The Ideation Bottleneck: Decomposing the Quality Gap Between AI-Generated and Human Economics Research

Research reveals AI-generated economics papers significantly underperform human-authored publications, with idea quality representing the primary bottleneck (71% of the gap) rather than execution quality. Analysis of 953 papers shows human research achieves 47.1% exceptional probability versus 16.5% for AI, with only 0.8% of AI papers surpassing median human quality on both dimensions.

• →AI-generated economics research substantially underperforms human publications in head-to-head quality comparisons.
• →Idea quality accounts for 71% of the performance gap between AI and human research papers.
• →Human papers achieve 47.1% mean exceptional probability compared to just 16.5% for AI-generated papers.
• →Only 0.8% of AI papers surpass the median human paper on both idea and execution quality simultaneously.
• →74% of AI papers employ difference-in-differences methodology, suggesting limited methodological diversity.