Adaptive Prompt Generation Using ChatGPT as Cognitive Scaffolding for Improving Spanish as a Foreign Language (ELE) Writing

Abstract

This study examines the use of generative artificial intelligence, specifically ChatGPT-3.5, as a pedagogical mediator to enhance writing in Spanish as a foreign language (ELE). A hierarchical five-level prompt system was implemented, manually integrated into the model's interface, simulating a conditional flow of progressive feedback. The architecture, grounded in cognitive scaffolding, was applied through controlled interactions with the Transformer model without additional training. The intervention, evaluated through a mixed quasi-experimental design (n=60), included automatic error analysis, codification of metacognitive strategies, and visualizations of textual behavior. The results show that the experimental group significantly reduced linguistic errors and activated more reflective processes than the control group. It is concluded that a technically structured system of hierarchical prompts can function as an adaptive computational tutoring system in AI-mediated educational environments.