2 May 2026 in Revista Interdisciplinaria de Educación, Salud, Actividad Física y Deporte
This study evaluated the impact of generative artificial intelligence tutoring on university students’ learning of differential calculus. A quasi-experimental pretest-posttest design with a retention measure was conducted at the Universidad Técnica Estatal de Quevedo (Ecuador) with 160 first-semester students nested in eight intact class sections (80 in the AI condition and 80 in the control condition), selected through non-probabilistic convenience sampling. The ten-week intervention combined graduated hints, independent solution attempts, and critical verification of AI responses. The primary outcome was total achievement (0-100), with conceptual understanding, procedural fluency, self-regulated learning, and cognitive load as secondary outcomes. Models adjusted for pretest and prior achievement estimated a 6.10-point advantage for the AI condition at posttest (95% CI [4.50, 7.70], p < .001, partial η² = .264) and 7.39 points at retention (95% CI [5.57, 9.21], p < .001). The standardized effect was larger for conceptual understanding (g = 0.86) than for procedural fluency (g = 0.43). Within the AI condition, fidelity to the verification protocol was positively associated with achievement (β = .18, p = .001), whereas uncritical copying was negatively associated (β = -.13, p = .026). Findings indicate that the benefit of AI tutoring depends on the quality of the pedagogical interaction rather than on mere access to the tool