Abstract
The rapid expansion of generative artificial intelligence (GenAI) in education has intensified debates about authorship, feedback, and the role of human judgment in writing instruction. While emerging research has examined AI-generated feedback in higher education and experimental contexts, fewer studies have investigated how rubric-aligned GenAI systems function within the practical realities of K–12 classrooms. This article presents a practice-oriented account of the implementation of CyberScholar, a rubric-aligned GenAI feedback tool introduced in four US middle school classrooms. Rather than evaluating learning outcomes or establishing causal impact, this work documents design decisions, ethical safeguards, teacher preparation, classroom mediation, and overall patterns of student engagement observed during implementation. Drawing on classroom observations, student and teacher surveys, focus group discussions, and platform interaction traces, the article highlights both affordances and limitations of rubric-aligned AI feedback. Findings suggest that such systems may increase the visibility of assessment criteria during drafting and extend iterative revision opportunities, but their pedagogical value is conditional upon teacher mediation, critical AI literacy, manageable feedback density, and institutional safeguards.