Most Americans are not highly interested in politics and, as such, many do not
specifically seek out political news. Yet they may encounter such information inciden-
tally on social media platforms, where posts about politics may still appear in users’
feeds as a result of algorithmic recommendations. We study the role of algorithmic
recommendations of political content at a time when prominent platforms have sought
to pull back from politics altogether. To do so, we take advantage of a policy shift: In
spring 2024, Meta announced that Instagram (along with Threads) would no longer
recommend posts about political or social topics to users from accounts they do not
already follow. In a large online field experiment, we randomly assign Instagram users
to opt back in to political recommendations and observe in follow-up surveys whether
our participants record changes in political knowledge, emotional experiences with
politics, and overall satisfaction with the platform. Concurrently, we document how
turning on recommendations affected content across several platform surfaces (Feed,
Reels, Explore) for users in general. Our study, spanning several months during an
unusually eventful period in American politics, sheds light on how algorithmic choices
shape citizens’ perceptions, knowledge, and attitudes about the election campaign.