Modeling AI-assisted writing: How self-regulated learning influences writing outcomes

Abstract

Academic writing is essential to academic and professional success, yet remains a challenge for many students. Artificial intelligence (AI) offers a potential solution, but most research on that possibility has focused on final written products rather than on the writing process. This study helps to fill that gap by modeling how key variables interact in generative AI-assisted writing processes, based on survey data from 1073 postgraduate students from 21 countries studying in the UK. We used structural equation modeling to categorize AI use into three levels, from basic to advanced: 1) for technical support, 2) for text development, and 3) for transformation. Self-regulated learning (SRL) strategies positively predicted all three types of AI use. Notably, while the most advanced use of AI (i.e., for writing transformation) significantly enhanced outcomes including critical thinking, motivation, and writing quality, whereas the most basic use (for technical support) did not predict such outcomes. This study further revealed that AI self-efficacy and writing self-efficacy were significant antecedents of self-regulation, suggesting the importance of supporting students’ self-efficacy in boosting self-regulation in AI use. This suggests that the key to writing-outcome improvement may not be to teach students different uses of AI, but to develop their self-regulation to the point that they can independently explore and apply advanced uses of this technology.