The IAGEI broadly refers to an individual’s psychological tendency and behavioral readiness to use generative AI tools (such as large language models) to search for, consult, and accept exercise-related guidance, plans, or suggestions generated by the AI (19). In this study, this concept is operationalized as a multidimensional construct. It not only refers to the frequency at which college students use such technologies but also, more importantly, encompasses the breadth of the types of exercise information they seek (e.g., personalized plans, movement correction, nutritional advice, etc.), as well as their perceptions of and trust in the quality of the information they receive, ultimately reflecting their intrinsic willingness to continue using and recommend the tool to others (20, 21).

Critically, IAGEI pertains specifically to engagement with generative AI, which represents a qualitative shift from traditional digital exercise platforms (22). Unlike static information repositories, rule-based fitness applications, or algorithmic recommenders operating within fixed parameters, generative AI introduces distinct intervention mechanisms (23). Its core affordances include open-ended dialogic interaction, allowing for multi-turn conversations that understand nuanced user constraints; adaptive reasoning and content generation, synthesizing contextualized plans rather than retrieving static information; and the capacity to simulate key coaching functions such as empathetic encouragement and collaborative goal-setting (24). Therefore, IAGEI captures the intention to engage with a tool that functions as an interactive, reasoning partner, making it a theoretically distinct precursor to behavior change compared to intentions toward more passive or rigid digital tools (25).

According to social cognitive theory, a positive intention to adopt information often indicates a deeper process of interaction and integration with the information (26), whereas, according to the theory of planned behavior, a firm behavioral intention is a key antecedent of actual behavior (27). Existing studies have shown that convenient, accurate, and high-quality exercise information support can enhance an individual’s exercise self-efficacy, reduce cognitive and behavioral barriers to starting and maintaining exercise, and thus positively impact sustained regular exercise (28–30). Compared with traditional digital exercise information sources such as fixed-function fitness apps or pre-recorded online tutorials, generative AI can engage in open-ended natural language interaction, flexibly adjusting the content, difficulty, and framing of its suggestions in response to users’ continuous feedback in real time (31). This dialogic, co-constructed form of guidance allows users to negotiate exercise goals, constraints, and preferences with the system, which may strengthen their perceived behavioral control and intention in line with the theory of planned behavior, beyond what is typically achieved by one-way information delivery channels (32). At the same time, the capacity of generative AI to simulate the functions of a virtual coach—for example by offering encouragement, reframing setbacks, and iteratively refining exercise plans—may create a qualitatively different motivational climate from that of static information platforms, thereby making it a potentially novel type of technology-enabled intervention rather than merely a new delivery medium for existing content (33). Based on this, this study hypothesizes that IAGEI will positively predict EA.

This study is principally grounded in self-determination theory (34), which provides the core mechanism—autonomous motivation—for understanding sustained behavioral engagement. AM, derived from self-determination theory (35), refers to the psychological quality of engaging in a behavior for intrinsic reasons, such as interest, enjoyment, or value alignment, rather than external pressure or rewards. It is a key predictor of behavioral persistence (36). On one hand, according to the “basic psychological needs” sub-theory within self-determination theory, individuals have fundamental psychological needs for autonomy, competence, and relatedness (34). Generative AI, by providing highly personalized and interactive exercise information, can directly support users’ autonomy feeling that the behavior is self-chosen and competence feeling capable of effectively executing plans (34, 37). The satisfaction of these needs is a crucial situational factor that promotes the internalization of external motivation and enhances levels of AM (38). Compared with conventional digital tools that provide pre-set exercise plans, generative AI can engage in iterative question–answer cycles in which students explain their situational constraints, receive tailored suggestions, and immediately ask follow-up questions, thereby allowing them to experience a stronger sense of authorship over the final plan (39). From a self-determination theory perspective, this co-creation process may be especially conducive to autonomy-supportive communication and to the internalization of exercise goals, which helps explain why the present model emphasizes intention to adopt generative AI, rather than generic exposure to digital exercise content, as the starting point (40). On the other hand, both self-determination theory itself and a wealth of empirical research have clearly pointed out that compared to controlled motivation (35), AM can significantly predict the long-term adherence to various health behaviors, as it is closely linked to higher enjoyment of behavior, stronger psychological vitality, and greater perseverance in the face of setbacks (41). IAGEI may create an information environment that supports basic psychological needs, which in turn nourishes individuals’ AM, and this high-quality motivation drives sustained exercise behavior (42). Therefore, this study hypothesizes that AM plays a significant mediating role in the relationship between IAGEI and EA.

To complement conventional variable-centered analyses and uncover the complex system of interdependencies among specific psychological and behavioral dimensions, this study employs network analysis (43, 44). This method moves beyond examining relationships between latent constructs to model the web of pairwise associations among observable dimensions (45, 46). Its value lies in its capacity to reveal which specific facets of constructs (e.g., which dimension of motivation or adherence) are most centrally connected within the observed system, offering a nuanced map of the psychological-behavioral architecture (47). Building on this, we construct a network model among the dimensions of IAGEI, AM, and EA. The goal is not to infer causality but to achieve a more granular understanding: First, to reveal which specific IAGEI characteristics show the strongest conditional associations with which types of AM and EA dimensions (48); Second, to exploratorily identify nodes (dimensions) with high connective influence, which may represent promising focal points within the associative network for future hypothesis-driven research (49). Thus, the network analysis serves as a valuable exploratory supplement (50), providing a detailed visualization of the systemic interrelations that can inform future theoretical refinement and intervention design (51).

Figure 1 illustrates the conceptual model depicting the hypothesized relationships among these variables.

From April to June 2025, an online survey was conducted using stratified random cluster sampling. One undergraduate university was randomly selected from each of four provinces: Guizhou, Jiangxi, Guangdong, and Henan. In each university, four classes were randomly selected at each grade level, totaling 48 classes and 48 classes of students as survey participants. All students in the selected classes were invited to participate in the online questionnaire survey. All participants provided informed consent and voluntarily agreed to participate. The inclusion criteria for participants were: (1) ability to understand and cooperate with the survey; (2) voluntary participation in the study. Exclusion criteria included: (1) incomplete questionnaire responses (less than 80% completion); (2) refusal to participate in the survey. A total of 1,968 questionnaires were collected. After removing invalid questionnaires due to patterned responses and failure to pass lie-detection questions, the missing values in the remaining data were handled using multiple imputation via chained equations (MICE) (52). The imputation was conducted using the mice package in R. Ultimately, 1,878 valid questionnaires were obtained, yielding an effective response rate of 95.4%.

Among the 1,878 college students included in the analysis (as shown in Table 1), the distribution of their demographic characteristics is as follows: 752 males (40.0%) and 1,126 females (60.0%); grade distribution included 537 first-year students (28.6%), 576 s-year students (30.7%), 414 third-year students (22.0%), and 351 fourth-year students (18.7%); regarding their residence, 795 students (42.3%) lived in rural areas, while 1,083 students (57.7%) lived in urban areas.

SPSS 27.0 was used to conduct item discrimination tests, item-total correlation analysis, and exploratory factor analysis on the 418 pre-survey samples. Subsequently, AMOS 26.0 was used to perform a confirmatory factor analysis to assess the structural validity and reliability of the IAGEI scale. In the formal survey, 1,878 valid questionnaires were analyzed using SPSS 27.0 with the PROCESS macro (model 4) for mediation analysis, with gender, age, grade, and residence as covariates. The network analysis was performed in R 4.3.2, using a mixed graphical model for estimation and applying the EBICglasso algorithm for regularization to streamline the network structure. The hyperparameter γ for the extended Bayesian information criterion (EBIC) was set to 0.5. For node centrality evaluation, the expected influence was chosen as the measure of node impact, as it shows good stability across mixed-polarity networks. Finally, the Bootstrap sampling method was used for 1,000 iterations to calculate 95% confidence intervals for edge weights to assess estimation accuracy and test the stability of centrality indicators. The acceptable stability coefficient (CS) was set to CS > 0.25, with CS > 0.5 indicating good stability.

This study used Harman’s single-factor test to analyze common method bias. The results showed that 11 factors had eigenvalues greater than 1, with the first factor accounting for 21.24% of the total variance. Since this value is below the 40% threshold, it indicates that there is no significant standard method bias in the data of this study.

A correlation analysis was conducted on the means of IAGEI, EA, and AM. The results showed that IAGEI, EA, and AM were positively correlated with each other (as shown in Table 2).

Using SPSS 27.0 and PROCESS 4.1 macro (Model 4), IAGEI was set as the independent variable, EA as the dependent variable, and AM as the mediator to test the mediation effect of AM. Gender, age, grade, and residence were included as covariates in the model. The results (Table 3) showed that IAGEI was significantly positively correlated with AM (β = 0.362, p < 0.001), and AM also significantly predicted EA (β = 0.320, p < 0.001). After including AM in the model, IAGEI still showed a positive association with EA (β = 0.299, p < 0.001). The Bootstrap test showed that the indirect effect of AM was 0.116, with a 95% confidence interval that did not include zero, indicating a statistically significant association consistent with mediation. This cross-sectional finding supports the hypothesized theoretical model in which AM plays a mediating role. As shown in Table 4. This effect accounted for 27.78% of the total effect.

This study used network analysis to explore the associative structure between the dimensions of IAGEI, AM, and EA. The network estimation results showed that the network contained 54 non-zero edges, with a network density of 0.98, and a non-zero edge proportion of 98.18%, indicating a highly interconnected system among the observed variables. In the network structure (as shown in Figure 2), the strongest connections within dimensions were found in the following relationships: EA’s “Behavioral Habits (BH)” and “Emotional Experience (EI)” were most strongly related (r = 0.252), followed by IAGEI’s “Attitude toward Use (ATU)” and “Adoption Intention (ADI)” (r = 0.250). In terms of cross-dimensional associations, two key connections are noteworthy: EA’s “Emotional Experience (AE)” and AM’s “Intrinsic Motivation (IM)” showed a significant positive correlation (r = 0.085), and IAGEI’s “Perceived Fit (PF)” and EA’s “Emotional Experience (AE)” also showed a close connection (r = 0.080). The network analysis results suggest that positive emotional experiences are not only strongly related to intrinsic motivation but are also significantly influenced by the degree of personalized information fit, forming a complex interaction network among the variables. Detailed values of all edges in the cross-sectional network are provided in Supplementary Table S5.

The centrality analysis results (Figure 3) show that Intrinsic Motivation (IM) under the AM dimension (EI = 1.537) and Behavioral Habit (BH) under the EA dimension (EI = 1.245) are the two nodes with the highest expected influence values. This suggests that motivation driven by intrinsic interest and enjoyment, as well as a regularized exercise behavior pattern, may play the most central driving roles in the entire psychological-behavioral system, having the most substantial overall influence on other variables. Supplementary Table S6 provides detailed centrality indicators for all nodes at two time points.

To assess the reliability of the estimated network, this study used the Bootstrap sampling method (1,000 iterations) to test the stability of the network structure and centrality indicators. The results showed that the stability coefficients (CS-coefficient) for edge weights and shortest path distances were both 0.75, indicating that the estimation of network connectivity strength is very stable. Regarding node centrality indicators, the stability coefficients for expected influence and strength were both 0.361, which are above the acceptable threshold of 0.25, suggesting that the centrality ranking of core nodes is reliable. Overall, the network model in this study demonstrates good stability in estimating edge weights and key influential nodes, supporting subsequent analysis and discussion. The Bootstrap confidence intervals for edge weights and centrality difference test results are provided in the Supplementary materials.

This study confirmed that IAGEI has a significant positive correlation with EA. This result is consistent with the findings of a cross-sectional study (58). The theory of planned behavior provides a theoretical foundation for understanding this relationship (27), emphasizing that behavioral intention is the most direct predictor of actual behavior (59). Once college students form a stable IAGEI, their psychological readiness to incorporate the technological tool into their exercise planning may transform into sustained information inquiry and plan execution, ultimately enhancing the regularity and persistence of exercise behaviors (60). At the mechanism level, generative AI, by providing highly personalized and instant-feedback exercise plans, effectively addresses the information asymmetry and execution uncertainty that college students face during exercise (61–63). Unlike traditional fixed-mode training plans, AI can dynamically adjust content based on the individual’s real-time status (64). This precise adaptation feature significantly improves the feasibility and acceptance of the exercise plan, further promoting the formation of exercise habits by enhancing self-efficacy (65). Recent research has highlighted the growing importance of AI in promoting physical activity, particularly in improving mental health. For example, a comprehensive review has demonstrated significant evidence supporting the mental health benefits of sustained physical activity, which aligns with our findings on exercise adherence (66). This reinforces the idea that AI-driven interventions could be instrumental in enhancing both physical and mental health by providing personalized, adaptive exercise plans.

Additionally, the increasing role of AI in providing personalized guidance is discussed in recent studies on explainable AI in clinical settings. While these studies primarily focus on clinical decision support, they offer valuable insights into how AI can facilitate user trust and deliver tailored advice (67). These insights are directly applicable to the domain of exercise adherence, where personalized AI-driven solutions are critical for encouraging sustained engagement and adherence to exercise routines. In the context of Chinese higher education, this mechanism is particularly prominent. Faced with intense academic pressure and fragmented time, traditional exercise modes are often challenging to execute effectively (68). Generative AI, with its 24/7 availability and personalized customization, meets the dual needs of flexibility and efficiency among contemporary college students (69). It offers a feasible technological solution for maintaining regular exercise under limited time and space conditions, reflecting the practical value of the Healthy China strategy powered by digital technologies in the campus environment (70–72).

Consistent with our primary theoretical framework (SDT), the cross-sectional mediation analysis indicated a significant associative pattern in which AM links IAGEI and EA. An SDT-informed explanation involves two interrelated phases: motivation formation and behavior maintenance. Generative AI creates an information environment that supports autonomy and competence needs by providing highly personalized exercise plans (73). When students perceive that the AI system can offer “tailored” training recommendations based on their physical fitness level, equipment conditions, and time constraints (74, 75), this precise fit not only enhances the tool’s perceived value but also promotes the internalization of external motivation by respecting individual choice and boosting confidence in execution, thus creating the necessary conditions for the development of AM (76–78).

In the behavior maintenance phase, AM’s promotion of EA reflects the sustained effect of high-quality motivation. Unlike controlled motivation driven by external rewards or punishments, AM, rooted in interest identification and value internalization, can trigger more positive emotional experiences and greater behavioral resilience (79). In the context of Chinese higher education, amid anxiety over credit pressure and social comparisons, exercise driven by AM is more likely to maintain stability because it aligns with students’ intrinsic needs for self-development rather than becoming an additional burden (80, 81). This associative pattern suggests that generative AI may not only be directly linked to exercise behavior by providing information support but, more importantly, fosters high-quality motivation through creating a needs-supportive environment, thus providing intrinsic motivation for behavior persistence (82). This mechanism offers a new perspective for understanding the deeper role of intelligent technologies in promoting health behaviors. Future campus health promotion should focus on cultivating students’ autonomous identification with exercise through technological means, rather than relying solely on external incentives (83, 84).

The exploratory network analysis provided a detailed map of the associative relationships between the dimensions of IAGEI, AM, and EA. The high connectivity (network density = 0.98) observed in the network model indicates a tightly interrelated system, which may reflect the inherent synergy between these psychological and behavioral constructs in the context of exercise. This systemic perspective complements the variable-centered mediation analysis. The strong internal connections within dimensions reveal the inherent consistency of each psychological construct. The close relationship between EA’s “Behavioral Habit (BH)” and “Affective Experience (AE)” supports the core idea of habit formation theory (85). As exercise behaviors become automated through repeated execution, the conscious effort required diminishes, and the accompanying positive emotional experience further reinforces the behavior’s sustainability (86). This finding is particularly significant among Chinese college students—faced with heavy academic pressures—solidifying exercise into a daily habit that requires no extra willpower and deriving emotional pleasure from it becomes a key factor in maintaining long-term exercise (87, 88). Similarly, the strong connection between “Attitude toward Use (ATU)” and “Adoption Intention (ADI)” in IAGEI supports the basic assumption of the Technology Acceptance Model, where an individual’s positive evaluation of technology directly translates into a willingness to use it (89). Among the cross-dimensional connections, the association between AE and Intrinsic Motivation (IM) enriches self-determination theory by highlighting the interaction between motivation and emotion (35). Positive emotional experiences provide an emotional foundation for cultivating IM by satisfying basic psychological needs, and this mechanism is particularly significant in the context of Chinese culture, which emphasizes the “integration of body and mind” in health concepts (90). At the same time, the connection between Perceived Fit (PF) and AE (r = 0.080) suggests that the personalized content provided by generative AI is not only practically valuable but also capable of eliciting positive emotional responses by precisely matching individual needs, offering new evidence for understanding the emotional value of technological fit (91).

The results of the exploratory network centrality analysis show that Intrinsic Motivation (IM) and Behavioral Habit (BH) exhibit the most substantial expected influence (EI values of 1.537 and 1.245, respectively), suggesting they may be influential nodes within the observed associative network. The observed high centrality of IM is consistent with the core idea of self-determination theory (35). As the purest form of AM, IM provides lasting intrinsic motivation for behavior persistence by fulfilling an individual’s needs for autonomy, competence, and relatedness (92). Its position in this cross-sectional network suggests it as a potential candidate for exercise motivation derived from interest and enjoyment, which not only directly influences behavior persistence but also generates ripple effects by activating other psychological processes, such as enhancing emotional experience and promoting cognitive evaluation (93, 94). The high centrality of BH reflects the importance of habit formation theory in EA research. When exercise behavior becomes automated through repeated execution, its demand for cognitive resources decreases, which is particularly valuable in the high cognitive load and fragmented time context of Chinese college students (95). The formation of habits not only directly sustains exercise frequency but also establishes a “behavior-emotion” positive feedback loop through its strong connection to emotional experience, a mechanism particularly prominent in Chinese culture, which emphasizes the integration of knowledge and action (96). In the context of Chinese higher education, the well-known positions of these two core nodes hold significant practical implications. Facing heavy academic pressure and a competitive environment, the primary challenges to college students’ EA have shifted from a lack of knowledge to a lack of motivation and difficulty forming habits (97). IM, by satisfying students’ needs for autonomy and competence, provides intrinsic motivation for EA, while BH helps students overcome willpower depletion through its automation mechanism. Together, they constitute the key psychological resources for addressing the unique challenges faced by Chinese college students.

This study makes threefold contributions at the theoretical level: First, it successfully integrates the fields of technology acceptance and health behavior by constructing and validating a theoretical model in which IAGEI serves as the starting point, and AM mediates the process, ultimately influencing EA. This model provides a novel and comprehensive theoretical framework for understanding how digital technologies empower health behavior promotion (98). Second, the study introduces network analysis, an advanced method that goes beyond the traditional variable-centered paradigm. It visually reveals the central roles of Intrinsic Motivation (IM) and Behavioral Habit (BH) within the psychological-behavioral system, confirming the core propositions of self-determination theory and habit formation theory and deepening our understanding of the complex interactions among variables from a system perspective (35). Lastly, this study elevates the role of generative AI from an “information tool” to a “motivation catalyst,” uncovering the deep mechanism through which it cultivates AM by fulfilling basic psychological needs (99). This significantly enriches and expands the application of self-determination theory in the context of intelligent human-machine interactions.

At the practical level, the findings of this study provide a clear roadmap for using artificial intelligence to promote physical exercise among college students in the context of Chinese higher education. The study suggests that educators and health promoters should not merely view generative AI as a “broadcast station” for delivering training plans, but should leverage its personalized and interactive features to transform it into a “motivational partner” that supports students’ autonomous decision-making and enhances their sense of competence, thereby shifting exercise behavior from an external requirement to an intrinsic need (100). More importantly, the core intervention targets identified through network analysis—Intrinsic Motivation (IM) and Behavioral Habit (BH)—guide the design of efficient and cost-effective precision interventions. Practical efforts should focus on creating sports activities that stimulate students’ intrinsic interest and on using AI to support habit-forming training, thereby achieving a “small effort with big results” in promoting exercise. This provides concrete empirical evidence for implementing the Healthy China strategy and for the deep integration and innovative application of digital technologies in campus sports.