Answers are grounded in this paper and cite their sources.
The rapid expansion of artificial intelligence (AI)–based writing tools in education has outpaced systematic understanding of how pedagogical intentions, theoretical commitments, and technical customization are translated into AI system design. Customization is conceptualized here as the deliberate modification of AI system behavior to align outputs with pedagogical intent. In response to this gap, this systematic review synthesizes 23 empirical studies from recent AI-in-education research to examine how AI systems are customized for writing pedagogy. Specifically, the review analyzes patterns of alignment among pedagogical goals, theoretical principles, customization methods, and implementation modes. Findings indicate that although pedagogical aims have increasingly shifted beyond surface-level correction toward writing processes, feedback literacy, and higher-order development, customization practices remain predominantly driven by performance-oriented theories and technical optimization. Prompt engineering and fine-tuning dominate existing approaches, with limited evidence that learning-centered, sociocultural, or developmental theories are operationalized within AI system behavior. Across studies, customized AI systems demonstrate benefits in improving writing quality, learner engagement, and assessment reliability; however, persistent challenges related to feedback inaccuracy, learner overreliance, and ethical transparency are consistently reported. By identifying systematic misalignments and underexplored research spaces, particularly in longitudinal design, interactional analysis, and theory-informed customization, this review reveals a structural gap between educational aspirations and current AI design practices, specifically the disconnection between learning-oriented pedagogical goals and the performance-driven theories and modular customization techniques that predominantly operationalize them. The findings underscore the need to move beyond automation-focused models toward developmentally oriented, theoretically grounded AI systems capable of supporting sustained learning in writing pedagogy.
No takes yet. Share an insight, caveat, or question.
Ying Luo (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: