I employ Latent Dirichlet Allocation (LDA), a machine learning algorithm, to divide the FOMC meeting minutes from 2002 to 2013 into nine economic topics. The aim is to analyse whether the increased monetary policy transparency via the adoption of forward guidance in the post-meeting statements, and its greater use during and after the Global Financial Crisis, have transformed the contents of the minutes over time, and whether we can observe greater proportion of contents devoted to monetary policy (defined as references to current and future setting of the fed funds rate). The results show there is no significant variation in the proportion of topics over time but still we can observe an increase in the proportion of the monetary policy topic starting with the firmer use of forward guidance during and post-crisis. In addition, this paper highlights the particular suitability of LDA as a tool for Central Banks’ documents analysis.