Teachers’ innovative teaching behaviors (ITB) refer to the adoption and implementation of novel pedagogical strategies, tools, and approaches that deviate from traditional, teacher-centered methods. These behaviors are characterized by a focus on student-centered learning, active participation, real-world problem-solving, and the thoughtful integration of technology (Xin et al., 2022). In higher vocational institutions, ITB are particularly important because they help align classroom learning with workplace demands and support students’ transition into specific occupations.

In the context of vocational universities, ITB typically include: Technology-enhanced instructional features, such as the use of digital tools, learning management systems (LMS), interactive whiteboards, simulations, virtual reality (VR), and artificial intelligence (AI) to create dynamic, practice-oriented learning experiences (Henderson et al., 2017). Active and collaborative learning designs, including think-pair-share, flipped classrooms, debates, case studies, and peer instruction, which require students to actively construct knowledge rather than passively receive information (El-Bassiouny and El-Naggar, 2023). Supportive classroom climate, in which students feel respected, valued, and safe to take intellectual and practical risks, which is crucial for emotional engagement (Tasiouli and Lyra, 2024). Differentiated and personalized instruction, where teaching methods, content, and assessments are tailored to the diverse learning paces, experiences, and career interests of vocational students (Iqbal et al., 2020). These instructional features are expected to influence a set of psychological drivers on the student side—such as motivation, self-efficacy, learning satisfaction, and trust in AI—which in turn shape students’ learning engagement. Student learning engagement (LE) is a multidimensional construct that extends beyond mere participation. It typically encompasses three core dimensions: behavioral, emotional, and cognitive engagement (Fredricks et al., 2004; Wang and Eccles, 2013). Behavioral engagement refers to students’ participation in academic and practice-based activities, including attendance, effort, persistence, and time-on-task in both classrooms and training workshops. Emotional engagement involves students’ affective reactions to teachers, peers, and the learning environment—such as interest, enthusiasm, enjoyment, and a sense of belonging to their vocational program. Cognitive engagement pertains to students’ psychological investment in learning, characterized by a willingness to exert mental effort, employ deep learning strategies, and develop self-regulation skills (Pohl, 2020; Wong et al., 2024). High levels of engagement across these dimensions are associated with improved critical thinking, problem-solving abilities, long-term retention of knowledge, and better preparation for the workplace (Pang, 2022; Shcheglova et al., 2019).

Existing research indicates several mechanisms through which ITB can promote each dimension of LE among vocational university students: Behavioral engagement: Active learning strategies inherently require students to be present and participate. Flipped classrooms, for instance, necessitate pre-class preparation and in-class application, directly impacting attendance, effort, and persistence (Rutkienė et al., 2022). The use of interactive technologies such as polling systems, collaborative online platforms, or AI-supported practice systems can also increase participation and time-on-task (Malekjafarian and Gordan, 2024). Cognitive engagement: Problem-based and project-based learning challenge students to think critically, analyze complex information, and apply theoretical knowledge to authentic vocational situations, thereby promoting deeper cognitive processing (Yu, 2024; Zheng et al., 2009). Technology-enhanced simulations or virtual labs allow students to experiment and learn from mistakes in a safe environment, encouraging sophisticated metacognitive strategies (Asare et al., 2023). Personalized learning paths can ensure that students are appropriately challenged, leading to sustained cognitive effort (Lin et al., 2024). Emotional engagement: Innovative methods that make learning more relevant, interactive, and enjoyable can significantly enhance students’ interest and motivation. For example, using real-world case studies or gamification can make abstract concepts more relatable and exciting, fostering a positive emotional connection to the subject matter and to the chosen occupation (Redondo-Rodríguez et al., 2022). A supportive, respectful classroom climate cultivated by ITB further strengthens emotional engagement by promoting belonging and confidence. Taken together, these findings suggest that ITB should be positively associated with vocational students’ learning engagement across all three dimensions. In line with this multidimensional view, the present study distinguishes behavioral, emotional, and cognitive engagement conceptually, while modeling LE as a higher-order construct in the structural model. Accordingly, we propose the following hypotheses about statistical associations in our cross-sectional data:

In addition, when we consider a composite measure of learning engagement, we expect:

These hypotheses refer to expected patterns of association rather than causal effects, given the cross-sectional design of the study.

While a direct link between ITB and LE is intuitive and supported by prior research, the mechanisms underlying this relationship warrant further exploration, especially in vocational higher education. One key psychological driver is learning satisfaction (LS), defined as students’ subjective evaluation of their educational experiences, processes, and outcomes (Wiers-Jenssen et al., 2002). When vocational students perceive teaching methods as engaging, relevant to their future careers, and supportive of their learning needs—characteristics typically associated with ITB—their satisfaction is likely to increase, which in turn may fuel greater engagement (Djudin, 2018).

Learning satisfaction is a crucial indicator of the quality of educational experiences from the student perspective. It reflects students’ contentment with various aspects of their academic and training life, including teaching quality, curriculum relevance to industry, learning resources, assessment methods, and the overall learning environment. High LS has been linked to increased motivation, academic achievement, persistence, and positive word-of-mouth recommendations for the institution (Esteban and del Cerro, 2020). Among the many factors influencing LS, teacher-related variables—particularly teaching style, instructional quality, and the perceived innovativeness of teaching—consistently emerge as significant predictors.

In vocational universities, ITB are expected to enhance LS through several mechanisms. Innovative teaching methods are typically more interactive, participatory, and practice-oriented than traditional approaches (Freeman et al., 2014). For example, incorporating real-world problems (problem-based learning) can enhance cognitive engagement by requiring students to apply knowledge in authentic work contexts (Maftuh et al., 2023), which increases students’ perceptions of relevance and usefulness. The use of collaborative technologies and AI-supported platforms can foster behavioral engagement through peer interaction and repeated practice, while also supporting emotional engagement by creating a supportive learning community (Li et al., 2024). Personalized feedback facilitated by technology, engaging multimedia content, and active learning tasks that give students a sense of agency can significantly enhance their satisfaction with the teaching and learning process (Manalu, 2024). For vocational students in particular, when teaching is perceived as closely aligned with industry standards and future employment, satisfaction tends to rise. Therefore, ITB are expected to have both direct and indirect effects on LE. Directly, innovative teaching designs can increase behavioral, emotional, and cognitive engagement by making learning more active, meaningful, and professionally relevant. Indirectly, ITB can improve LS, which then motivates students to invest more effort, experience more positive emotions, and think more deeply about course content and practical skills. For instance, a well-designed flipped classroom in a vocational course (ITB) may lead students to feel more prepared and perceive in-class practice as more valuable (higher LS), which in turn encourages them to participate more actively and think more critically during simulations or lab sessions (higher LE; Han et al., 2021; Akçayır and Akçayır, 2018).

Based on this reasoning, we propose the following hypotheses:

Consistent with Figure 1 (Parallel multiple mediation model with AI Trust and Learning Satisfaction), these hypotheses refer to indirect (dashed) pathways from ITB to LE via LS, complementing the direct (solid) paths from ITB to the engagement dimensions in our cross-sectional model. We specified AIT and LS as parallel mediators rather than imposing a serial ordering (AIT → LS or LS → AIT) because, given the cross-sectional design, there is no strong theoretical or temporal basis to justify a directional causal sequence between AIT and LS; accordingly, modeling them in parallel provides a more defensible test of their simultaneous mediating roles.

A total of 513 vocational university students were recruited by convenience sampling from several higher vocational institutions in Guangdong Province, China. Guangdong is a highly developed coastal province in southern China, with relatively advanced digital infrastructure and early adoption of AI in higher education, which makes it a suitable context for studying AI-mediated teaching and learning. The sampling frame consisted of full-time students enrolled in vocational bachelor and higher vocational diploma programs. Inclusion criteria were: (a) currently enrolled in one of the participating institutions; (b) able to complete an online questionnaire in Chinese; and (c) provided informed consent. Questionnaires with excessive missing data or patterned responses were excluded from analysis. Among the 513 valid respondents, 217 (42.3%) were male and 296 (57.2%) were female; Regarding academic year, 157 (30.6%) were first-year students, 167 (32.5%) second-year students, 189 (36.8%) third-year students, which sum to the total valid sample of 513.

Prior to analysis, we conducted data-quality screening to ensure the independence and credibility of responses. During this process, a number of questionnaires were identified as highly repetitive (i.e., duplicated cases with identical response patterns across a large proportion of items), suggesting non-independent submissions. These duplicated/repetitive cases were removed, together with questionnaires that showed excessive missing data or patterned responses (e.g., straight-lining). After these exclusions, the final analytic sample comprised N = 513.

This study was reviewed and approved by the Academic Ethics Committee of Zhongshan Institute, University of Electronic Science and Technology of China (2024011418). All procedures complied with the ethical standards of the institutional research committee and with the 1964 Helsinki Declaration and its later amendments.

All scales were originally developed in English and then translated into Chinese using a standard forward–backward translation procedure. Two bilingual experts independently translated the items into Chinese; discrepancies were resolved through discussion. A third bilingual expert, who was blind to the original instruments, back-translated the Chinese version into English. The back-translated items were compared with the originals to ensure semantic equivalence. Prior to the main survey, the questionnaire was piloted with a small sample of vocational students to check clarity and timing. Based on the pilot feedback, minor wording adjustments were made.

Confirmatory factor analyses (CFA) were conducted for all multi-item scales to verify the measurement structure in the present sample. Composite reliability (CR) and average variance extracted (AVE) indices were also computed to assess convergent and discriminant validity. Detailed psychometric results (factor loadings, CR, AVE, and HTMT values) are reported in Tables 1–3. Unless otherwise noted, all items were rated on a 5-point Likert-type scale. A 5-point format was chosen to balance measurement sensitivity with respondent burden and to be consistent with common practice in Chinese higher education research.

Data were collected using the online platform Questionnaire Star (Wenjuanxing), which is widely used for survey research in China. Course instructors and program coordinators in the participating vocational universities distributed the survey link to eligible students via institutional learning management systems and class communication groups. Before accessing the questionnaire, students were presented with an online information sheet describing the purpose of the study, procedures, voluntary nature of participation, potential risks and benefits, and data-protection measures. Students who agreed to participate clicked an “I agree” button to provide electronic informed consent. Participation was anonymous; no identifying personal information (e.g., student ID, name, phone number) was collected. Data were stored on password-protected servers and were accessible only to the research team. After providing consent, participants first completed demographic questions (e.g., gender, age, academic year, major), followed by the Innovative Teaching Behaviors scale, the Artificial Intelligence Trust scale, the Learning Engagement scale, and the Learning Satisfaction scale (Table 3).

Confirmatory factor analyses (CFAs) were conducted in AMOS to examine the measurement properties of the four constructs (ITB, AIT, LS, and LE). For hypothesis testing, we did not estimate a full structural equation model (SEM) with global model-fit indices. Instead, composite scores were computed for each construct and the proposed parallel mediation model was tested using ordinary least squares (OLS) regression–based mediation in PROCESS (Model 4; Hayes), with ITB as the predictor, AIT and LS as parallel mediators, and LE as the outcome. Indirect effects were evaluated using bias-corrected bootstrapping with 5,000 resamples, and effects are reported as unstandardized coefficients (B) with standard errors and 95% confidence intervals. All analyses were conducted using the final analytic sample N = 513.

In the present study, we did not conduct an additional latent method factor (LMF) test for common method variance. Given the large number of items and the multidimensional structure of the measurement model, adding an LMF would substantially increase model complexity and could lead to estimation problems (e.g., nonconvergence or improper solutions), thereby reducing interpretability. Instead, we implemented procedural remedies (e.g., anonymous and voluntary participation, emphasizing that there were no right or wrong answers, and minimizing evaluation pressure) and assessed construct distinctiveness statistically. Specifically, the Harman’s single-factor test indicated that the first factor accounted for 35.6% of the variance, suggesting that a single general factor did not dominate the covariance among items. In addition, the CFA results and HTMT ratios supported satisfactory discriminant validity among the focal constructs. Nevertheless, because all variables were measured via self-report in a single survey wave, common method variance cannot be completely ruled out and should be considered when interpreting the findings. Future research may apply an LMF approach under a more parsimonious measurement structure, or adopt multi-source and/or time-separated designs to further mitigate and test for method effects.

The present study followed the guidelines of Fornell and Larcker (1981) to assess the reliability and validity of the scale.

HTMT ratios (off-diagonal), with 95% bootstrap Cis (n = 5,000 samples, above diagonal). All values <0.85, supporting discriminant validity (Table 4).

This study found that ITB has a significant positive predictive effect on overall LE ( β = 0.5134 , p < 0.001 ) , a result highly consistent with the research conclusions of Wang and Eccles (2013) and Xin et al. (2022). The strong correlation between ITB and LE in this study ( r = 0.518 ) empirically supports the theoretical claim that innovative teaching is closely linked to student engagement. It is important to note, however, that our empirical model treated learning engagement as a single latent construct; the specific process mechanisms described below are literature-based and were not directly tested as separate mediational paths in our data.

Drawing on prior research, at least three mechanisms have been theorized through which ITB may enhance engagement. At the behavioral engagement level, previous work suggests that innovative teaching methods—such as flipped classrooms—can significantly improve students’ classroom participation (Rutkienė et al., 2022). When teachers employ interactive polling systems or collaborative online platforms (Malekjafarian and Gordan, 2024), students’ task engagement time is theorized to increase. Our finding of a robust association between ITB and overall LE is consistent with, but does not by itself prove, this behavioral mechanism (Table 6).

At the cognitive engagement level, studies by Yu (2024) and Zheng et al. (2009) indicate that problem-based and project-based learning promote deep cognitive processing. In particular, when teachers introduce AI-assisted simulation experiments or virtual laboratories (Asare et al., 2023), students can engage in trial-and-error learning in safe environments and develop more complex metacognitive strategies. Our data are compatible with this account insofar as higher ITB predicts higher composite engagement, but we did not include separate indicators of specific cognitive strategies in the structural model, so these mechanisms remain inferential rather than directly tested.

At the emotional engagement level, prior research has emphasized the role of gamified teaching in enhancing students’ interest and enjoyment (Redondo-Rodríguez et al., 2022). By using real case studies and interactive activities, innovative teaching methods can make abstract concepts more relevant and attractive, thereby fostering positive emotional connections to the subject. Again, our findings provide empirical support for a strong overall link between ITB and engagement, while the tripartite mechanisms across behavioral, cognitive, and emotional engagement are grounded in the literature rather than separately estimated in our data.

This study found that LS plays an important mediating role between ITB and LE, with the mediation effect accounting for 29.84% of the total effect. This finding not only validates Hypothesis 3 proposed in the literature review but also provides empirical evidence for the theoretical mechanism summarized there. Specifically, the path analysis shows that the indirect pathway ITB → LS → LE is significant, consistent with the research conclusions of Han et al. (2021).

Theoretically, Djudin (2018) argued that when students perceive teaching methods as attractive, relevant, and responsive to their learning needs—features that characterize innovative teaching behaviors—their satisfaction increases, which in turn drives deeper engagement. Our empirical results directly support this core mechanism at the latent-variable level: ITB significantly predicts LS, and LS significantly predicts LE, with a non-trivial proportion of the ITB–LE relation transmitted through satisfaction. At a more fine-grained level, prior studies emphasize the role of personalized feedback, multimedia content, and active learning tasks in enhancing satisfaction (Manalu, 2024). While our measures did not disentangle the unique contributions of each instructional feature, the significant mediating effect of LS suggests that students’ holistic evaluative experience of teaching is an important psychological bridge between ITB and engagement.

A significant contribution of this study is revealing the mediating role of AIT in the ITB–LE relationship, with AIT accounting for 22.17% of the total effect. At the empirical level, our path model shows that the chain ITB → AIT → LE is significant. This provides direct support for the theoretical claim by Zehner and Hahnel (2023) that students’ trust in AI systems’ fairness, accuracy, and educational value is crucial for their active participation in AI-supported learning.

Our teaching-focused framing of AIT also helps clarify the mechanistic interpretation of this mediation. When teachers use innovative methods to demonstrate AI tools—for example, explicitly modeling how to design effective prompts, how to critically question AI-generated responses, and how to cross-verify outputs with authoritative sources, as promoted by Miao et al. (2021) and Vázquez-Madrigal et al. (2024)—students are more likely to perceive AI as both useful and controllable. Our data show that higher ITB is associated with higher AIT, and that higher AIT in turn predicts greater overall engagement, but the specific cognitive, emotional, and behavioral pathways through which trust operates are inferred from theory rather than decomposed in the model.

With regard to the three dimensions of engagement, prior literature suggests that AIT may operate somewhat differently across behavioral, emotional, and cognitive facets. Conceptually, students who trust AI are more likely to (a) behaviourally persist in using AI tools rather than abandoning them after initial difficulty, (b) emotionally experience less anxiety and more curiosity when interacting with AI, and (c) cognitively invest effort in exploring, evaluating, and integrating AI-generated information into their own thinking. In this study, however, LE was modeled as a single latent construct that integrates behavioral, emotional, and cognitive items. As a result, we could not statistically estimate separate paths from AIT to each engagement facet, and our findings speak to AIT’s relationship with overall learning engagement rather than its differential effects. Future research should explicitly model three distinct engagement factors to test whether AIT is more strongly tied to emotional–cognitive engagement than to purely behavioral participation.

Beyond individual-level processes, our findings also connect to institutional readiness and policy. Furthermore, Chan and Hu (2023) warned that lack of trust leads to student resistance, disengagement, or merely superficial use of powerful educational tools. Our results suggest that “conditional trust” cultivated through innovative teaching behaviors—where students understand both the appropriate contexts for AI application and the need for critical vigilance—can mitigate these risks. Effective AI integration in higher education therefore requires targeted institutional policies that enhance technology readiness, promote a culture of peer support, and address educators’ AI-related concerns, so that classroom-level trust-building efforts by teachers are reinforced rather than undermined by the broader environment.

Finally, the present findings highlight that trust in AI should not be equated with uncritical acceptance. This study underscores the importance of critical trust formed through teacher guidance, rather than blind trust, as a key factor in promoting deep learning engagement. Proactive, ethically grounded integration of AI—guided by institutional oversight, curriculum reform, and equity safeguards—is essential to ensure its educational benefits while preserving integrity and humanistic values In this sense, teacher practices that foster reflective, well-informed trust function as a micro-level expression of broader ethical governance in AI integration.

Although this study provides valuable findings, several limitations should be acknowledged. First, as described in the literature review, the sample comes from universities in the Guangdong region, and regional characteristics may affect the generalizability of the results. Future research should expand to different regions and types of institutions to test the robustness of these patterns.

Second, this study relied exclusively on self-report measures, which may increase the risk of social desirability bias and common method variance. Although we collected the data anonymously, assured participants of confidentiality, and employed well-validated scales, these procedures cannot completely rule out the possibility that respondents answered in a socially desirable manner. Future research could address this limitation by combining self-reports with additional data sources—such as classroom observations, learning analytics, behavioral indicators of system use, or peer and instructor ratings—in order to triangulate the key constructs and further enhance the robustness of the findings.

Third, the study did not distinguish the differential impacts of different types of AI tools (such as large language models like ChatGPT, AI recommendation systems, or automated scoring systems) on trust and engagement. Different tools may evoke distinct patterns of perceived usefulness, fairness, and risk, which could in turn shape AIT and engagement in different ways. Subsequent studies should therefore examine specific AI applications and compare their effects on behavioral, emotional, and cognitive engagement.

Finally, our structural models treated learning engagement as a single latent construct. Although this approach aligns with many prior studies, it limited our ability to test whether ITB and AIT relate differentially to behavioral, emotional, and cognitive engagement. Future research should employ multidimensional engagement models and longitudinal or experimental designs to more rigorously identify the causal processes through which innovative teaching behaviors and AI trust jointly shape students’ learning experiences.