Effects of DeepSeek-assisted Writing Instruction on EFL Learners' Writing Performance

To investigate the effects of DeepSeek assisted writing instruction on different dimensions of learners’ English writing skills, a prospective, quasi-experimental study was conducted in 2025. Employing a mixed research design, the study involved 31 students over six weeks. The experiment consisted of three stages: baseline period, intervention period and follow-up period. During the intervention period DeepSeek was employed to assist EFL learners’ writing. Learners’ writing production of each period was collected and rated in a randomized fashion to minimize grading bias. Then a semi-structured interview was conducted to further elucidate proficiency-dependent perceptions. Results revealed that DeepSeek significantly enhanced micro-level writing skills (i.e., vocabulary, grammar, mechanics) across all proficiency groups. However, higher-proficiency learners demonstrated greater gains in macro-level performance, while lower-proficiency learners exhibited over-reliance on technical corrections. Thematic analysis highlighted divergent attitudes: advanced learners valued DeepSeek's in-depth logical reasoning scaffolds, whereas lower-proficiency learners valued error reduction and were confounded by cognitive over-load. These findings underscore the necessity of proficiency-tailored AI assistance together with teacher guidance and peer collaboration in EFL writing pedagogy.