Mentor or Examiner? A Critical Discourse Analysis of Ai-Generated Feedback in EFL Writing Education

Abstract

This study addresses a critical gap in understanding how artificial intelligence (AI) systems construct evaluative discourse in language education contexts. While AI-powered feedback tools increasingly supplement or replace human assessment in English as a Foreign Language (EFL) writing instruction, limited attention has been paid to the discursive mechanisms through which these systems position learners, construct authority, and shape pedagogical relationships. Drawing on Critical Discourse Analysis (CDA), this article develops a comprehensive framework for analyzing AI feedback discourse through a comparative analysis of two Large Language Models’ (LLMs’) responses to 63 undergraduate EFL descriptive essays. Employing Fairclough’s three-dimensional CDA framework, the analysis reveals distinct patterns in how AI systems construct their evaluative stance, distribute agency, and enact pedagogical authority in feedback. An emergent six-part taxonomy of discourse moves is identified: diagnostic positioning, prescriptive directives, facilitative suggestions, affective engagement, metalinguistic explanation, and comparative benchmarking. Findings indicate that the two LLMs employ contrasting discursive strategies—akin to a mentor versus an examiner—with significant implications for student positioning, learning autonomy, and the nature of pedagogical relationships in digitally mediated contexts. The proposed framework extends CDA methodology to AI-generated educational discourse and offers educators practical tools for critically evaluating AI feedback systems. As educational institutions rapidly adopt AI assessment tools, this taxonomy enables informed decisions about which discursive practices align with desired pedagogical values. The study concludes by discussing implications for student agency, pedagogical authority, and AI literacy in teacher education, and by recommending the development of more pedagogically-aligned AI feedback systems.