Chat Generative Pre-Trained Transformer (ChatGPT) has generated excitement and concern in education. While cross-sectional studies have highlighted correlations between ChatGPT use and learning performance, they fall short of establishing causality. This review examines experimental studies on ChatGPT’s impact on student learning to address this gap. A comprehensive search across five databases identified 69 articles published between 2022 and 2024 for analysis. The findings reveal that ChatGPT interventions are predominantly implemented at the university level, cover various subject areas focusing on language education, are integrated into classroom environments as part of regular educational practices, and primarily involve direct student use of ChatGPT. Overall, ChatGPT improves academic performance, affective-motivational states, and higher-order thinking propensities; it reduces mental effort and has no significant effect on selfefficacy. However, methodological limitations, such as the lack of power analysis and concerns regarding post-intervention assessments, warrant cautious interpretation of results. This review presents four propositions from the findings: (1) distinguish between the quality of ChatGPT outputs and the positive effects of interventions on academic performance by shifting from welldefined problems in post-intervention assessments to more complex, project-based assessments that require skill demonstration, adopting proctored assessments, or incorporating metrics such as originality alongside quality; (2) evaluate long-term impacts to determine whether the positive effects on affective-motivational states are sustained or merely owing to novelty effect; (3) prioritise objective measures to complement subjective assessments of higher-order thinking; and (4) use power analysis to determine adequate sample sizes to avoid Type II errors and provide reliable effect size estimates. This review provides valuable insights for researchers, instructors, and policymakers evaluating the effectiveness of generative AI integration in educational practice.

An examination of this early synthesis raises concerns about the quality of the primary studies included in the analysis.