This article examines emotional intelligence (EI) as a pedagogical mediator of the effectiveness of AI-supported formative assessment practices at a university. In a sample of 64 full-time female students (one university), the following were compared over the course of a semester (T0–T2): the intensity/quality index of AI practices (personalization, early warnings, adaptive feedback, analytical dashboards), EI indicators (general latent factor and subscales: self-regulation, emotion regulation, empathy, social awareness) and educational outcomes (rubric performance, observed engagement, satisfaction/course climate). Psychometric testing included CFA and invariance tests; effects were estimated in SEM/PROCESS with bootstrapped intervals (5,000–10,000 runs) and cluster-robust errors by instructor/group. Partial mediation was observed: the indirect path through EI was statistically significant, while the positive direct effect of AI remained. The mediation rate was highest for engagement (~53%) and satisfaction (~50%), and moderate for academic performance (~36%). In the component analysis, self-regulation made the largest contribution to the indirect effect, followed by emotion regulation and empathy; the contribution of social awareness was smaller and more inconsistent. Practical conclusion: the effectiveness of AI solutions increases when integrated with EI and feedback literacy development programs (self-regulation and feedback micromodules for students; supportive digital communication for teachers), as well as when feedback attribution metrics are included in analytical dashboards. The limitations of a single institution and a gender-homogeneous sample provide direction for cross-institutional panels, longitudinal RCTs, and expansion of objective behavioral data sets, with control for covariates and robustness testing of results across model specifications.