Designing a human-in-the-loop AI iterative process for research writing: a framework for preserving productive struggle

Abstract

This Hypothesis and Theory paper addresses a design problem at the center of AI-assisted research writing: which cognitive demands should AI reduce, and which must remain the writer’s own work. Generative AI can absorb the effort of drafting, organizing, and synthesizing prose, but that same effort is where understanding takes shape. Recent empirical work documents reduced neural engagement, weaker recall of one’s own writing, and a diminished sense of authorship among writers who rely heavily on AI assistance. The paper grounds this problem in cognitive load theory, which distinguishes intrinsic load (the element interactivity a task inherently requires) from extraneous load (demands that do not contribute to learning), and which treats germane processes as the working memory resources a writer allocates to intrinsic load. It draws on productive struggle and generative processing to explain why germane processing cannot be offloaded: when a tool performs the work, the processing does not transfer to the writer; it does not occur. Two research questions are posed. First, which cognitive demands in research writing constitute extraneous load that AI could reduce, and which require germane processing that must remain the writer’s own. Second, how can human-in-the-loop AI collaboration be designed across the phases of research writing so that this distinction is operationalized through human judgment? The contribution is a nine-phase conceptual framework mapped to the sections of a research paper, from curiosity and inquiry through references and submission. Each phase specifies the dominant cognitive demand, the AI role (ranging from minimal involvement to full collaboration), and design guidelines for tools and interactions. The framework addresses research writing from secondary school (grades 9–12) through postdoctoral and professional research, across disciplines. It treats phase-aware AI use as responsive pedagogy applied to research writing: instruction and tool design that respond to the cognitive demands of the task rather than to the tool’s capabilities. The paper draws implications for the design of responsive digital education, including tool selection, assessment, equity, and support for writers with cognitive differences. The framework is intended to preserve the cognitive work that makes research meaningful.