In this dissertation, I analyze U.S.A. presidential debates as a communicative arena where candidates attempt to persuade the electorate to vote for them. The study is based on discourse analysis grounded in pragma-dialectics and conversational epistemics and deontics to I elucidate how talk on U.S.A. presidential debates is structured. I show that candidates advance a pragmatic argumentative structure tied to the act of making a campaign promise and based on four distinct epistemic resources. Drawing on the first debates from 2000, 2008 and 2016, I show how candidates shape their speech to manage the disagreement space between themselves and the audience. Candidates discursively construct the relevant facts and delineate territories of knowledge by invoking epistemic resources and through the careful use of the epistemic modality. They manage their political commitments through the use of the deontic modality and commitment markers. Using these two aspects, candidates show that there is a problem and that they are the only person going to implement a desired policy. This results in a complex argumentative structure based on three distinct premises and six critical questions which defines the discourse in this arena of communication. First, there is a necessity premise, used to stress that some course of action has to be followed, which is checked through two critical questions: a) does the course of action solve a problem? and b) does the course of action have positive effects? Second, there is a commitment premise, used to stress that the speaker will enact the proposed course of action. Two critical questions are used to check the reasonableness of this premise: c) does the candidate intend to enact this course of action? and d) is the candidate able to enact this course of action? Last, there is a comparison premise, used to stress that the opponent will not enact the desired course of action. The critical questions which are tied to this premise are: e) is the candidate the only one to propose the desired policy? and f) is the opponent not able to execute this policy?