From AI Feedback to Peer Reflection: A Dual-Phase Reflective Writing Loop Model Using a Generative AI-Assisted Collaborative Dialogic Feedback

Abstract

In AI-driven writing education, EFL students increasingly
rely on automated feedback; however, most AI feedback
systems remain limited to surface-level corrections and
provide insufficient support for metacognitive regulation.
To address this limitation, this study proposes a
Generative AI-assisted Collaborative Dialogic Feedback
(GenAI-CDF) model that integrates metacognitive theory with
the academic writing process through two AI-supported
phases: an AI as tutor phase for individualized reflection
and an AI as peer phase for dialogic co-construction. Using
a quasi-experimental design, the effectiveness of the
GenAI-CDF model was compared with a conventional
ChatGPT-assisted dialogic feedback model (GPT-DF) across
three writing tasks: brainstorming, outlining, and
drafting. Results from ANCOVA and one-way ANOVA indicated
that students in the GenAI-CDF group significantly
outperformed those in the GPT-DF group in writing
performance, critical thinking, and metacognitive ability.
These findings demonstrate that aligning AI feedback with
metacognitive regulation transforms AI from a corrective
tool into a collaborative learning partner that actively
supports higher-order thinking and sustained engagement in
academic writing.