Structure as scaffolding: How rubric-anchored LLM feedback enhances revision engagement in academic argumentation

Generative AI can provide rapid, detailed feedback for student writing, but long unstructured comment lists may be difficult to use during short classroom revision windows. This study examined whether organizing comparable-priority large language model (LLM) feedback under rubric-aligned headings supports stronger revision engagement than the same feedback presented as a flat list and a criteria-aligned self-revision checklist. An intact-class quasi-experiment was conducted in an undergraduate English for Academic Purposes course (N = 105; three intact classes, one per condition). Students completed two timed classroom revision tasks with rubric-anchored LLM feedback (RA), matched-priority non-anchored LLM feedback (NA), or checklist-based self-revision, and outcome patterns were interpreted descriptively using adjusted posttest means. Process evidence was triangulated through revision traces for all students and suggestion-level uptake coding in a stratified subsample. Adjusted posttest means showed a stable ordering (RA > NA > control). Revision traces indicated more extensive between-draft change in both LLM conditions than in the checklist condition, with the largest magnitudes under RA. Uptake coding further suggested somewhat stronger criterion-referenced enactment under RA, especially for higher-order argument-related revisions. The findings position feedback organization as a practical design lever for making AI-generated writing feedback more usable in classroom revision.