An Observational Study on Challenges of Using AI tools in EFL Writing

This qualitative observational study examines the integration of AI tools (DeepSeek/ChatGPT) in tertiary EFL writing courses. Over a 6-week period, we tracked 87 Chinese learners’ experiences using classroom observations and semi-structured interviews, addressing three research questions: 1) What cognitive and behavioral challenges emerge during AI-mediated writing? 2) How does AI usage influence textual diversity? And 3) What competence gaps hinder effective AI integration in EFL writing? Findings reveal pervasive cognitive conflicts in reconciling AI outputs with original ideas (e.g., 82% reported voice appropriation concerns), textual homogenization in vocabulary and sentence structure (77% lexical overlap), and critical competence gaps in technical, evaluative, and ethical AI application (74%/68%/31% deficit rates). Notably, while AI enriched topic vocabulary (e.g., 33% lexical uniqueness gain via personalized prompts), it risked eroding learners’ voice agency. Furthermore, early technology dependency (avg. 4.33 self-initiated uses/session) shifted to critical avoidance (1.49 uses) as learners confronted AI’s limitations. We argue for pedagogically scaffolded AI training—emphasizing prompt personalization, critical evaluation, and ethical frameworks—to balance efficiency with originality. Implications for EFL writing curricula and teacher development are discussed.