Approaches to writing play an important role in both the writing processes and outcomes. However, little is known about whether L2 writers adopt different combinations of approaches in academic writing contexts and what factors predict such combinations. Hence, this study aimed to identify different profiles of approaches to writing in an L2 academic context and examine how they are predicted by writing self-efficacy and large language model (LLM) acceptance. To this end, a total of 578 Chinese graduate students were recruited to participate in the study. Latent profile analysis revealed three distinct writing profiles: unorganized (Profile 1), dissonant (Profile 2), and deep and organized (Profile 3), with the majority of students categorized under the dissonant profile. Additionally, multinomial logistic regression analysis revealed that writing self-efficacy positively predicted profile membership, with the strongest effect observed for Profile 3, followed by Profile 2 and then Profile 1. LLM acceptance also positively predicted profile membership, with the strongest effect for Profile 2, followed by Profile 3 and then Profile 1.

Self-efficacy plays a central role in L2 academic writing. However, limited attention has been paid to how technology-related factors and emotions jointly shape learners' self-efficacy in L2 academic writing. Drawing on control-value theory and self-efficacy theory, this study investigates the relationships among large language model (LLM) acceptance, enjoyment, anxiety, and self-efficacy in L2 academic writing. A total of 643 Chinese EFL graduate students participated in the study, with data collected via an online questionnaire. Structural equation modelling was employed to examine hypothesized direct and indirect relationships. The results of the direct effects analysis showed that LLM acceptance positively predicted self-efficacy, enjoyment, and anxiety. Moreover, enjoyment positively predicted self-efficacy, whereas anxiety negatively predicted it. The results of the indirect effects analysis further revealed that enjoyment served as a complementary mediator, whereas anxiety acted as a competitive mediator in the link between LLM acceptance and self-efficacy. These findings highlight the dual emotional pathways through which LLM acceptance shapes self-efficacy and offer pedagogical insights into fostering students' self-efficacy in the L2 academic writing context.

This study investigates the psychological mechanisms underlying engagement in AI-mediated EFL writing from a social cognitive theory perspective. Specifically, it examines how growth mindset, ideal L2 self, and enjoyment relate to engagement. An explanatory sequential mixed-methods design was employed. In the quantitative phase, survey data were collected from 519 university EFL learners from ethnic minority regions in China and analyzed using structural equation modeling. In the qualitative phase, semi-structured interviews were conducted with 20 participants to elaborate on learners’ experiences and explain the quantitative results. The quantitative findings revealed significant positive associations among growth mindset, ideal L2 self, enjoyment, and engagement in AI-mediated EFL writing. Mediation analyses further indicated that growth mindset was indirectly related to engagement through ideal L2 self and enjoyment, including both single-mediator pathways and a chained pathway via ideal L2 self followed by enjoyment. The qualitative findings further showed that learners maintained engagement by reframing challenges and feedback as learning opportunities, connecting writing to future self-guides, and experiencing enjoyment during iterative revision. This study advances understanding of engagement in AI-mediated EFL writing by highlighting the interplay between personal beliefs, motivational self-guides, and affective experiences, and it offers practical implications for designing AI-supported writing instruction that fosters sustained learner involvement.

© 2026 Optica Publishing Group. All rights, including for text and data mining (TDM), Artificial Intelligence (AI) training, and similar technologies, are reserved.