Online discussion platforms are designed to facilitate discussions remotely between users, and have grown substantially in popularity over the last couple of decades. This growth was supercharged by the recent rise of remote/hybrid work and schooling. While there are many different types of these platforms, they share a common structure: the platforms all present a ranked list of answers and comments which participants can engage with. A common objective of the ranking algorithms that these platforms share is to maximize user engagement. In this research, we identify a novel discussion feature which we formalize as the level of discussion consensus, and we study the impact of consensus on engagement, and how consensus can be leveraged to improve the performance of ranking algorithms to increase engagement in a discussion. In the first chapter of this dissertation, The Empirical and Theoretical Foundations of the Consensus Effect, we analyze real-world discussion data and empirically verify discussion consensus as an engagement driver in online discussions. The presence of the consensus effect suggests that ranking algorithms should consider not only comments that would induce engagement in the present period, but also ones that would maximize engagement by managing the desired level of consensus. Based on this insight, we propose a new dynamic model for ranking optimization, and a class of intuitive algorithms that, among other factors, account for the level of consensus when prescribing rankings that maximize engagement using a limited lookahead. Building off the foundation of the first chapter, in the second chapter of this dissertation, The Consensus Effect in Real Time: A Field Experiment, we put the new consensus aware dynamic model to the test in a field experiment. Partnering with the online discussion board that supplied us with the data we used to glean our empirical results, we run a controlled field experiment where the ranking algorithm used for discussions is randomly selected between our consensus aware algorithm and the control, which is the algorithm currently used in practice (which does not account for the level of discussion consensus). This experiment demonstrates in real time how the engagement in discussions managed by the consensus aware algorithm outperforms the engagement in discussions managed by the current (control) algorithm. Finally, in the third chapter of this dissertation, Core-Periphery Evolution of a Group: A Driver of the Consensus Effect, we take a deeper look at the network structure of online discussions. We find that in some discussion groups, a clear core-periphery structure forms, where a select few members at the cores of discussions are responsible for much of the engagement, while the rest of the discussants remain at the periphery of the active discussion. Notably, we find that whether an individual belongs to the core or the periphery of a discussion, sensitivity to the consensus effect can vary quite significantly. Given these differences between the core and the periphery, we show how accounting for structural differences in sensitivity to consensus can be employed to further increase discussion engagement.