Building rhetorical scaffolding: Closing the envisioning gulf in LLM-generated feedback for argumentative writing

Abstract

Large Language Models (LLMs) are increasingly used as writing assistants, yet their feedback often defaults to superficial, sentence-level corrections that fail to develop students' rhetorical skills. This paper addresses the cognitive challenges writers face when using LLMs, conceptualized as the "gulf of envisioning" - the gap between a writer's intentions and their ability to formulate effective prompts. It is argued that this gulf is particularly wide for complex, under-specified tasks like argumentative writing. Drawing on a qualitative synthesis of research in Human Computer Interaction (HCI) and Computer-Assisted Language Learning (CALL), "rhetorical scaffolding" is proposed as a framework of structured, dialogic prompting strategies for pedagogical intervention. This paper demonstrates how these strategies can bridge the envisioning gulf by transforming the AI-writer interaction from a corrective monologue into a metacognitive partnership.