<p>Self-regulated learning (SRL) enhances academic performance and fosters lifelong learning skills, yet many students struggle to regulate their own learning effectively. Generative AI chatbots hold promise for scaffolding SRL through adaptive interaction, but most prior studies have targeted isolated SRL phases and relied on domain-specific chatbots requiring technical expertise, limiting their scalability in ordinary classrooms. To address this gap, the present study embedded SRL scaffolding into a general-purpose AI chatbot and examined its effects on students’ academic writing performance in a project-based English for Academic Purposes (EAP) context. Using a quasi-experimental pretest-posttest design with 205 undergraduates across intact classes, the study compared the experimental group that received structured chatbot prompts covering the full SRL cycle (forethought, performance, self-reflection) with a control group using the same chatbot without SRL scaffolding. The findings revealed three key outcomes. First, chatbot-assisted SRL scaffolding was associated with significant improvements in students’ writing performance, with particular gains in task fulfillment and discourse competencies, highlighting the value of comprehensive, cyclical SRL support. Second, SRL skills partially mediated the relationship between chatbot use and writing performance, suggesting that improvements occurred both directly and indirectly through strengthened self-regulation. Third, AI literacy moderated the relationship between chatbot-assisted SRL scaffolding and writing performance, with structured chatbot prompts helping compensate for lower literacy levels and thereby narrowing performance gaps. These findings contribute theoretically by extending SRL frameworks into generative AI-enhanced learning contexts. Practically, they demonstrate the feasibility of embedding SRL scaffolding in general-purpose chatbots as accessible, scalable tools for improving writing performance and fostering learner development.</p>