Ahead of the 2026 midterm elections, researchers at the Annenberg School for Communication recently published two studies that analyze the impact of social media during elections.
Meta organized a 17-person team — including Communication professors Deen Freelon and Sandra González-Bailón — to analyze the company’s data and determine how it’s platforms influenced the 2020 elections. This year, the team published the results of two studies that focused on the spread of untrustworthy information and deceptive online networks during that election cycle.
“What we found in 2020 is that reach was highly concentrated: a handful of networks accounted for most of their users exposed, and much of that exposure happened because ordinary users reshared that content,” González-Bailón wrote in an Oct. 7 announcement.
One of the studies, published in Science Advances, measured how much unreliable source content users of platforms such as Facebook and Instagram were exposed to and the effects of reducing such exposure. The other study, published in Nature Human Behavior, reported that 37 million Facebook and 3 million Instagram users viewed misleading political content about the elections in 2020.
Looking ahead to next month’s midterm elections, González-Bailón explained that the current question is whether artificial intelligence platforms are now detecting and demoting untrustworthy posts.
Freelon added that the influx of content and stimuli on social networks contributes to the reason why any single form of media has only “a remote chance” of substantially altering beliefs.
He also commented on a lack of distinction between the political and the apolitical in modern social media spaces.
“People often discuss topics that have political implications but never bring up government, elections, or the courts,” Freelon added.
According to González-Bailón, the team found that a small group of accounts and networks produced a large share of the problematic content they examined. She stated that while “platform design matters,” so do the online behaviors of ordinary users who engage with this content.
Looking ahead, the researchers cannot continue their work to the same extent because they no longer have access to the platform data Meta previously allowed them to obtain.
“The timing couldn’t be worse,” González-Bailón wrote in the announcement, “We’re being asked to judge these risks with little evidence and access to platform data.”
González-Bailón and Freelon will continue to research these topics through Penn’s Center for Information Networks and Democracy and the Annenberg School’s Politics, Identities, and Communication Lab.
