Students’ perceptions of generative AI in EFL writing: Strategies, self-efficacy, satisfaction and behavioural intention

Abstract

This study examined students’ perceptions and attitudes towards generative artificial intelligence (AI) tools in language learning, particularly in English as a Foreign Language (EFL) writing. Employing a cross-sectional design, data were collected from 399 Saudi university EFL students to assess self-reported associations between AI chatbots and writing-related attitudes. The findings indicate that these chatbots were associated with increased self-efficacy, satisfaction and behavioural intention among participants. Participants reported high levels of satisfaction and self-efficacy, moderate levels of AI writing strategies (operationalised usefulness) and strong intention to reuse. The structural equation model results showed that AI writing strategies were significantly associated with satisfaction (β = .784), self-efficacy (β = .525) and behavioural intention (β = .353), where satisfaction and self-efficacy served as mediators of the strategies–intention relationship. These findings highlight the potential of AI to support confidence, develop attitudes and possibly grow skills for second language writing, particularly with younger people who are familiar with technology. Because outcomes were cross-sectional and self-reported, these estimates represent associations that should not be viewed as causal effects. The findings indicate the promise of AI to cultivate students' confidence and positive attitudes towards second language writing, especially when working with younger university learners with a high level of digital exposure. Implications for practice or policy: Universities should integrate approved generative AI into writing centres and language labs with short disclosures and reflective rationales. Departments should design assessments and modules on permitted, ethical and non-coercive AI uses such as brainstorming, vocabulary and revisions. Low-stakes AI-assisted practice should be encouraged, while high-stakes assessments remain under instructor control. Institutions should publish course-level AI policies outlining permitted and forbidden uses, disclosure requirements, privacy considerations and integrity standards.

Ähnliche Arbeiten