Empowering GPT as a processual writer: Didactext-guided prompting improves knowledge access, iterative revision, and overall textual quality

Large language models are increasingly used as writing assistants, but their application often relies on holistic prompting that overlooks the recursive and cognitive dimensions of writing. This article investigates how guided prompting based on the Didactext model empowers GPT-4 to function as a processual writer, enhancing literacy processes in educational contexts. By decomposing writing into four recursive phases—knowledge access, planning, production, and revision—we demonstrate empirical improvements in reasoning depth, iterative refinement, and overall output quality. Building on GPT-4’s advanced capabilities in multimodal reasoning and steerability, the study adopts a hybrid experimental design with 150 mini-essay titles generated under guided and unguided conditions. Overall, guided prompts achieved higher textual quality, with raters observing clearer structure, deeper reasoning, and more precise use of evidence. Bias analyses also indicated a reduction in stereotypical content, though not its total elimination. These findings offer novel evidence of how AI can be used to simulate human cognitive writing processes and support literacy development, particularly in the revision phase. Implications include the design of AI-powered tutoring tools capable of encouraging gradual and proactive writing practices while reducing bias in diverse linguistic contexts.