Twitter allows political candidates to broadcast messages directly to the public, some of which spread virally and potentially reach new audiences and supporters. During the 2014 U.S. gubernatorial election, 74 candidates posted 20,580 tweets, of which, 10,946 were retweeted a total of 139,315 times. Using content analysis, automated classification and regression analysis, we show that actors with different levels of network influence tend to promote different types of election content, but that the convergence of their choices and actions lead to information flows that reach the largest audiences. We also show that actors with middle-level influence, in terms of the number of followers they have, tend to be the most influential in the diffusion process. Our work provides empirical support for the theoretical framework of negotiated diffusion, which suggests that information flows are the result of the convergence of top-down forces (structures and powerful gatekeepers) and bottom-up forces (collective sharing of actors with varying degrees of influence).