Scholars of democratic politics differ in their conceptualization of the role of media in political campaigns and how it subsequently affects political behavior and electoral outcomes. There is therefore no consistent theoretical framework to guide defining, measuring, and analyzing election coverage, particularly in emerging democracies. We apply topic models and supervised learning methods to a news corpus of almost 100,000 news articles and approximately 137,000 tweets from South Africa’s 2014 election. We use a theoretically informed classification of election coverage to demonstrate how variation in “narrow” or “broad” conceptions of elections generate variation in codings of coverage of campaign events and activities reported in the media. Our analysis results in radically distinct representations of political actors and institutions in the electoral landscape: a narrow classification includes cues to race, party, and incumbent performance in coverage; a broad definition reflects public policy concerns and service delivery outcomes. Further, examining messaging directly from political actors on social media, we find social media messages by and large reflect the news reported in traditional media and further demonstrate politicians’ diverse appeals. We discuss the challenges and opportunities our research and method pose for how scholars conceive of media’s role in elections based on their theoretical priors about the form, content, and impact of campaigns in emerging democracies. Our results also provide methods, evidence, and lessons learned for replication to study (electoral and non-electoral) events in other developing country settings.