The Amazon is a complex object of policy that comprises environmental, economic, social, and sovereignty concerns. Despite this complexity, governments are often portrayed as having a single understanding of the region as a political problem. In this article, we investigate how the Amazon has been constructed as a problem in 6240 Brazilian presidential speeches since 1985 using supervised machine learning. Conceptually, we develop a framework that accounts for how important transnational actors, such as presidents, construct objects of policy as particular problems depending on where and when they participate in politics. Empirically, we find that presidents often construct the Amazon as an environmental problem when speaking far away from the region, whereas they usually construct it as problems of economic integration or social development when speaking within in the Amazon. Furthermore, presidents increasingly mix problem-constructions to represent the Amazon as a complex and multifaceted object of policy.