The accelerated integration of generative artificial intelligence (AI) into higher education has created an urgent need for teachers who can effectively leverage AI to create individualised and responsive learning experiences (Abbasi et al., 2024). However, despite significant institutional investments in professional development (PD), evidence across Spain (Galindo-Domínguez et al., 2024), Hungary (Dringó-Horváth et al., 2025), and the UK (Atkinson-Toal & Guo, 2024) suggests that the majority of in-service teachers lack confidence in and ability of integrating AI into pedagogical practices. This persistent gap suggests that PD programmes alone may be insufficient, pointing to the need for research examining how broader workplace conditions, such as leadership practices, collegial relationships, and institutional support structures, shape teachers’ beliefs and practices.

Social Cognitive Theory (SCT) provides a useful lens for understanding part of this phenomenon. According to Bandura (1986), self-efficacy beliefs develop through the reciprocal interaction of personal cognition, environmental factors, and behavioural experiences. Prior research has established that transformational leadership positively predicts a range of teacher outcomes, including knowledge sharing (Hoang & Le, 2024), engagement with learning cultures (Long & Xia, 2025), and collaborative behaviours (Schmitz et al., 2025). In particular, Liu and Hallinger (2024) noted that middle leaders may influence teachers more directly than principals do in higher educational contexts. However, the interplay of these factors within the nascent context of AI integration in higher education has yet to be fully elucidated. Moreover, existing research has largely assumed that teacher collaboration uniformly benefits professional outcomes, yet this assumption warrants empirical scrutiny in the context of emerging technologies where collaborative norms may not be established.

This research addresses these gaps by examining a mediation model in which transformational leadership enacted by academic associate deans predicts teachers’ AI competence self-efficacy, with perceived organisational support and teacher collaboration as potential mediators. This research is conducted in China, where the digital transformation reform of higher education has required teachers to integrate AI into routine work in the past five years. Characterised by hierarchical administrative structures and collectivist cultural values, the Chinese context offers a distinctive setting that may reveal mechanisms obscured in Western contexts and provide a comparative reference point for European educators.

Using an explanatory sequential mixed-methods design, we first collected survey data from 405 teachers across five universities in one Chinese province. Squares Structural Equation Modelling (PLS-SEM) revealed that transformational leadership (TL) was a strong predictor of perceived organisational support (OS) (β=0.681, p<.001), teacher collaboration (TC) (β=0.310, p<.001), and teachers’ AI competence self-efficacy (AI_SE) (β=0.308, p<.001). Mediation analysis confirmed OS as a significant indirect pathway (β=0.127). However, contrary to expectations, TC demonstrated weak mediating effects (VAF=0.121), suggesting a potential disconnect between collaborative activity and self-efficacy development, a pattern we term the “collaborative paradox”.

Given that the distinct pedagogical environments of STEM versus non-STEM disciplines may moderate the influence of TL, a multi-group analysis was conducted to rigorously examine divergences in the structural paths between TL and key outcomes. STEM teachers showed significantly stronger associations between TL and all outcome factors when compared to non-STEM teachers (TL to OS: β_STEM=0.836 vs. β_Non-STEM=0.423, p<.001; TL to TC: β_STEM =0.604 vs. β_Non-STEM=0.183, p<.001; TL to AI_SE: β_STEM=0.589 vs. β_Non-STEM=0.171, p=.012). In contrast, TC predicted AI_SE only among non-STEM teachers (β_Non-STEM=0.352 vs. β_STEM=-0.020, p_diff=.025).

The qualitative analysis identified four themes that clarify mechanisms underlying these patterns: middle leaders’ roles in facilitating cognitive reframing and resource gatekeeping, an interaction-transformation gap in collaboration, and discipline-based cognitive logic filtering. These findings suggest that STEM teachers’ self-efficacy derives primarily from mastery experiences and vertical institutional support, whereas non-STEM teachers rely more heavily on horizontal peer relationships for emotional scaffolding against technology-related anxiety.

This study employed an explanatory sequential mixed-methods design, in which quantitative findings guided subsequent qualitative inquiry. Three research questions guided the investigation: RQ1: To what extent does transformational leadership predict teachers’ AI competence self-efficacy in Chinese universities? RQ2: Do perceived organisational support and teacher collaboration mediate the relationship between transformational leadership and teachers’ AI competence self-efficacy? RQ3: Do these relationships vary across disciplinary contexts (STEM vs. non-STEM)? In the quantitative phase, survey data were collected from 405 teachers from five comprehensive universities located within one province in eastern China. Participants were recruited through institutional contacts, yielding a convenience sample. While this sampling strategy limits generalisability beyond the study context, it enables access to a sufficient sample size for the planned analyses while controlling for regional policy variation. The study included four scales: the Teachers AI Competence Self-efficacy Scales (Chiu et al., 2024); the Global Transformational Leadership Scale (Carless et al., 2000); the Perceived Organisational Support Scale (Eisenberger et al., 1997); and the Teacher Collaboration Scale (Geijsel et al., 2009). Data were analysed using PLS-SEM, which was chosen for its suitability for predictive modelling and its robustness with complex models involving multiple mediators (Hair et al., 2019). Multi-group analysis examined whether path coefficients differed significantly between STEM and non-STEM teachers. The qualitative phase sought to explain the mechanisms underlying the quantitative findings particularly the unexpected weakness of the collaboration pathway (Cohen et al., 2002). Using purposive sampling informed by maximum variation principles, we recruited 11 teachers representing diverse disciplines, career stages, and levels of AI engagement. Participants were initially identified through survey respondents who indicated willingness to be interviewed. Additional participants were recruited through snowball referrals to ensure disciplinary diversity. Semi-structured interviews, lasting 30 to 60 minutes each, explored participants’ experiences with leadership, collaboration, and AI integration. Data were analysed using constructivist grounded theory methods (Charmaz, 2014), proceeding through initial, focused, and theoretical coding. This iterative process yielded four core themes that illuminated the quantitative patterns. This sequential design strengthens the study’s validity by enabling triangulation: qualitative findings contextualise and explain statistical relationships, while quantitative results ensure that qualitative interpretations are grounded in broader patterns (Cohen et al., 2002).

This study offers three principal contributions to understanding how organisational conditions relate to teachers’ AI competence self-efficacy. First, the findings highlight TL as a robust predictor operating through multiple pathways, with perceived OS serving as the primary mediating mechanism. Interpreted through SCT, this suggests that middle leaders shape teachers’ self-efficacy beliefs not only through direct encouragement but also by cultivating environments perceived as supportive of professional risk-taking and experimentation with new technologies. Second, the weak mediating role of TC challenges prevailing assumptions that collegial interaction benefits technology integration. The qualitative findings suggest that collaboration around AI remains largely superficial, characterised by information exchange rather than deep pedagogical inquiry. Thus, it fails to provide the vicarious experiences or social persuasion that SCT identifies as sources of self-efficacy. This “collaboration paradox” invites reconsideration of how institution’s structure collaborative opportunities for emerging technology domains. Third, the pronounced disciplinary heterogeneity carries practical implications. For STEM teachers, whose self-efficacy appears grounded in mastery experiences and vertical support structures, institutions might prioritise hands-on experimentation opportunities and clear administrative endorsement. For non-STEM teachers, who benefit more from peer relationships, facilitated learning communities and mentorship programmes may prove more effective. Several limitations warrant acknowledgment. The cross-sectional design precludes causal inference. Longitudinal research is needed to establish temporal precedence in the future. The convenience sample from one Chinese province limits generalisability, and self-reported self-efficacy may not correspond to actual competence or classroom practice. Additionally, the collectivist, high power distance context of Chinese higher education may amplify leadership effects in ways that differ from European settings. Despite these limitations, this study provides a theoretically grounded, empirically supported framework that European educators and policymakers can use as a comparative reference when designing institution-level strategies to support teachers’ AI integration.

Abbasi, B. N., Wu, Y., & Luo, Z. (2024). Exploring the impact of artificial intelligence on curriculum development in global higher education institutions. Education and Information Technologies, 30(1), 547-581. https://doi.org/10.1007/s10639-024-13113-z Atkinson-Toal, A., & Guo, C. (2024). Generative Artificial Intelligence (AI) Education Policies of UK Universities. Enhancing Teaching and Learning in Higher Education, 2, 70-94. Bandura, A. (1986). Social foundations of thought and action. Englewood Cliffs, NJ, 1986(23-28), 2. Carless, S. A., Wearing, A. J., & Mann, L. (2000). A short measure of transformational leadership. Journal of business and psychology, 14(3), 389-405. Charmaz, K. (2014). Constructing grounded theory (introducing qualitative methods series). Constr. grounded theory. Chiu, T. K. F., Ahmad, Z., & Çoban, M. (2024). Development and validation of teacher artificial intelligence (AI) competence self-efficacy (TAICS) scale. Education and Information Technologies, 30(5), 6667-6685. https://doi.org/10.1007/s10639-024-13094-z Cohen, L., Manion, L., & Morrison, K. (2002). Research methods in education. routledge. Dringó-Horváth, I., Rajki, Z., & T. Nagy, J. (2025). University teachers’ digital competence and AI literacy: Moderating role of gender, age, experience, and discipline. Education Sciences, 15(7), 868. Eisenberger, R., Cummings, J., Armeli, S., & Lynch, P. (1997). Perceived organizational support, discretionary treatment, and job satisfaction. Journal of Applied psychology, 82(5), 812. Galindo-Domínguez, H., Delgado, N., Losada, D., & Etxabe, J.-M. (2024). An analysis of the use of artificial intelligence in education in Spain: The in-service teacher’s perspective. Journal of digital learning in teacher education, 40(1), 41-56. Geijsel, F. P., Sleegers, P. J., Stoel, R. D., & Krüger, M. L. (2009). The effect of teacher psychological and school organizational and leadership factors on teachers' professional learning in Dutch schools. The elementary school journal, 109(4), 406-427. Hoang, T. N., & Le, P. B. (2024). The influence of transformational leadership on knowledge sharing of teachers: the roles of knowledge-centered culture and perceived organizational support. The Learning Organization, 32(2), 328-349. https://doi.org/10.1108/tlo-08-2023-0144 Liu, S., & Hallinger, P. (2024). The effect of department leadership on teacher professional learning in China: A multilevel moderated mediation model. Educational Management Administration & Leadership. https://doi.org/10.1177/17411432241232541 Long, Y., & Xia, Y. (2025). Leveraging transformational and instructional leadership for teacher professional development: A dual-mediation model of teacher self-efficacy and organisational culture. Teaching and Teacher Education, 164. https://doi.org/10.1016/j.tate.2025.105087 Schmitz, M.-L., Antonietti, C., Consoli, T., Gonon, P., Cattaneo, A., & Petko, D. (2025). Enhancing teacher collaboration for technology integration: the impact of transformational leadership. Computers & Education, 234. https://doi.org/10.1016/j.compedu.2025.105331