This research examines whether and how policymakers’ perceptions influence the negotiation processes and outcomes of IMF stabilization programs. It “opens the black box” of financial negotiations between the IMF, Argentina, and Brazil in the early postwar period, seeking to understand why two countries with broadly similar economic and political characteristics achieved distinct outcomes in financial negotiations—Argentina secured successive programs while Brazil did not. Unlike broad structural studies and traditional ideological narratives or historical accounts, this research takes a granular approach and employs innovative methodologies to engage with historical sources. It posits that “sentiment”—from both the IMF and program countries—plays a pivotal role in shaping interactions beyond economic and (geo)political observables. Since sentiment is not directly observable, this study is the first to operationalize a method using cutting-edge artificial intelligence algorithms for extracting it at scale from thousands of contemporary confidential policy sources produced by the IMF and the US, Argentine, and Brazilian governments. The findings reveal that the IMF learns and develops “soft information” over time through human interactions. This “soft information”—encompassing perceptions of trust, credibility, and capability—informs IMF lending decisions. extending beyond purely economic and (geo)political determinants.