The Effects of AI‐Powered Genre‐Based Writing Instruction on Low‐Proficiency EFL Writers' Ideal L2 Writing Self, Writing Enjoyment and <scp>L2</scp> Writing Performance: A Quasi‐Experimental Mixed‐Methods Study

ABSTRACT This mixed‐methods study examined the effects of DeepSeek‐powered genre‐based instruction (GBI) on argumentative writing performance, ideal L2 writing self, and writing enjoyment among 71 low‐proficiency EFL undergraduates. A quasi‐experimental design was employed, with an experimental group ( N = 32) receiving 6 weeks of AI‐powered GBI and a control group ( N = 39) receiving traditional instruction. Quantitative data included writing tests (holistic and CAF indices) and motivation questionnaires, while qualitative data came from semi‐structured interviews. The experimental group significantly outperformed the control group in holistic writing scores ( F = 45.38, p &lt; 0.001, = 0.40) and across multiple CAF indices. They also showed significantly enhanced ideal L2 writing self and writing enjoyment ( p &lt; 0.001, large effect sizes). Thematic analysis revealed enhanced writing confidence, perceived quality improvement, and concerns about AI over‐reliance. The findings demonstrate that AI‐powered GBI effectively enhances writing outcomes and motivational dispositions of low‐proficiency EFL learners, while highlighting the need for teacher‐mediated critical AI literacy.