From AI coaching to athletic engagement: the mediating roles of cognitive adaptability and growth mindset in sports students Authors: Dongmin Zhao, Khalid Abdullah Alotaibi, Shazia Rehman, Xin Li Published: 2026 ABSTRACT BackgroundAI-supported coaching tools are increasingly incorporated into sport education, yet limited research has examined how students’ perceptions of these technologies relate to their cognitive and motivational functioning. This study investigates perceived AI-supported coaching and its relationship to athletic learning engagement among sports-major students, with cognitive adaptability and a growth mindset as mediators.MethodsA cross-sectional survey was conducted in March 2025 at universities in Dezhou, Shandong Province, that offer sports degree programs. The final analytic sample comprised 1,794 sports-major students. Validated Chinese measures assessed perceived AI-supported coaching, cognitive adaptability, growth mindset, and athletic learning engagement. Structural equation modeling with bootstrapped indirect effects was used to examine mediation pathways, and sensitivity analysis was conducted to assess the robustness of indirect effects across alternative model specifications.ResultsPerceived AI-supported coaching was positively associated with cognitive adaptability and, to a lesser extent, with engagement. Cognitive adaptability was positively associated with both growth mindset and engagement. The indirect effect of AI-supported coaching on engagement via cognitive adaptability was significant, as was the sequential pathway through cognitive adaptability and growth mindset. In contrast, the indirect pathway through growth mindset alone was not significant. Sensitivity analyses indicated that the pattern and significance of indirect effects remained stable across alternative model specifications, supporting the robustness of the mediation findings.ConclusionsThe findings suggest that cognitive adaptability serves as a meaningful intermediary linking perceived AI-supported coaching with students’ engagement in athletic learning, whereas growth mindset primarily operates within a sequential cognitive–motivational pathway. These results underscore the importance of considering both cognitive and belief-based processes when integrating AI-supported tools into sport education. Introduction AI-supported tools are becoming increasingly embedded in sport training and physical education, with applications spanning performance analytics, motion tracking, automated video tagging, and injury-risk monitoring [1]. Recent reviews highlight that artificial intelligence (AI) is now used across multiple domains in sport, including biomechanics, performance enhancement, and sports medicine, transforming how athletes and coaches access and interpret performance information [2–4] Within higher education, technology-enhanced pedagogy has also expanded rapidly in physical education, with evidence that digital tools can support motivation, tactical understanding, and cognitive learning when appropriately integrated into curricula [5, 6]. For university sports majors preparing for future roles as athletes, coaches, or physical education teachers, AI-supported coaching systems are becoming a salient part of their learning environment. Yet, despite this technological growth, most existing work has centered on technical or performance-related outcomes. Much less is known about how sports-major students perceive AI-supported coaching and how such perceptions relate to their internal cognitive and motivational processes. In particular, the psychological implications of AI-supported coaching for students’ learning engagement remain insufficiently theorized, representing a critical gap that the present study addresses. AI-supported coaching in sport refers to digital systems that provide automated or semi-automated performance feedback through tools such as video-analysis platforms, motion-tracking technologies, and machine–learning–based performance indicators [7]. These systems present learners with detailed information about movement patterns, tactical decisions, and performance errors, requiring them to interpret complex feedback and consider alternative pathways for improvement [8]. Research on technology-enhanced and AI-mediated learning environments shows that students who view such feedback as useful tend to engage more actively in evaluating errors, experimenting with new strategies, and adjusting their responses during practice [9, 10]. In sport education specifically, digital feedback systems have been associated with learners’ willingness to shift techniques, revise tactical choices, and respond adaptively to novel performance information [11, 12]. Together, these trends suggest that students who perceive AI-supported coaching more positively are more likely to demonstrate higher cognitive adaptability—characterized by flexible thinking and the capacity to revise strategies in response to new information. Accordingly, the first hypothesis proposes that perceived AI-supported coaching will be positively associated with cognitive adaptability. Cognitive adaptability refers to an individual’s capacity to shift strategies, reinterpret feedback, and adjust behavior when encountering new, complex, or changing task demands [13]. In learning environments that require continuous interpretation of performance information, such as AI-supported coaching, adaptable students are more likely to critically evaluate feedback, explore alternative solutions, and modify their training approaches when necessary [6]. Studies in sport and motor-learning contexts show that athletes with higher cognitive adaptability tend to process feedback more effectively, experiment with corrective strategies, and remain engaged in practice tasks even when performance demands fluctuate [14, 15]. Research in physical education similarly indicates that cognitive flexibility aligns with greater involvement in deeper learning, stronger persistence, and greater readiness to adjust techniques in response to instructional inputs [16, 17]. Because AI-generated feedback often highlights performance discrepancies and suggests multiple avenues for refinement, students with higher adaptability may be particularly willing to engage with such information and invest effort in applying it to future practice [18, 19]. Taken together, these perspectives support the expectation that cognitive adaptability is positively associated with athletic learning engagement and may mediate the relationship between perceptions of AI-supported coaching and students’ engagement in sport-education settings. A growth mindset is the belief that athletic abilities and technical skills can be developed through sustained effort, practice, and effective strategy use rather than being fixed traits [20]. In sport education, students who endorse growth-oriented beliefs tend to approach training with greater persistence, interpret errors as learning opportunities, and show stronger commitment to improving specific movement or tactical skills [21, 22]. Studies with sport majors indicate that growth mindset is associated with adaptive motivational states, including higher practice engagement, challenge-seeking behavior, and willingness to refine techniques following corrective input [23, 24]. At the same time, mindset is shaped by how students process feedback rather than by feedback availability alone; research suggests that the capacity to reinterpret errors flexibly and adjust strategies—hallmarks of cognitive adaptability—often precedes the development of more constructive, growth-oriented beliefs in sport settings [25]. Thus, in AI-supported coaching environments, students who demonstrate greater adaptability may be more inclined to view performance information as actionable and improvable, aligning with a stronger growth mindset. In turn, a growth-oriented belief system is consistently linked to greater engagement in athletic learning tasks. Taken together, these patterns support the expectation that a growth mindset may operate as a second, downstream mediator, following cognitive adaptability, in the association between perceived AI-supported coaching and athletic learning engagement among sports-major students. Bringing these perspectives together, a sequential process in which cognitive adaptability precedes growth mindset is theoretically plausible in sport-education contexts. Adaptable students—those who reinterpret feedback flexibly, explore alternative movement patterns, and adjust training strategies—tend to construct more adaptive explanations for performance fluctuations, which aligns with growth-oriented beliefs about skill development [25]. Evidence from sport pedagogy and athlete development research indicates that flexible cognitive processing often precedes motivational shifts, with athletes who demonstrate greater adaptability showing greater openness to improvement and stronger belief in their capacity to refine skills over time [14]. In turn, a growth mindset has been consistently associated with deeper engagement in learning, higher effort investment, and greater persistence in response to setbacks among university sport students [26–28]. When viewed together, these findings suggest that students who adaptively interpret AI-generated feedback may be more inclined to adopt growth-oriented beliefs, which, in turn, align with stronger engagement in athletic learning tasks. Accordingly, the present study proposes that cognitive adaptability and a growth mindset may operate together in a sequential pathway, with perceptions of AI-supported coaching corresponding to athletic learning engagement among sports-major students. Cognitive adaptability and a growth mindset were conceptualized as sequential rather than parallel mechanisms, grounded in self-regulated learning and social-cognitive perspectives. Self-regulated learning theory suggests that adaptive cognitive processes, such as flexible strategy use, monitoring, and adjustment, often precede and shape learners’ motivational interpretations of competence and improvement [29]. When students successfully adapt their cognitive strategies in response to feedback or novel learning demands, they are more likely to perceive abilities as improvable and effort-responsive, reinforcing incremental beliefs characteristic of a growth mindset [20]. From a social cognitive perspective, repeated experiences of effective cognitive adjustment can strengthen mastery-oriented beliefs and perceptions of skill malleability [30]. Further evidence indicates that growth mindset beliefs are closely linked to self-regulatory processes and adaptive learning behaviors [25]. Accordingly, cognitive adaptability may function as a proximal cognitive mechanism through which AI-supported coaching experiences are cognitively processed, with growth mindset emerging as a downstream motivational interpretation of these adaptive learning experiences. While both constructs are conceptually related, they reflect distinct domains, cognitive regulation versus motivational belief systems, thereby supporting examination of a layered cognitive-motivational pathway. Nevertheless, alternative parallel pathways remain plausible and warrant examination in future longitudinal research. This study Despite growing integration of AI-supported coaching in sports education, empirical work has not yet examined how sports-major students interpret these technologies in relation to their cognitive adaptability, growth-oriented beliefs, and engagement. Prior research has tended to isolate these constructs—focusing separately on AI feedback, adaptability, or mindset—without considering how they may operate together within technologically enriched training environments. This has left a limited understanding of the psychological processes through which AI-supported coaching aligns with students’ learning involvement. Addressing this gap, the present study proposes an integrative framework in which cognitive adaptability and growth mindset function as interconnected mechanisms linking students’ perceptions of AI-supported coaching with their athletic learning engagement. By situating AI-supported coaching within a cognitive–motivational pathway, this study contributes to sport education research by clarifying how students process AI-generated feedback and how these interpretations correspond with their engagement in skill development. More broadly, the findings offer a theoretically grounded model that can inform the design of psychologically responsive AI-supported learning environments for sports-major students. Methods Study design and setting This study employed a cross-sectional survey design conducted in March 2025 in Dezhou City, Shandong Province, China. All universities in Dezhou offering undergraduate and graduate programs in sports were invited to participate. Data were collected online via a secure survey platform, ensuring accessibility across campuses and reducing administrative burden on participating institutions. Sampling approach and recruitment procedures A multi-institutional recruitment approach was used to obtain a broad, diverse sample of sports majors across universities in Dezhou City, Shandong Province. The sampling frame comprised all universities in Dezhou that offered undergraduate and graduate sports-related programs, including sports training, physical education, sports rehabilitation, and sports science tracks. Since no central registry of sports-major students was available, a practical, institution-based recruitment strategy was adopted. Each eligible university was formally contacted through departmental administrators, who assisted in distributing the online survey link to all enrolled students in sports majors via official student WeChat groups, academic notice boards, and university learning management systems. This resulted in a voluntary response sample drawn from the full accessible population, which is an appropriate and commonly accepted approach in large-scale survey research in higher education where complete enumeration lists are not publicly available. A total of 2,740 students received the survey invitation. Of these, 1,794 students provided valid and complete responses and were included in the final analysis. The final sample, therefore, reflects a 65.4% valid response rate across participating institutions. This sampling approach ensured broad representation across the sports-major student population in Dezhou while maintaining feasibility and ethical standards for participant recruitment in multi-university research. Eligibility criteria Students were eligible to participate if they met the following criteria: Enrolled as a full-time undergraduate student in a sports-related major (e.g., physical education, sports training, sports rehabilitation, sport science) at a university located in Dezhou, Shandong Province.Able to read and understand Chinese sufficiently to complete the questionnaire.Provided online informed consent prior to beginning the survey. Enrolled as a full-time undergraduate student in a sports-related major (e.g., physical education, sports training, sports rehabilitation, sport science) at a university located in Dezhou, Shandong Province. Enrolled as a full-time undergraduate student in a sports-related major (e.g., physical education, sports training, sports rehabilitation, sport science) at a university located in Dezhou, Shandong Province. Able to read and understand Chinese sufficiently to complete the questionnaire. Able to read and understand Chinese sufficiently to complete the questionnaire. Provided online informed consent prior to beginning the survey. Provided online informed consent prior to beginning the survey. Participants were excluded if: They were enrolled in non-sport majors or in postgraduate programs.They submitted duplicate responses, detected via device ID, timestamps, and pattern checks.Their survey contained > 10% missing items or showed signs of inattentive responding (e.g., uniform responses, completion time < 5 min). They were enrolled in non-sport majors or in postgraduate programs. They were enrolled in non-sport majors or in postgraduate programs. They submitted duplicate responses, detected via device ID, timestamps, and pattern checks. They submitted duplicate responses, detected via device ID, timestamps, and pattern checks. Their survey contained > 10% missing items or showed signs of inattentive responding (e.g., uniform responses, completion time < 5 min). Their survey contained > 10% missing items or showed signs of inattentive responding (e.g., uniform responses, completion time < 5 min). Pilot testing Prior to the main data collection, a pilot test with 42 sports-major students from one participating university was conducted to assess the clarity of the questionnaire, translation equivalence, cultural appropriateness, the technical functioning of the online survey, and the estimated response time. The pilot sample comprised undergraduate sports majors from the same academic context as the main study population, thereby ensuring the instrument's suitability for the target group. Feedback indicated the items were clear and comprehensible. Minor linguistic refinements were made to ensure conceptual consistency with validated Chinese versions of the scales. The internal consistency reliability of all pilot-tested scales (Cronbach’s α > 0.80 across constructs) supported proceeding with full-scale data collection. Data collection procedure Data collection occurred between March 1 and March 30, 2025. The online survey included a study information sheet, informed consent, screening questions, a main questionnaire assessing perceived AI-supported coaching, cognitive adaptability, growth mindset, and athletic learning engagement, and demographic information (age, gender, university, program, year of study). Participants completed the survey individually using personal mobile phones or computers. The system allowed one submission per device to minimize duplication. Completion time ranged from 7 to 12 min. Sample size and power considerations A total of 2,740 sports-major students were initially invited. After applying exclusion criteria, the final analytic sample consisted of 1,794 students, yielding a valid response rate of 65.4%. For structural equation modeling (SEM), recommended sample size rules include: A minimum of 10–20 participants per estimated parameter, or A minimum sample of 500 for complex mediation models [31]. Monte Carlo simulations for a dual-mediation model with medium path coefficients (β ≈ 0.20–0.30) show that sample sizes above 1,000 yield > 0.90 power to detect both direct and indirect effects. Thus, the achieved sample (n = 1,794) far exceeded SEM requirements, providing excellent statistical power for detecting mediation pathways. Data screening and cleaning To ensure the quality and integrity of the dataset, several checks for attention and response validity were implemented. Of the 2,740 initially invited participants, 2,164 submitted surveys. First, 26 duplicate entries, identified through matching device IDs, identical timestamps, and repeated response patterns, were removed. Next, 41 cases with more than 10% missing data were excluded. Surveys exhibiting inattentive or low-quality responding were then identified through straight-lining patterns, contradictory item responses, and completion times below 5 min, a threshold set at less than half of the median completion time observed in the pilot test. This step resulted in the exclusion of 79 cases. An additional 24 respondents were removed for failing eligibility criteria, as indicated by their enrollment in a non-sports major. After all exclusions, the final analytic sample comprised 1,794 valid cases, yielding high-quality data suitable for statistical analysis. Missing data handling Patterns of missing data were evaluated using Little’s MCAR test, which indicated data were missing completely at random (MCAR) (p > 0.05). Since item-level missingness was minimal (< 3% across all variables), missing values were addressed using Full Information Maximum Likelihood (FIML) during SEM estimation. FIML preserves sample size and produces unbiased parameter estimates under MCAR and MAR conditions. Ethical considerations This study was approved by the Institutional Review Board (IRB) of Dezhou University (Approval No. 5991/25). All procedures adhered to the ethical principles of the Declaration of Helsinki. Participation was entirely voluntary, and informed consent was obtained from all participants before data collection. Students were informed of the study purpose, confidentiality protections, and their right to withdraw at any time without penalty. No identifying information was collected, and all data were stored on encrypted servers accessible only to the research team. Measures Demographic information Participants provided demographic information, including age, gender, year of study, program major, and university affiliation. The final sample of 1,794 sports-major students had a mean age of 19.4 years (SD = 1.8), with 58.7% identifying as male (n = 1,053) and 41.3% as female (n = 741). Students were distributed across academic years: 27.6% first-year, 25.1% second-year, 24.8% third-year, and 22.5% fourth-year. The sample encompassed all major sports-related disciplines offered at participating institutions, including sports training, physical education, sports rehabilitation, and sports science, ensuring coverage of the diverse academic pathways within sports education programs in Dezhou. No single discipline dominated the sample, indicating a broadly distributed representation across sports-related majors. Perceived AI-supported coaching The Perceived Usefulness (PU) subscale from Davis’ Technology Acceptance Model (TAM) [32] was adapted to assess students’ perceptions of the usefulness and benefits of AI-supported coaching tools for improving training and learning outcomes. The PU subscale contains 6 items (e.g., “Using AI-based coaching tools enhances my performance”) rated on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). Higher scores indicated stronger perceived usefulness of AI-supported coaching. The PU scale has shown strong internal consistency (Cronbach’s α > 0.85) and good construct validity in digital learning and AI-assisted contexts [32, 33]. Chinese-adapted versions further support its reliability and unidimensional structure among university students and users of AI-supported learning technologies [34–36]. Cognitive adaptability Cognitive adaptability was assessed using the Cognitive Flexibility Inventory (CFI) [37]. The present study adapted the Chinese version of this scale to assess the cognitive adaptability of sports students [38]. The CFI includes 20 items across two dimensions—Alternatives and Control—and is rated on a 7-point Likert scale (1 = strongly disagree, 7 = strongly agree), with higher scores reflecting greater cognitive adaptability. Sample items include: “I can find workable solutions to difficult situations” (Alternatives) and “I have difficulty controlling my thoughts when facing challenging situations” (Control; reverse-coded). The original scale demonstrated strong internal consistency (α = 0.90) and good convergent and discriminant validity [37]. The Chinese version of the CFI has been validated in university student samples, showing satisfactory reliability (α ranging from 0.86 to 0.92) and a stable two-factor structure supported by confirmatory factor analysis [39, 40] Growth mindset Growth mindset was assessed using an adapted version of Dweck’s Implicit Theories of Intelligence Scale [41], contextualized to athletic skill development. The scale comprised six items (three entity and three incremental statements) rated on a 6-point Likert scale (1 = strongly agree, 6 = strongly disagree), with entity items reverse-coded so that higher scores indicated a stronger growth mindset. A sample incremental item is: “No matter who you are, you can significantly improve your athletic ability.” A sample entity item is: “You have a certain level of athletic ability, and you cannot really do much to change it” (reverse-coded). Previous studies have reported acceptable to good internal consistency for this scale (α typically ranging from 0.70 to 0.85) and have provided evidence of construct validity through associations with motivation, engagement, and adaptive learning outcomes [41]. Chinese versions of the scale have been widely validated among adolescents and sports-major university students, with factor-analytic evidence supporting its conceptual structure in sport and educational contexts [42, 43]. In the measurement model, entity items were reverse-coded, and all six items were modeled as a single latent growth mindset factor, with higher scores indicating stronger incremental beliefs. Factor loadings indicated that reverse-coded and positively worded items loaded appropriately on the intended construct, suggesting no evidence of substantial method effects. Athletic learning engagement Athletic learning engagement was measured using the Athlete Engagement Questionnaire (AEQ) developed by Lonsdale et al. [44]. The current study adopted the Chinese version of AEQ [45]. The AEQ contains 16 items across four dimensions—confidence, dedication, vigor, and enthusiasm—rated on a 5-point Likert scale (1 = almost never, 5 = almost always). Sample items include: “I am enthusiastic about my sport” (enthusiasm) and “I try hard to improve in my sport” (dedication). The original AEQ demonstrated strong internal consistency (α = 0.84–0.91) and good construct validity [44]. The Chinese version of the AEQ has been validated among sports-major students, showing high reliability (α > 0.80) and excellent factorial validity supported by confirmatory factor analysis [45, 46]. Statistical analysis Data analyses were conducted using SPSS (v. 27) for preliminary statistics and AMOS 26 for structural equation modeling (SEM). Prior to hypothesis testing, all variables were examined for normality, outliers, and multicollinearity. Skewness and kurtosis values fell within acceptable ranges (± 2), and no multicollinearity was detected (VIF < 3). Internal consistency reliability was evaluated using Cronbach’s α and composite reliability (CR), while average variance extracted (AVE) and standardized factor loadings were used to assess convergent validity. Discriminant validity was verified using the Fornell–Larcker criterion and HTMT ratios. A two-step SEM approach was adopted. First, a confirmatory factor analysis (CFA) was conducted to evaluate the measurement model, including factor loadings, CRs, AVEs, and overall model fit. Model fit was assessed using multiple indices: χ2/df, CFI, TLI, RMSEA, and SRMR, following established cutoffs (e.g., CFI/TLI ≥ 0.90, RMSEA ≤ 0.08) (L. [47]). The CFA was specified as an integrated measurement model that simultaneously included all study constructs (four-factor model), with each item loading on its respective latent construct. After establishing measurement adequacy, the structural model was estimated to test direct and indirect effects among perceived AI-supported coaching, cognitive adaptability, growth mindset, and athletic learning engagement. Sequential mediation was examined using a bias-corrected bootstrapping procedure (5,000 resamples). Indirect effects were considered significant when the 95% confidence intervals (CIs) did not include zero. The sequential mediation pathway was tested as: AI-supported coaching → cognitive adaptability → growth mindset → athletic learning engagement In addition, individual mediation paths were evaluated to distinguish the cognitive adaptability’s independent mediating effect and the growth mindset’s independent mediating effect from the full chain effect. Standardized path coefficients (β), p-values, and bootstrapped CIs were reported for all direct, indirect, and total effects. To examine the robustness of the mediation findings, a sensitivity analysis was conducted by estimating alternative structural models and assessing whether the mediation effects remained stable under varying specifications. All analyses were performed using Full Information Maximum Likelihood (FIML) to handle minimal missing data. Results Common method bias assessment To assess the potential influence of Common Method Bias (CMB), Harman’s single-factor test was first conducted. An unrotated exploratory factor analysis revealed that the first factor accounted for 23.7% of the variance, well below the recommended threshold of 40%, indicating that common method bias was unlikely to be a significant concern. In addition, a common latent factor (CLF) was included in the measurement model to capture variance shared among all indicators. The comparison between standardized loadings with and without the CLF showed differences of less than 0.10, further suggesting that common method bias did not substantially affect the results. Measurement diagnostics and reliability Table 1 presents the preliminary diagnostic estimates of the study variables. All constructs showed acceptable normality, with skewness and kurtosis values falling within recommended ranges. Multicollinearity was not an issue, as VIF values (1.62–2.03) were well below the cutoff of 3. Reliability indicators were strong, with Cronbach’s alpha values between 0.87 and 0.91 and composite reliability above 0.89. Convergent validity was also supported, as all AVE values exceeded 0.50. Overall, the results in Table 1 indicated that the measurement properties of all variables were adequate for further structural analysis.Table 1Preliminary diagnostic estimates of the study variablesConstructSkewnessKurtosisVIFαCRAVEAI-supported coaching−0.42−0.181.780.890.900.601Cognitive adaptability−0.36−0.222.030.910.920.622Growth mindset−0.28−0.111.620.870.890.588Athletic Learning Engagement−0.510.071.920.900.920.607VIF Variance Inflation Factor. α: Cronbach’s alpha. CR Composite Reliability. AVE Average Variance Extracted Table 1Preliminary diagnostic estimates of the study variablesConstructSkewnessKurtosisVIFαCRAVEAI-supported coaching−0.42−0.181.780.890.900.601Cognitive adaptability−0.36−0.222.030.910.920.622Growth mindset−0.28−0.111.620.870.890.588Athletic Learning Engagement−0.510.071.920.900.920.607VIF Variance Inflation Factor. α: Cronbach’s alpha. CR Composite Reliability. AVE Average Variance Extracted Preliminary diagnostic estimates of the study variables VIF Variance Inflation Factor. α: Cronbach’s alpha. CR Composite Reliability. AVE Average Variance Extracted Bivariate correlations and measurement discriminance Table 2 presents the construct correlations and discriminant validity assessment using the Fornell–Larcker criterion. The square roots of AVE (diagonal values) exceeded the corresponding inter-construct correlations, indicating satisfactory discriminant validity among all variables. Correlational analysis showed that AI-supported coaching was moderately correlated with cognitive adaptability (r = 0.466, p < 0.001) and small correlations with growth mindset (r = 0.287, p < 0.001) and athletic learning engagement (r = 0.323, p < 0.001). Cognitive adaptability was moderately to strongly correlated with growth mindset (r = 0.538, p < 0.001) and athletic learning engagement (r = 0.488, p < 0.001). Growth mindset was strongly correlated with athletic learning engagement (r = 0.561, p < 0.001). Overall, the values in Table 2 confirmed both adequate discriminant validity and theoretically consistent association patterns among key study variables.Table 2Construct correlations and discriminant validity (Fornell–Larcker Criterion)12341. AI-supported coaching0.7752. Cognitive adaptability0.466***0.7893. Growth mindset0.287**0.538***0.7674. Athletic Learning Engagement0.323**0.488***0.561***0.779***p < 0.001, **p < 0.05. Bold shows discriminant validity estimates Table 2Construct correlations and discriminant validity (Fornell–Larcker Criterion)12341. AI-supported coaching0.7752. Cognitive adaptability0.466***0.7893. Growth mindset0.287**0.538***0.7674. Athletic Learning Engagement0.323**0.488***0.561***0.779***p < 0.001, **p < 0.05. Bold shows discriminant validity estimates Construct correlations and discriminant validity (Fornell–Larcker Criterion) ***p < 0.001, **p < 0.05. Bold shows discriminant validity estimates Model fit analysis The CFA fit indices reported in Table 3 refer to the overall four-factor integrated measurement model estimated simultaneously. The results demonstrated that the model achieved an acceptable and well-fitting structure. The χ2/df ratio was 1.09, which is well below the recommended threshold of 3, indicating excellent model parsimony. Both the Comparative Fit Index (CFI = 0.92) and the Tucker–Lewis Index (TLI = 0.92) met the recommended minimum of 0.90, demonstrating good incremental fit. Additionally, the Root Mean Square Error of Approximation (RMSEA = 0.041) and the Standardized Root Mean Squared Residual (SRMR = 0.036) were below the recommended cutoff of 0.08, indicating satisfactory absolute fit. Collectively, the indices in Table 3 confirmed that the measurement model demonstrated a strong overall fit to the data.Table 3Measurement model fit indicesFit IndexObserved ValueRecommended Threshold\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi 2/df$$\end{document}(1170.06/1074)1.09 < 3Comparative Fit Index0.92 ≥ 0.90Tucker-Lewis Index0.92 ≥ 0.90Root Mean Square Error of Approximation0.041 < 0.08Standardized Root Mean Squared Residual0.036 < 0.08 Table 3Measurement model fit indicesFit IndexObserved ValueRecommended Threshold\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi 2/df$$\end{document}(1170.06/1074)1.09 < 3Comparative Fit Index0.92 ≥ 0.90Tucker-Lewis Index0.92 ≥ 0.90Root Mean Square Error of Approximation0.041 < 0.08Standardized Root Mean Squared Residual0.036 < 0.08 Measurement model fit indices Sequential mediation analysis The structural model demonstrated several significant associations among the study variables, keeping age and gender as controlled (Fig. 1). The results indicated that perceived AI-supported coaching was positively related to cognitive adaptability (β = 0.313, p < 0.001). However, its path to growth mindset was small and not statistically significant (β = 0.08, p = 0.137). AI-supported coaching also showed a modest, positive association with athletic learning engagement (β = 0.11, p < 0.01). Cognitive adaptability demonstrated substantial positive associations with growth mindset (β = 0.455, p < 0.001) and athletic learning engagement (β = 0.273, p < 0.001). Additionally, a growth mindset was positively and statistically significantly associated with athletic learning engagement (β = 0.368, p < 0.001). With respect to indirect effects: The indirect path from AI-supported coaching → cognitive adaptability → engagement was statistically significant (indirect effect = 0.085, 95% Boot CI [0.042, 0.137]). Whereas, the indirect path from AI-supported coaching → growth mindset → engagement was not statistically significant, as the confidence interval included zero (indirect effect = 0.029, 95% Boot CI [− 0.015, 0.072]). Moreover, the sequential indirect pathway through both mediators: AI-supported coaching → cognitive adaptability → growth mindset → engagement, was statistically significant (indirect effect = 0.052, 95% Boot CI [0.018, 0.094]).Fig. 1Path analysis (Sequential Mediation) – Direct and indirect effects are presented with standardized β coefficients and 95% Boot Confidence Interval (CI) with 5000 resamples. *** shows significance at p < 0.001 while** shows significance at p < 0.05. Solid lines indicate a significant path, whereas dotted lines indicate an insignificant path. IV stands for independent variable, M1 for mediator one, M2 for mediator two, and DV for dependent variable Path analysis (Sequential Mediation) – Direct and indirect effects are presented with standardized β coefficients and 95% Boot Confidence Interval (CI) with 5000 resamples. *** shows significance at p < 0.001 while** shows significance at p < 0.05. Solid lines indicate a significant path, whereas dotted lines indicate an insignificant path. IV stands for independent variable, M1 for mediator one, M2 for mediator two, and DV for dependent variable Moreover, age and gender were included as control variables in the structural model; however, neither showed a statistically significant association with athletic learning engagement, and their inclusion did not alter the pattern of relationships among the primary study variables. Sensitivity analysis A sensitivity analysis was conducted to examine the robustness of the mediation findings. Consistent with methodological recommendations for evaluating the stability of indirect effects in SEM [48, 49], the sequential mediation model was compared with alternative model specifications, including a direct-only model and parallel mediation structures. The hypothesized model demonstrated good overall fit (χ2/df = 1.09, CFI = 0.92, TLI = 0.92, RMSEA = 0.041, SRMR = 0.036). Comparable levels of fit were observed for the direct-effects-only model (χ2/df = 1.21, CFI = 0.91, TLI = 0.91, RMSEA = 0.046, SRMR = 0.041) and the parallel mediation model (χ2/df = 1.14, CFI = 0.92, TLI = 0.92, RMSEA = 0.043, SRMR = 0.038), indicating no meaningful deterioration in model fit across specifications. Importantly, the substantive mediation results remained stable across models. The indirect effect of perceived AI-supported coaching on athletic learning engagement via cognitive adaptability remained statistically significant in both the hypothesized sequential model (β = 0.085, 95% Boot CI [0.042, 0.137]) and the parallel mediation model (β = 0.082, 95% Boot CI [0.039, 0.132]). The sequential indirect pathway through cognitive adaptability and growth mindset also remained significant in the hypothesized model (β = 0.052, 95% Boot CI [0.018, 0.094]), whereas the indirect path through growth mindset alone was consistently non-significant (β = 0.027, 95% Boot CI [− 0.014, 0.071]). Because some alternative models were non-nested, model comparisons were based on comparative fit indices and parameter-estimate stability rather than formal χ2 difference testing. Overall, these results indicate that the mediation findings were robust to reasonable variations in model specification. Discussion The present study examined how perceived AI-supported coaching was related to athletic learning engagement among sports-major students, with particular attention to the roles of cognitive adaptability and growth mindset. The findings indicated that perceived AI-supported coaching was positively associated with cognitive adaptability and, to a lesser extent, with athletic learning engagement. Cognitive adaptability was significantly associated with both growth mindset and engagement, and growth mindset was positively associated with engagement. Mediation analyses further showed that cognitive adaptability served as a significant intermediary linking perceived AI-supported coaching to engagement, both independently and as part of a sequential pathway involving growth mindset. In contrast, the indirect path through growth mindset alone did not reach statistical significance. Collectively, these results highlight the importance of cognitive and motivational processes in understanding how students’ perceptions of AI-supported coaching align with their learning-related engagement in sport settings. The first hypothesis proposed that perceived AI-supported coaching would be positively associated with cognitive adaptability. The findings supported this expectation, as students who reported higher perceptions of AI-supported coaching also demonstrated higher levels of cognitive adaptability. This pattern aligns with prior research suggesting that digital and AI-assisted learning environments may be associated with learners’ flexibility in strategy use, problem re-framing, and responsiveness to feedback. For example, studies on AI-driven feedback systems and adaptive learning platforms have shown that exposure to technologically enriched training environments is often associated with greater cognitive flexibility and adaptive problem-solving tendencies among general and sports-major students [50, 51]. Research in sport-related learning engagement has similarly documented that interactive, data-driven analytic tools are associated with athletes’ ability to adjust techniques and refine decision-making under variable conditions [12, 52]. More broadly, work in educational technology has reported that AI-enhanced instructional systems tend to be associated with improvements in learners’ strategic shifting and openness to alternative solutions, both central components of cognitive adaptability [53, 54]. The present finding is therefore statistically consistent with a growing body of literature indicating that AI-supported learning environments are associated with cognitive adaptability through increased exposure to dynamic performance information and structured feedback. The results supported the second hypothesis, indicating that students who perceived AI-supported coaching more positively also reported higher engagement, indirectly through greater cognitive adaptability. This finding aligns with prior work showing that adaptive cognitive capacities often underlie the connection between technology-enhanced learning environments and learners’ behavioral engagement. For instance, research on adaptive and personalized digital learning systems has shown that exposure to feedback-rich environments is associated with improved cognitive flexibility and more adaptive responses to learning challenges, which in turn correspond with higher engagement and persistence [53, 55, 56]. Studies in sport science similarly suggest that sport students who demonstrate higher cognitive adaptability—such as the ability to adjust strategies, re-evaluate errors, and respond dynamically to instructional inputs—tend to display greater engagement and investment in training [9, 10, 57]. Furthermore, meta-analytic evidence indicates that feedback-rich learning contexts, including video-based and interactive analytical tools, are positively associated with both adaptive cognitive processing and learners’ sustained involvement in practice activities [58]. Taken together, these findings support the present result that cognitive adaptability functions as a meaningful pathway through which perceived AI-supported coaching aligns with athletic learning engagement. The third hypothesis, which proposed that perceived AI-supported coaching would be indirectly related to athletic learning engagement through growth mindset alone, was not supported. Although growth mindset was positively associated with engagement, perceived AI-supported coaching was not statistically significantly associated with growth mindset, resulting in a non-significant indirect effect. This pattern is consistent with previous research indicating that a growth mindset tends to develop primarily through sustained mastery experiences, meaningful feedback, and long-term motivational socialization rather than through isolated perceptions of technological tools or instructional formats [25, 59]. Studies examining digital or AI-enhanced learning environments have similarly reported that, while such environments may facilitate performance monitoring or strategic adjustment, they do not consistently align with shifts in learners’ implicit beliefs about ability unless paired with explicit mindset interventions or instructional framing [59–61]. In sport contexts, research suggests that growth mindset is shaped more strongly by coaching climate, interpersonal feedback, and mastery-oriented environments than by exposure to specific technologies alone [62, 63]. Therefore, these findings help contextualize the present non-significant pathway, suggesting that a growth mindset may require deeper motivational inputs and cannot be readily inferred from students’ perceptions of AI-supported coaching systems in isolation. The sequential mediation hypothesis, which proposed that perceived AI-supported coaching would be indirectly related to athletic learning engagement through cognitive adaptability and a growth mindset, was supported. Although the indirect effect was modest in magnitude, it was statistically significant, suggesting that students who reported more favorable perceptions of AI-supported coaching also showed higher engagement, mediated by greater cognitive adaptability and a stronger growth mindset. This layered pattern is consistent with emerging evidence that cognitive and motivational dispositions often operate in combination rather than in isolation when linking contextual features to engagement. Recent structural models have shown, for example, that cognitive flexibility can function as a mediator between self-regulatory capacities and adaptive outcomes such as resilience and school functioning [64, 65], while growth mindset has been linked to learning engagement through serial mediating mechanisms involving stress appraisals, emotions, or future-oriented motivational processes [64]. The present findings align with this broader pattern by indicating that, in the context of AI-supported coaching, cognitive adaptability and a growth mindset together formed a meaningful sequential pathway through which perceptions of technologically enriched training environments were associated with students’ reported engagement in athletic learning activities. The sequential pathway identified in this study may be interpreted as a layered cognitive–motivational process in which adaptive cognitive processing serves as a proximal mechanism shaping subsequent motivational beliefs about learning and improvement. From a self-regulated learning perspective, students who demonstrate greater cognitive adaptability in response to AI-supported coaching may be more likely to interpret challenges as opportunities for skill development, thereby reinforcing incremental beliefs characteristic of a growth mindset [20, 29]. Repeated experiences of effective cognitive adjustment may be associated with stronger mastery-oriented interpretations of performance and ability, consistent with social cognitive accounts of belief formation in learning contexts [30]. This pattern is consistent with the possibility that AI-supported coaching was associated with adaptive cognitive engagement with learning tasks, with growth mindset functioning as a secondary motivational reinforcement that may support sustained engagement [25]. Such a layered interpretation aligns with contemporary models of learning that emphasize dynamic interactions between cognitive regulation and motivational belief systems in technology-enhanced educational environments. Although the sequential indirect effect was modest in magnitude, such effects may still hold theoretical significance given the multifaceted and context-dependent nature of engagement. In complex sport-learning environments, engagement is shaped by multiple interacting cognitive, motivational, and contextual factors; therefore, even modest but consistent indirect pathways may reflect meaningful psychological processes. The present findings are consistent with the interpretation that AI-supported coaching was associated with engagement by facilitating adaptive cognitive processing, with a growth mindset functioning as a complementary motivational mechanism that may reinforce sustained involvement in training and learning activities. This layered interpretation extends prior research by highlighting how cognitive and motivational dispositions operate in combination within technology-enhanced sport learning environments rather than as isolated predictors of engagement. In addition to the significant pathways, it is important to consider the non-significant direct association between perceived AI-supported coaching and growth mindset. Although a growth mindset was positively associated with athletic learning engagement, students’ perceptions of AI-supported coaching were not significantly associated with growth-oriented beliefs. One possible explanation is that a growth mindset reflects a relatively stable belief system shaped by prolonged mastery experiences, a coaching climate, and socialization processes, rather than by short-term perceptions of technological tools. AI-supported coaching systems may provide performance feedback and analytic insights, but without explicit pedagogical framing, such feedback may be insufficient to shift students’ implicit beliefs about ability development substantially. Measurement and design factors may also have contributed to this non-significant finding. Growth mindset was assessed using a generalized belief-based scale, which may be less sensitive to contextual influences such as perceptions of AI-supported coaching. Moreover, the cross-sectional design limits the ability to capture gradual changes in belief that may emerge over time with sustained exposure to AI-supported systems. Notably, the significant sequential pathway through cognitive adaptability suggests that AI-supported coaching may be indirectly associated with a growth mindset by first shaping how students cognitively interpret and adapt to feedback. Future research employing longitudinal or intervention-based designs, as well as sport-specific mindset measures and explicit mindset-supportive instructional framing, may better capture the conditions under which AI-supported coaching contributes to growth-oriented beliefs. Moreover, the sample's diversity with respect to gender, academic year, and sports specialization may also have implications for interpreting the findings. Students at different stages of training or enrolled in different sports-related programs may vary in their exposure to structured coaching, learning demands, and performance expectations, which could influence how AI-supported coaching is perceived and utilized. Although subgroup analyses were beyond the scope of the present study, the consistency of the overall model across a large and heterogeneous sample supports the robustness of the observed relationships. Future research could further examine whether specific pathways differ across academic levels or disciplinary tracks. This study advances theory in three key ways. First, by integrating perceived AI-supported coaching with cognitive adaptability, growth mindset, and engagement, it extends technology-acceptance perspectives into the psychological domain of sport education. Prior research applying TAM in physical education has focused mainly on technology-use intentions rather than on internal cognitive or motivational processes; the present model therefore broadens this theoretical scope. Second, the findings highlight cognitive adaptability as a central mechanism statistically linking perceptions of AI-supported coaching with engagement. This contributes to the sport psychology literature by positioning cognitive adaptability not only as a performance-relevant skill but also as a psychological pathway associated with technologically enriched learning environments and sport students’ motivational orientations. Third, the results clarify the role of a growth mindset within AI-supported training contexts. The non-significant direct association between AI-supported coaching and growth mindset, combined with its contribution only within the sequential pathway, is consistent with the interpretation that mindset may function as a downstream motivational disposition, shaped in part by sport students’ cognitive adaptability rather than by perceptions of technology alone. This nuanced positioning complements existing work, indicating that mindset shifts typically require sustained mastery experiences or targeted pedagogical framing. Together, these contributions provide a theoretically grounded framework for understanding how AI-enhanced coaching environments are associated with cognitive and motivational processes in sport education, and offer a foundation for future investigations that more explicitly integrate AI-coaching research with contemporary motivational and cognitive theories. An additional contextual consideration relates to the characteristics of sports-major university students. Compared with general university students, sports majors typically engage in regular, structured sports participation, receive systematic training, and possess greater domain-specific knowledge and physical competence. Their academic programs often emphasize skill refinement, performance monitoring, and iterative feedback, which may heighten sensitivity to coaching inputs and performance analytics. Moreover, students majoring in sports tend to exhibit higher intrinsic motivation for physical activity and skill development, as their academic and career trajectories closely align with their sport participation. These characteristics may make AI-supported coaching tools particularly relevant and effective for this population, as they align with students’ existing training routines and goal-oriented learning contexts. At the same time, these population-specific features suggest that the observed relationships may not directly generalize to non-sports majors or students with limited sport involvement, underscoring the importance of examining AI-supported coaching across diverse learner groups in future research. Study limitations and future directions Several limitations of the present study should be acknowledged. First, the cross-sectional design precludes firm conclusions about causal directionality among perceived AI-supported coaching, cognitive adaptability, growth mindset, and athletic learning engagement. Although the proposed mediation pathways were theoretically grounded and supported by structural equation modeling and sensitivity analyses, longitudinal or experimental designs are needed to establish temporal ordering and causal mechanisms underlying these relationships more rigorously. Although the sequential pathway was theoretically grounded, the cross-sectional design does not allow confirmation of temporal ordering; future longitudinal or experimental research should examine whether cognitive adaptability and a growth mindset operate sequentially or in parallel over time. Second, the sample comprised exclusively Chinese university students enrolled in sports-related majors. Cultural norms, educational practices, and sport-training environments in China may shape students’ perceptions of AI-supported coaching and related cognitive and motivational processes in ways that differ from other cultural contexts. As a result, caution is warranted when generalizing the present findings to students from other countries, age groups, or non-university athletic populations, such as youth athletes or professional practitioners. Future research should examine whether the proposed model replicates across diverse cultural settings and developmental stages. Third, all study variables were assessed using self-report measures, which may introduce response bias despite procedural and statistical controls for common method variance. Although Harman’s single-factor test and the common latent factor approach suggested that common method bias was unlikely to influence the results substantially, future studies could benefit from incorporating multi-source data, objective performance indicators, or coach-reported assessments of engagement and adaptability. Study implications The findings offer several implications for working with sports-major students. First, because engagement was linked indirectly to perceived AI-supported coaching through cognitive adaptability, AI tools may be most effective when integrated into training activities that require students to interpret feedback, consider alternative tactical or technical responses, and revise their approaches accordingly. Embedding brief reflective tasks after AI feedback, such as identifying alternative movement strategies or evaluating multiple training options, may strengthen this adaptive processing. Second, the results indicate that a growth mindset did not function as a standalone mediator, suggesting that mindset may not shift solely through exposure to AI tools. For sports programs seeking to foster growth-oriented beliefs, explicit pedagogical framing (e.g., emphasizing skill-development trajectories and normalizing errors as learning opportunities) may be necessary alongside AI-supported analysis. Third, because many sports majors progress into coaching or teaching roles, incorporating training in integrating AI-generated feedback with psychological skills (e.g., instructional framing, reflective dialogue, strategy adjustment) could help prepare them to use technology in a more psychologically informed manner. Finally, the combined mediation pattern suggests practical value in providing targeted support for students who show low adaptability or engagement; AI feedback reports may serve as useful anchors in mentoring conversations aimed at strengthening adaptive decision-making and constructive interpretation of performance information. Beyond these general implications, the findings also offer more specific guidance for the design and implementation of AI-supported coaching in physical education and sports training contexts. AI coaching tools can be optimized by incorporating features that explicitly promote cognitive adaptability, such as adaptive feedback that encourages comparisons among alternative techniques, scenario-based performance simulations, and iterative goal-adjustment functions. To support growth-oriented thinking, AI systems could integrate explanatory feedback emphasizing effort–improvement links, individualized progress tracking over time, and reflective prompts that frame errors as part of skill development rather than as performance deficits. With respect to future research, studies could differentiate among types of AI coaching tools (e.g., video-based feedback systems, wearable sensor analytics, or AI-driven performance dashboards) to examine whether their psychological effects on cognitive adaptability and growth mindset vary by design. Longitudinal and intervention-based studies are also needed to assess how sustained exposure to AI-supported coaching influences belief formation, adaptive cognition, and engagement in athletic learning over time. Such approaches would allow for stronger causal inference and provide a more comprehensive understanding of how AI-supported coaching can promote continuous skill development and long-term sports participation. Conclusion This study examined how perceived AI-supported coaching relates to athletic learning engagement among sports-major students, with attention to the roles of cognitive adaptability and growth mindset. The findings showed that cognitive adaptability served as a key intermediary variable linking perceptions of AI-supported coaching to engagement, whereas growth mindset contributed only as part of a sequential pathway. These results underscore the importance of cognitive and motivational processes in understanding students’ responses to technologically enriched training environments. By integrating AI-supported coaching with established psychological constructs, the study offers a framework for future research on how digital tools are interpreted and utilized within sport education and training contexts.