Öğretmenlerin yabancı dil olarak İngilizce yazmada yapay zekâ geri bildirimin doğruluğuna ve iyileştirme sürecine yönelik görüşlerinin ve geliştirme uygulamalarının incelenmesi

This multiple case study examines how English preparatory class instructors refine AI-generated feedback to render it more teacher-like in both technical and pedagogical terms, how they perceive the accuracy of AI-generated feedback for both error-coded and content-related on student essays based on their refinement insights, and how they evaluate the process of refining it. As prior studies emphasize the need for a teacher filter before classroom use, teachers' refinements, documented through color-coded and structured reviews, served as the primary data sources to explore this issue. Their holistic accuracy ratings enabled exploring the extent of required teacher intervention basing their judgements on their reviews, while the interview with teachers further supported the findings by eliciting practical suggestions for enhancing the classroom usability of AI feedback. The analysis revealed a key limitation of AI-generated error-coded feedback: excessive focus on grammatical correctness leading AI to overlook meaning, clarity, and structural issues, and more. These limitations reduced teachers' trust in error-coded feedback and increased their reluctance to use it, as they found the process of refinement more burdensome. In contrast, content feedback often missed repetitive ideas due to its lack of focus on holistic, paragraph-level processing. However, teachers found content feedback more useful, noting it would be fully effective only when supplemented with priority areas specific to each student's needs. These findings can inform future efforts to enhance the accuracy and classroom applicability of AI-generated feedback. Keywords: AI-generated writing feedback, EFL Teachers' refinement practices, Suggestions for improvement, Accuracy evaluations