The Lawfare Podcast · The Lawfare Institute

Lawfare Daily: What’s Influencing Politics Online? X’s Algorithm, Creators, and the New Persuasion Machine

March 31, 2026·48 min·4 clips

Helene Widmer's study shows X's algorithm pushes users' political opinions to the right when they switch feeds.

Renee DiResta opens by framing the episode around a central question: is it the algorithm or the influencer that shapes political persuasion online, or is it both? She introduces two guests whose recent research addresses different sides of this question. Professor Helene Widmer of the Paris School of Economics co-authored a paper published in Nature examining the political effects of X feed algorithm. Her study took advantage of X offering both chronological and algorithmic feed options, allowing researchers to randomize users into both conditions. The key finding was asymmetric: users who switched from the chronological feed to the algorithmic feed shifted their political opinions to the right on policy priorities, views on Trump criminal investigations, and opinions on the Ukraine war. However, users who switched from the algorithmic feed back to chronological showed no corresponding shift back. Widmer explains this asymmetry through follow behavior: the algorithm surfaces content from accounts users do not yet follow, users then follow those accounts, and when the algorithm is turned off, those follows persist. The study also found that the algorithm demotes news outlets, reducing their presence from about 25 percent of posts in the chronological feed to roughly 12 percent in the algorithmic feed, while upweighting political activists. Widmer notes this is consistent with a prior Meta study that found no depolarizing effect from turning off the algorithm, since that study only examined the turn-off direction. Nathaniel Lubin, founder of the Better Internet Initiative, presents findings from a field experiment conducted during the 2024 U.S. presidential election. Researchers had participants follow different types of content creators, some primarily political and some primarily cultural or apolitical, who had been trained to integrate fact-checked pro-social content into their feeds. The study found that on a per-video basis, the apolitical cultural creators had larger treatment effects on shifting political knowledge and opinions than the overtly political creators. Lubin attributes this partly to parasocial connection: audiences continued following the cultural creators at higher rates even after the study ended, suggesting a deeper trust bond. The more political creators effects appeared to be driven more by frequency and repetition, similar to advertising. The conversation then turns to broader implications. Widmer argues the asymmetry finding challenges the defense that algorithms are politically neutral because they simply show users what they already want. DiResta raises the issue that news organizations do not engage in back-and-forth replies on social media, which disadvantages them in engagement-weighted algorithmic feeds compared to individual influencers and activists. They discuss how platform demographics have fractured, with X in particular now skewing notably more conservative than other platforms. Lubin notes that the broader platform architecture, including account recommendations at signup and interface design, contributes to influence beyond just the feed algorithm itself. Both researchers agree that the stickiness of algorithmic effects, where learned behaviors and network connections persist even after the algorithmic input is removed, should make policymakers cautious about assuming that simply offering an opt-out from algorithmic feeds is sufficient to address the problem.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from The Lawfare Podcast