As generative AI becomes increasingly present in language classrooms, AI literacy in English language education is often framed in terms of technical skills, tool use, or academic integrity, implicitly positioning AI use in terms of success or failure. This study adopts a practice-oriented perspective to examine how EFL students engage in academic writing when composing with generative AI in classroom-based CALL contexts. Drawing on qualitative data from approximately 90–100 undergraduate students, the study analyzes students’ prompts, paired non-AI and AI-assisted drafts, and observable writing decisions in introduction and conclusion tasks embedded in regular academic writing coursework. Rather than evaluating writing quality or AI performance, the analysis focuses on how students shape AI assistance, negotiate control over writing decisions, and respond when AI support aligns—or fails to align—with their writing intentions. Using an inductive–abductive analytic approach, the findings identify recurring patterns of appropriation, modification, and resistance to AI-generated content, suggesting that what counts as “successful” AI use emerges through situated writing practices rather than technological outcomes alone, with implications for CALL pedagogy and writing instruction.