To revise or to recommend: an exploratory sequential mixed method study on GenAI formative feedback on Reading assessment in secondary schools

Purpose: While evaluation-based feedback has been widely advocated in language assessment research for its perceived utility, this study examined and compared the effectiveness of three types of GenAI formative feedback within secondary education reading assessments. Design/methodology/approach: This study employed an exploratory sequential mixed methods design, guided by the Student-Feedback Interaction Model. Six teachers evaluated GenAI formative feedback provided to 516 Year 9 students on constructed-response reading tasks. Feedback types included: (1) answer revisions, (2) reading strategy recommendations, and (3) additional task recommendations. Qualitative data were collected from post-evaluation interviews. Findings: Contrary to prevailing feedback research that positions recommendation as more impactful than revision, MANCOVA tests reveal GenAI revision feedback was rated significantly more effectively than recommendation feedback across classical, fiction, and technical writing reading genres. Qualitative data from post-evaluation interviews revealed key limitations of GenAI’s recommendation on reading strategies and tasks: (1) repetition and copied materials; (2) limited development of knowledge transfer; (3) irrelevance to high-stakes exams; (4) generic and non-contextualized recommendations; and (5) overly complex language for lower proficiency students. Originality/value: This study provides the first empirical evidence demonstrating that GenAI formative feedback proves more effective for answer revision than strategy recommendation. It advances genre-specific insights by revealing GenAI’s greater efficacy in classical literature tasks compared to fiction or technical texts. Practically, the research proposes a feedback-literate GenAI integration model that addresses five limitations of recommendation-based approaches (including repetitiveness, transferability gaps, and exam misalignment), offering educators actionable principles to implement AI tools critically while maintaining pedagogical intentionality.