Abstract: In the 2020 election cycle, political campaigns in the United States spent over $3 billion on online advertising. But how the internet platforms helped campaigns target and influence voters is largely shrouded from scrutiny. In this study we examine four platforms — Facebook, Google, Youtube, and TikTok — where political content was widely distributed to assess how these platforms amplified or moderated the political speech. Drawing on Rancière’s conception of politics, we assess how platforms shape the distribution of the “politically sensible” on their sites.
Platforms have provided limited mechanisms for outside researchers to examine their conduct. By crawling platforms between September and November 2020, we create a unique dataset of 800,000 image, text, and video political ads from Facebook, Google, and YouTube, along with 500,000 TikTok videos. We conduct a comparative analysis to understand what is defined as political by each platform, locating important divergences in who is allowed to place political ads, as well as what we can investigate as public with the available platforms’ transparency tools. Furthermore, we use appropriate machine learning techniques to understand how content was distributed to users or why it was moderated. We find patterns regarding the reach, cost, and demographic targeting of ads, which are not always explainable by the available data platforms provide. The same applies for moderation practices, locating different treatment of seemingly similar content. Overall, we locate platform-specific techniques that lead to opacity, such as reporting of uninformative content metrics, inadequate explanations, and non-disclosure of key advertising metadata. We then use these findings to develop policy recommendations that make online political advertising more transparent and accountable.