Mediation Without a Mind: A Sociocultural Reconsideration of Student Feedback Literacy for Generative AI Feedback in ELT Writing

Abstract

Generative artificial intelligence increasingly provides immediate feedback to English language learners, yet learners’ capacity to interpret, evaluate, and act on such feedback, known as feedback literacy, remains undertheorised in relation to this new source. This conceptual paper adopts a theory adaptation approach, using Carless and Boud’s four-dimensional model of student feedback literacy as the domain theory and Vygotsky’s sociocultural theory as the method theory. It examines how appreciating feedback, making judgements, managing affect, and taking action are reshaped when feedback is generated by general-purpose artificial intelligence rather than by teachers or peers. The paper argues that the model remains useful, but that each dimension faces additional strain because generative artificial intelligence functions as a mediating artefact and cannot be assumed to perform the pedagogical role of a more knowledgeable other. Although it can provide responsive assistance, it does not reliably identify a learner’s zone of proximal development, interpret developmental needs, or provide support that is carefully adjusted and gradually withdrawn. Learners must therefore undertake more of the evaluative, emotional, and regulatory work required to use feedback effectively. In English language writing contexts, target-language proficiency influences how strongly these limitations are experienced, creating a proficiency paradox in which learners who depend most on artificial intelligence feedback may have fewer linguistic resources to judge its accuracy, relevance, and appropriateness. The paper uses illustrative ELT writing scenarios and indicative proficiency benchmarks to anchor this argument, suggesting that the paradox may be most acute for learners at CEFR A2-B1, while recognising that such boundaries are gradual rather than fixed. It concludes that feedback literacy for generative artificial intelligence should be deliberately developed through teacher-supported mediation, guided verification, reflective revision, comparison of human and artificial intelligence feedback, and selective adaptation of suggestions.