Generative AI in higher education: a meta-analysis of intellectual and social–emotional outcomes

Background Generative AI has rapidly entered higher education, yet evidence on its effects across learning domains remains fragmented. This study synthesized experimental evidence on whether generative AI improves intellectual outcomes and social emotional outcomes among higher education students. Methods A systematic review and meta-analysis was conducted on 33 experimental and quasi-experimental studies involving 3,394 participants. The included studies were published between 2021 and 2024, and the literature search encompassed the period from database inception to January 26, 2026. Random-effects models in Stata 18 were used to estimate pooled effects for intellectual outcomes and social emotional outcomes, and publication bias, sensitivity analyses, and moderator analyses were also performed. Results Generative AI demonstrated a significant positive effect on intellectual outcomes (Hedges’ g = 1.096, 95% CI 0.087 to 2.104, p = 0.033) and a smaller but significant positive effect on social emotional outcomes (Hedges’ g = 0.301, 95% CI 0.048 to 0.553, p = 0.020). The difference between the two outcome domains was not statistically significant (coefficient = -0.739, p = 0.252). Publication bias was detected for intellectual outcomes but not for social emotional outcomes. Sensitivity analyses indicated that social emotional findings were relatively stable, whereas intellectual outcomes were more sensitive to individual studies. Moderator analyses showed that functional type, intervention duration, and knowledge domain did not significantly explain between-study heterogeneity at the omnibus level. Conclusion Generative AI appears to be a promising educational resource in higher education, particularly for intellectual outcomes, although the current evidence base remains heterogeneous and methodologically uneven. Systematic review registration PROSPERO CRD420261361622 .