The Impact of Generative Artificial Intelligence on Learning Outcomes in Higher Education: A Meta‐Analysis
TL;DR
GenAI significantly improves students' learning outcomes, with a particularly prominent positive effect on professional skills and emotional attitude, and potentially favourable instructional combinations for different learning objectives, such as enhancing academic or skill‐based outcomes.
Abstract
In recent years, Generative Artificial Intelligence (GenAI) has been increasingly adopted in higher education to support undergraduate learning. Despite its rapid integration into educational settings, the effects of GenAI on learning outcomes remain inconclusive owing to varying study designs and outcome measures. This study aims to synthesise empirical evidence on the effectiveness of GenAI in improving undergraduate students' learning outcomes across three dimensions: academic performance, professional skills and emotional attitude. A meta‐analysis was conducted on 35 experimental and quasi‐experimental studies published between 2022 and 2024, retrieved from Web of Science (SCI‐Expanded and SSCI), EBSCO (ERIC) and Scopus databases. The analysis examined the overall effects of GenAI and explored the moderating variables, including discipline type, teaching model, intervention duration, learning method and GenAI interaction method. We test the impact of GenAI on undergraduate learning outcomes across three dimensions: academic performance, professional skills and emotional attitude. The findings reveal that: (1) GenAI significantly improves students' learning outcomes, with a particularly prominent positive effect on professional skills ( g = 0.72) and moderate positive effects on academic performance ( g = 0.46) and emotional attitude ( g = 0.44). (2) The effects of GenAI are not universal, but are significantly moderated by key factors, including intervention duration, teaching model and GenAI interaction method. (3) The findings identify potentially favourable instructional combinations for different learning objectives, such as enhancing academic or skill‐based outcomes. However, these subgroup results should be interpreted cautiously in relation to sample size, intervention duration and outcome measurement. These findings provide methodological guidance for university faculty to effectively integrate GenAI into classroom teaching and offer evidentiary support for policies aimed at enhancing teachers' AI instructional competency.