Rabble-Rousers in the New King's Court: Algorithmic Effects on Account Visibility in Pre-X Twitter

TL;DR

The paper investigates how Twitter's pre-X recommendation algorithm influenced account visibility for political actors using real-user feeds and counterfactual comparisons. It demonstrates that right-leaning accounts appeared more visible in algorithmic feeds, but this effect largely stemmed from behaviors correlated with algorithmic rewards—namely posting more agitating content and closer proximity to the platform's owner, Elon Musk. It also uncovers a centralization pattern around Musk and finds that legacy-verified accounts were less visible in the algorithmic feed, while Twitter Blue verification showed no effect. The study uses bipartite network analysis with balanced matching to isolate account-level drivers, highlighting incentives that may shape online discourse and trust/safety dynamics, while noting limitations in causal inference and temporal scope.

Abstract

Algorithmic effects on social media platforms have come under recent scrutiny, with several works reporting that right-leaning accounts tend to receive more exposure. In this paper, we expand upon this body of work using data collected from user feeds after Twitter's change of ownership but before its re-branding to X. We replicate findings from prior work regarding the increased exposure of right-leaning accounts to wider audiences in algorithmically curated compared to reverse-chronological feeds, and, crucially, we further unpack this effect to understand what correlated (and did not correlate) with these differences. Our results reveal that right-leaning accounts benefited not necessarily due to their political affiliation, but possibly because they behaved in ways associated with algorithmic rewards; namely, posting more agitating content and receiving attention from the platform's owner, Elon Musk, who was the most central network account. We also demonstrate that legacy-verified accounts, like businesses and government officials, received less exposure in the algorithmic feed compared to non-verified or Twitter Blue-verified accounts. We discuss implications of these findings for the intersection between behavioral incentives for algorithmic reach and online trust and safety.

Figures (10)

• Figure 1: Balanced network visualizations.
• Figure 2: CCDFs for node importance measures across different feeds.
• Figure 3: CCDF of degree change from chronological to engagement feed.
• Figure 4: Violin plots by account political leaning and verification status.
• Figure 5: Gemini annotation performance by number of human annotators.