Abstract
Academic writing remains one of the most cognitively demanding skills for English as a Foreign Language (EFL) learners, whose difficulties with linguistic accuracy, structural coherence, and lexical precision are compounded by limited access to individualized feedback in resource-constrained educational settings. This study evaluates the potential impact of Dola AI—a conversational agent powered by Large Language Models (LLMs)—on the academic writing performance of secondary-school EFL learners in Ecuador. A quasi-experimental, non-equivalent control group design with pre-test and post-test assessments is proposed, with a stratified sample of 120 ninth- and tenth-grade students (mean age 14.8 years; proficiency levels A2–B1 according to the Common European Framework of Reference for Languages, CEFR) from Quintiliano Sánchez Rendón School. The experimental group (n = 60) will engage in Dola AI–mediated writing activities over twelve weeks, while the control group (n = 60) will receive conventional instruction. Writing performance will be evaluated across four dimensions—linguistic accuracy, structural coherence, lexical complexity, and rhetorical appropriacy—using a validated analytical rubric (maximum score 100; inter-rater reliability assessed via Intraclass Correlation Coefficient, ICC = .87, 95 % CI [.78, .92]). It is hypothesized that AI-mediated scaffolding will be associated with moderate-to-large improvements across all four writing dimensions. Learners will be encouraged to integrate Dola AI into the pre-writing, drafting, and revision stages as a formative support tool rather than a content-generation surrogate. Implications for curriculum design, teacher training, and responsible AI integration are discussed.

This publication has 14 references indexed in Scilit: