Embedding constrained, learner-initiated GenAI feedback in collaborative L2 writing: Targeted grammar gains with a two-week follow-up

Abstract

Research on generative artificial intelligence (GenAI) in second-language writing has largely examined commercial tools such as Grammarly or general-purpose chatbots, typically in individual workflows. Evidence remains limited on purpose-built classroom systems that embed constrained, learner-initiated GenAI feedback within technology-mediated collaborative writing, align that feedback with explicitly taught forms, and assess outcomes beyond an immediate posttest. This study evaluated a pedagogical intervention using Collabowrite, a custom web application for collaborative story writing. Its core GenAI feature provides form-prioritized feedback as local edit suggestions with brief first-language explanations and requires learners to implement changes manually. Participants were 36 undergraduates from four intact classes of a compulsory two-year English-as-a-foreign-language program at a private university in Japan that does not offer an English major. A one-group pretest-posttest design with a two-week delayed posttest assessed performance using three rotated parallel 40-item recognition-focused multiple-choice grammar tests targeting consistent tense, pronouns, conjunctions, and articles. Mixed-effects models indicated higher total grammar test scores at posttest and delayed posttest than at pretest, with no clear difference between posttest and delayed posttest. Because the study used a one-group pretest-posttest design without a comparison group, the observed gains cannot be attributed specifically to GenAI and may reflect regular instruction, test practice, or novelty effects. The study contributes classroom implementation evidence for a pedagogical configuration that combines explicit focus on form, meaning-focused collaborative writing, and learner-initiated, form-prioritized GenAI feedback, and provides a basis for controlled follow-up work.