From Tool to Tutor: Socratic AI Tutoring, Metacognitive Engagement, and Prior Knowledge as Determinants of Learning Gains in Gateway STEM Courses

Gateway STEM courses introductory courses such as Algebra carry disproportionately high failure and withdrawal rates, creating a critical bottleneck in undergraduate STEM pipelines. Intelligent tutoring systems (ITS) have demonstrated consistently positive learning effects over conventional instruction, yet the mechanisms underlying these gains, particularly for students entering with limited prior preparation, remain incompletely theorized. This study presents a quasi-experimental pretest-posttest framework comparing a Socratic AI Tutoring System (SATS) with traditional instructor-led instruction across two intact sections of a gateway STEM course (target N = 120-200). Drawing on Social Constructivism, Cognitive Load Theory, and Socratic Pedagogy, the study tests four hypotheses: that SATS produces higher learning gains (H1), that metacognitive engagement mediates this effect (H2), that prior knowledge moderates the treatment effect (H3), and that the mediation pathway is itself moderated by prior knowledge (H4). Data was analyzed by using ANCOVA, Hayes’ PROCESS Models 4, 1, and 7 respectively. Findings demonstrate that Socratic AI tutoring enhances academic performance especially among low-prior-knowledge learners, with metacognitive engagement serving as the primary mechanism of effect.