What Are We Missing in AI‐Supported L2 Writing Research? A Systematic Review From 2014 to 2024

ABSTRACT Background While existing reviews have mapped the efficacy of pre‐GenAI tools or, more recently, explored GenAI research trends and user perceptions, a comprehensive synthesis focused specifically on the pedagogical strategies and instructional designs employed in empirical studies across both eras is needed. Objectives This systematic review analyzes AI‐supported L2 writing trends and examines the integration of technology, content, and pedagogy within instructional designs, guided by the TPACK framework from 2014 to 2024. Methods 61 empirical studies were analysed. We employed a dual‐coding framework, with one lens identifying general research trends and the second using the TPACK framework to examine specific instructional design elements like theoretical grounding and learning activities. Results A dramatic surge in publications occurred post‐2022. The research landscape is overwhelmingly concentrated in formal higher education settings, with a narrow instructional focus on academic writing genres. A significant “theory‐practice” pattern was identified: while Sociocultural Theory was the most cited framework, the dominant instructional format was individual activity, suggesting a reinterpretation of the AI as a “social partner” or “More Knowledgeable Other.” Over half of the studies lacked an explicit theoretical framework. AI is used primarily for language optimization and explanation, targeting the revision and editing stages, while assessment remains focused on the final writing product. Conclusions The data reveal a disconnect between technological potential and instructional practice, evidenced by the research's narrow scope (limited to formal settings and generative capabilities), often atheoretical basis, and its focus on AI as a remedial tool.