This study explores the educational potential of a narrative-based artificial intelligence model, Alma de Tezontle, designed to enhance environmental awareness among university students. Using the GPT Personalization Methodology grounded in life stories and dialogic interaction, the model was shaped as a symbolic agent with a reflective voice. In this experience, the model adopts the perspective of an ancient stone that has witnessed the pollution of the Santiago River in Mexico, generating conversations that weave together ecological crisis with cultural, emotional, and territorial dimensions. The intervention followed a quasi-experimental design, involving an experimental group of sixty students who interacted with the GPT and a control group of thirty-two who took the same course through a traditional format. To assess changes, an adapted version of the Andalusian Ecobarometer was applied, and non-parametric statistical tests were conducted. The results reveal significant differences in three key dimensions: a reduction in beliefs that downplay the severity of environmental problems, an increase in pro-environmental behaviors, and heightened concern about climate change. Gender analysis suggests that women were more likely to translate their concern into concrete actions, while men showed more noticeable shifts in belief. These findings confirm that customized GPT models can serve as ethical dialogic agents in educational settings, capable of activating learning processes that engage not only cognition, but also empathy, awareness, and a deeper commitment to sustainability.
