node strength, intermediary centrality, and closeness centrality, respectively (see Figure 3). This indicates robust stability for these centrality measures, with node strength demonstrating the highest stability.

The accuracy of edge weights was estimated using nonparametric bootstrapping. Results indicate that the 95% CIs for edge weights in the network are narrow, suggesting that the estimated edge weights in this study are highly precise (Figure 4).

Figure 5 displays the bridge strength for each node. EI1 “Emotion Awareness” (bridge strength=0.427) exhibits the highest bridge strength, followed by ALS3 “Comfort with AI” (bridge strength=0.242), ALS1 “Assistance” (bridge strength=0.129), and EI2 “Self-Emotion Management” (bridge strength=0.118), suggesting they may serve as bridging nodes in this network.

This study is the first to use network analysis methods within the NUR.S.E.S. framework to examine the associations among EI, AILS, and AS among undergraduate nursing students. The network analysis identified that emotion use (EI4) and assessment ability (AS3) showed the highest strength centrality, indicating their prominent positions within the network. Emotion perception (EI1) had the highest bridge strength, suggesting it may serve as a connecting node between the EI domain and the AILS and AS domains. These findings provide insights into the psychological mechanisms underlying the development of AS among nursing students, offer a concrete explanation of how the 6 dimensions of the NURSES framework interact at the psychological level, and provide a critical foundation for the future development of targeted, comprehensive interventions that address the full spectrum from emotional foundations to technical confidence.

This study conducted a descriptive analysis of nursing students’ EI, AILS, and AS. Results indicate that undergraduate nursing students’ AILS (mean 4.68, SD 0.78) and AS (mean 4.60, SD 0.59) were generally at a moderately high level, while EI (mean 3.57, SD 0.55) scored relatively low. This situation highlights a key phenomenon: nursing students possess strong confidence in their ability to learn and master AI technologies and have established a foundational level of AS. However, the development of their EI has not kept pace, consistent with the findings of Bewersdorff et al [38]. From a specific-dimension perspective, within AILS, the “Comfort Level with Artificial Intelligence (AILS3)” dimension scored the highest (mean 4.96, SD 0.98), indicating that nursing students experience relatively positive feelings when interacting with AI. This finding aligns with the research results of Volpato et al [48]. In terms of EI, the “Emotional perception (EI1)” dimension had the lowest mean score 3.41 (SD 0.54), suggesting that nursing students have significant room for improvement in their ability to accurately identify and understand their own and others’ emotions. This weakness may directly impact nursing students’ empathetic communication and emotional regulation in high-pressure clinical settings, posing a potential risk to patient safety; failure to promptly identify patients’ anxiety or distress may delay nursing decisions.

The network structure constructed in this study showed that emotional perception (EI1), defined as the ability to recognize one’s own and others’ emotional states, was strongly connected to AILS and AS nodes. In this sample, emotional perception had the highest bridge strength, indicating that it was relatively well-connected to nodes in other communities. This pattern is consistent with Fu’s [49] finding that emotional competence is associated with mental health and with deeper engagement with AI in educational contexts. The high bridge strength suggests that the ability to recognize emotions may be associated with trust in AI technology, which aligns with previous research showing that individuals with higher EI tend to monitor emotional cues and manage stress effectively [50]. Specifically, the positive connection between emotional perception and the assistance dimension (AILS1) indicated that students who reported greater awareness of their own confusion, anxiety, or sense of accomplishment during technology learning were also more likely to view AI as a helpful tool [51]. This pattern is consistent with the observation by Mosleh et al [35]. that AI integration in teaching creates opportunities for emotional recognition and regulation, which may be easier for students with stronger emotional perception.

The centrality measures showed that emotional use (EI4) and assessment (AS3) had the highest strength centrality, indicating that they were highly connected nodes within the network [52]. This pattern suggests that AS is associated not only with cognitive factors but also with emotional factors. Emotional use, defined as the ability to use emotional information for planning, creativity, and problem-solving, was positively connected with critical evaluation of AI outputs [13]. This finding is consistent with the observation that some individuals turn to GenAI for emotional guidance [50] and that emotional use may help nursing students engage in critical thinking when faced with complex AI outputs [53]. Evaluation ability, which concerns the judgment of AI-generated content [33], was closely connected with emotional use in the network. This pattern aligns with the finding by Deng and Chen [19] that AI-driven environments can support cognitive flexibility and independent judgment, possibly through emotional resilience. These exploratory observations invite further investigation using longitudinal or experimental designs to examine whether and how emotional use and evaluation skills might be related to AS [54].

Network analysis results showed that EI, represented by emotional perception (EI1), had a positive connection with AILS, represented by assistance (AILS1) and comfort with AI (AILS3). For example, the edge weight between EI1 and AILS1 was 0.22. AILS also showed positive connections with AS, represented by usage (AS2) and evaluation (AS3). This pathway pattern, in which EI connected to AILS and AILS in turn connected to AS, is consistent with a mediating structure [55]. However, due to the cross-sectional design, this pattern should be interpreted as exploratory. It suggests that AILS may occupy a bridging position, but formal mediation analysis using longitudinal data would be required to test this hypothesis [49]. Prior research has shown that EI is associated with self-efficacy and with cognitive engagement with AI [49]. The connection pattern observed here is compatible with the idea that confidence in technical skills (AILS4) is associated with more frequent use of AI tools (AS2) [56]; that comfort with AI (AILS3) and anthropomorphic interaction (AILS2) are associated with lower anxiety and more exploration [25]; and that the belief that AI serves as an effective assistant (AILS1) is associated with reduced operational anxiety [53]. These observations are hypothesis-generating and require replication in independent samples with stronger designs. The significance of this approach for clinical practice lies in the fact that only by simultaneously strengthening both the emotional foundation and technical confidence can knowledge be translated into safe and effective clinical practice.

The most significant negative correlation was found between “Ethics (AS4)” and “Usage (AS2)” in the network. This indicates that at the current stage, nursing students who are more concerned about AI ethical issues (such as data privacy) may be more cautious in their use of AI tools, even exhibiting lower usage frequency [57]. Although this contradicts findings showing positive correlations between EI and reasons for adopting AI services or adoption/usage intentions [58], it reflects the development of critical thinking. When nursing students recognize the potential risks of technology, they may transition from blind tool users to critical thinkers, adopting more rational, selective use behaviors [59]. This finding resonates with the perspective of Ricon [60] that premature or excessive enthusiasm in medical AI may obscure ethical concerns, while measured caution is a prerequisite for responsible innovation. Participants in one study expressed concerns about potential privacy breaches and the risk of personal data theft [61]. Therefore, educators should not pursue unrestricted usage frequency but instead focus on cultivating students’ high-level application skills grounded in strong ethical vigilance.

Second is the negative correlation between “Emotional Intelligence in Others (EI3)” and “Anthropomorphic Interaction (AILS2).” This negative correlation may reflect multiple mechanisms. First, individuals with high EI are able to clearly distinguish the fundamental differences between human emotional connections and AI-simulated interactions. The ability to manage others’ emotions manifests as the capacity to perceive, influence, and regulate others’ emotions in interpersonal interactions, whereas the tendency toward anthropomorphic interaction is a psychological inclination to ascribe human characteristics to AI and interact with it in a social manner. Nursing students with high EI have a deeper understanding that, although AI can simulate empathetic responses, it lacks genuine emotional intent and reciprocity. Consequently, they consciously avoid viewing AI as an emotional substitute to prevent the development of inappropriate emotional dependence, which is consistent with Dakakni and Safa [62]. Second, cultural factors may play a role. In the context of Eastern culture, which emphasizes interpersonal harmony and emotional connection, nursing education places a high value on “authentic interpersonal interaction.” Nursing students with high EI may have internalized these professional values, leading them to adopt a more cautious or even resistant attitude toward anthropomorphizing AI. Third, from the perspective of clinical practice, this negative correlation carries positive ethical implications. The core of nursing practice lies in establishing trust within therapeutic interpersonal relationships. If nursing students are overly inclined to anthropomorphize AI and blur the boundaries between humans and machines, it may diminish their sensitivity to the emotional needs of real patients. Therefore, the lower tendency toward anthropomorphic interaction exhibited by nursing students with high EI can be viewed as an adaptive strategy for upholding the core values of the nursing profession [63].

The findings of this study provide empirical support for the NUR.S.E.S. framework, revealing the underlying psychological mechanisms behind its 6 dimensions. Specifically, emotional perception (EI1), as the strongest bridge node, empirically demonstrates the central role of E (EI) and S (social-emotional comfort) as the emotional cornerstones within the framework [64]; emotional application (EI4) and assessment (AS3), as core driving nodes, correspond to E and R (defect identification), respectively, indicating that emotional competence and technical critical thinking share an intrinsic synergistic mechanism; the various dimensions of AILS occupy key positions in the network that connect emotion and cognition, validating the theoretical premise within the framework that S (AILS) serves as the core driving force [65]. Ultimately, the synergy of all these elements achieves the framework’s ultimate goal—S (shape the future)—that is, shaping the future [27].

In addition, this study provides an initial exploratory mapping of the interrelated factors associated with AS among nursing undergraduates, and these findings may offer insights for future research. From a practical perspective, nodes with high centrality, such as emotional use and evaluation, could exert influence on the overall network configuration through their connections with other nodes. Bridge nodes, such as emotional perception and comfort with AI, may transmit influences across different domains, making them potential targets for hypothesis-driven interventions. Accordingly, nursing educators might consider several tentative strategies, fostering students' emotional perception skills, which appear to link EI with AILS; creating low-anxiety AI learning environments to enhance comfort with AI; helping students set personalized learning goals and reflect on their emotional responses to strengthen emotional use and critical evaluation; and introducing case-based learning that combines clinical scenarios with AI outputs to ground technical critique and ethical decision-making in empathy for patients’ emotional needs. Such a multidimensional approach, addressing emotional, cognitive, and skill components simultaneously, may support the development of nursing students as future professionals who can shape AI integration in nursing practice. However, all suggestions require empirical validation through longitudinal or experimental designs.

This study is the first to visualize, through network analysis, the patterns of association among EI, AILS, and AS in nursing students. It identified nodes that showed high centrality and bridging strength (emotional perception, emotional application, evaluation, comfort with AI). The observed association pattern suggests that emotional competence may be closely linked with technological confidence, which in turn may be linked with AS. These exploratory findings offer preliminary clues for future hypothesis-driven research. However, they do not establish causality, nor do they directly identify effective intervention targets. Any educational implications drawn from these results are speculative and require empirical validation [56].

This study has several limitations. First, the cross-sectional design cannot establish causal relationships between variables. Although the network pattern is consistent with some causal pathways, a reverse pathway (eg, higher AS leading to greater AILS) is equally plausible. Future research should use longitudinal or intervention designs to test dynamic relationships. Second, the sample originated from a single university, limiting generalizability. Cross-regional and cross-cultural comparisons are needed. Third, self-report scales may introduce bias; future studies could incorporate behavioral data and qualitative interviews.

Despite these limitations, this study offers profound theoretical insights and precise practical guidance for cultivating a new generation of nursing professionals who are “technically proficient, confident, and emotionally intelligent” in the era of AI. Future education and research should focus on developing integrated intervention programs that combine emotional education with technical training, empowering nursing students to embrace the challenges of the intelligent health care era fully.