The research team built its conceptual model on two well-established theoretical pillars. The first is the Unified Theory of Acceptance and Use of Technology, known as UTAUT, which explains technology adoption through four core determinants: performance expectancy, the belief that a tool improves one’s performance; effort expectancy, the perceived ease of using it; social influence, the sense that important others expect one to use it; and facilitating conditions, the perception that infrastructure and support exist to make use possible. The second pillar is Social Cognitive Theory, which frames human behavior as the product of personal beliefs, environmental factors, and actions interacting in a continuous loop. By fusing these frameworks, the researchers could trace a full causal pathway: from the psychological and contextual drivers of generative AI adoption, through actual use of the tools, and onward to observable innovative behavior, with mediating variables operating in the middle of the chain.

Data came from an online survey administered in China, which yielded 291 valid responses from Generation Z students in higher education. The sample size is modest by big-data standards but well suited to the analytical strategy the team deployed. Ethics approval was granted by the Academic Committee of the School of Economics and Management at the Institute of Disaster Prevention in February 2025, and informed consent was obtained from all participants, who were told their involvement was voluntary and that they could withdraw at any time. The survey instrument captured students’ perceptions across the UTAUT determinants, their actual generative AI use, their knowledge-sharing habits, their creative self-efficacy, and their self-reported innovative behavior, allowing the researchers to model the entire adoption-to-innovation pathway in a single framework.

The analytical approach is one of the study’s most distinctive technical contributions. Rather than relying on a single statistical technique, the team executed a dual-stage analysis combining partial least squares structural equation modeling, or PLS-SEM, with an artificial neural network, or ANN. PLS-SEM is a variance-based technique that estimates networks of simultaneous regression equations linking latent constructs, which are abstract psychological variables like performance expectancy that cannot be observed directly but are inferred from multiple survey indicators. This makes PLS-SEM ideal for testing whether a hypothesized causal structure fits the data. Its limitation, however, is that it assumes the relationships between constructs are linear and additive. Human psychology rarely cooperates with such assumptions, and that is precisely where the neural network enters the picture.

Artificial neural networks are machine-learning models inspired loosely by the architecture of biological brains. They consist of layers of interconnected nodes that transform inputs into outputs through weighted connections and nonlinear activation functions, allowing them to capture complex, nonlinear patterns that classical regression equations miss. In this study, the ANN served as a complementary second stage: after PLS-SEM confirmed which pathways were statistically significant, the neural network ranked the relative importance of the predictors of innovative behavior without imposing linearity constraints. This dual-stage design, pairing a theory-driven structural model with a data-driven learning model, is increasingly seen as best practice in behavioral research, because it lets researchers test causal theory while also letting the data reveal nonlinear effects and importance rankings that the structural model alone would conceal.

The PLS-SEM results affirmed all four UTAUT determinants as positive drivers of generative AI adoption. Performance expectancy, effort expectancy, social influence, and facilitating conditions each significantly increased the likelihood that students adopt these tools, which is consistent with the broader technology-acceptance literature but notable in the generative AI context, where the tools are conversational, iterative, and qualitatively different from earlier educational software. More consequentially, actual generative AI use significantly enhanced students’ innovative behavior, confirming that adoption is not merely a matter of convenience but a genuine catalyst for generating, promoting, and realizing new ideas in academic work. The pathway from use to innovation, in other words, held up under rigorous structural testing.

The study then probed what operates between use and innovation, and two mediators emerged. Knowledge sharing, the extent to which students exchange ideas, insights, and information with peers, was identified as a positive driver of innovative behavior and a mediator of the relationship between generative AI use and innovation. Creative self-efficacy, a student’s belief in their own capacity to produce creative outcomes, played the same dual role. Both findings enrich the understanding of these mediators in educational settings: generative AI use appears to stimulate innovation partly by opening channels of knowledge exchange and partly by strengthening students’ confidence in their own creative abilities. The tools do not simply hand students finished answers; they seem to activate social and psychological processes that make students more inclined and more able to innovate.

The ANN analysis delivered the study’s most striking technical insight. When the neural network ranked the predictors of innovative behavior, creative self-efficacy emerged as the pivotal factor, underscoring its central role in fostering innovation among Generation Z students. This nonlinear importance ranking complements the structural findings: while PLS-SEM confirmed that generative AI use, knowledge sharing, and creative self-efficacy all significantly predict innovative behavior, the ANN revealed that the personal belief in one’s own creativity carries exceptional weight in the innovation equation. For educators, this suggests that the psychological lever matters as much as the technological one, and that interventions aimed at building students’ creative confidence may amplify the innovative payoff of generative AI adoption more effectively than simply providing access to the tools.

The implications for universities are actionable and timely. The confirmation that facilitating conditions positively influence adoption means that infrastructure, institutional support, and accessible training are prerequisites, not afterthoughts. The mediating roles of knowledge sharing and creative self-efficacy suggest that universities should design learning environments where generative AI use is paired with collaborative exchange and confidence-building exercises, rather than treated as a solitary activity. If the goal is to cultivate innovative behavior, the study indicates that effective integration requires attention to the full psychological chain: expectancy beliefs that motivate adoption, actual use that catalyzes innovation, and the social and self-belief mechanisms that carry the innovative impulse from tool to outcome.

The research also carries honest limitations that temper its conclusions. The cross-sectional survey design captures perceptions at a single point in time, so causal claims rest on the theoretical model rather than longitudinal observation, and self-reported measures of innovative behavior may not fully align with objectively assessed creativity. The sample of 291 students in China, while adequate for the analytical techniques, may not generalize to all Generation Z populations or all educational systems. The authors note that the study advances the burgeoning literature on generative AI in education by highlighting its capacity to cultivate innovative behavior among students and by offering insights for its effective integration into learning environments. As universities continue to debate the role of these tools, this dual-stage analysis offers a technically rigorous data point: the question is no longer only whether students will use generative AI, but whether institutions will build the conditions, the sharing cultures, and the creative confidence that turn that use into genuine innovation.

Subject of Research: The impact of generative AI adoption on the innovative behavior of Generation Z students in higher education

Article Title: Generative AI adoption and innovative behavior among Gen Z students: a deep-learning based dual-stage PLS-SEM-ANN analysis

Article References: Zhang, J., Quoquab, F., Xu, Z., Zhang, J., Mohammad, J., & Yan, B. (2026). Generative AI adoption and innovative behavior among Gen Z students: a deep-learning based dual-stage PLS-SEM-ANN analysis. Humanities and Social Sciences Communications. https://doi.org/10.1057/s41599-026-09524-7

Image Credits: AI Generated

DOI: 10.1057/s41599-026-09524-7

Keywords: generative AI, Generation Z, higher education, innovative behavior, UTAUT, Social Cognitive Theory, PLS-SEM, artificial neural network, creative self-efficacy, knowledge sharing, technology adoption, China