Many central banks make policy decision collectively in the form of a committee. With the increased public communications of policy actions, wording over policy statements became an important tool for committee members to express their views. By analyzing sentiments expressed in transcripts of the Federal Open Market Committee (FOMC) meetings and FOMC statements using a large language model, we explore the interaction among FOMC members over statement language for various topics identified by word clustering. Specifically, we examine how statement language aggregates preferences of individual participants expressed during FOMC meetings. While the preference of the “median” voting member has received a lot of attention as representing the FOMC’s view, actual monetary policy deliberations are more nuanced and the statement language was modified occasionally to reduce the number of dissents. To examine the financial market effect of a surprise in statement language, we contrast sentiments expressed in the released statement with those from the private sector’s expectation of statement language using the proprietary data on the statement draft circulated among financial market participants before FOMC meetings. We find that the surprise component in the sentiment on the forward-looking component of the policy is associated with a monetary policy shock identified by high-frequency movements in the bond market. Our findings suggest that an unexpected shift in the relative influence of a voting member can be a novel source for a monetary policy shock if it leads to a surprise in statement language.