This article investigates the scientization of central banks and analyzes the evolving knowledge production. Existing literature suggests that scientized knowledge production either becomes globally more similar over time or is driven by local political economy considerations. However, research on the structure of central bank knowledge production is lacking. To address this, I conduct semantic network analysis on 75 000 central bank working papers and articles from top economics journals. Findings show global organizational forms of knowledge production have diffused, but semantics remain localized. The semantic structure becomes increasingly clustered over time, with a notable division between the Federal Reserve System (FED) and non-FED clusters. Only Federal Reserve produces knowledge aligned with top academic journals. Cluster differences are not solely due to mandates or political contexts but depend on specific policy environments. This research illuminates the evolution of knowledge production within central banks and underscores the influence of organizational and policy contexts.