Subscriptions and external links help drive resentful users to alternative and extremist YouTube channels

Abstract

Do online platforms facilitate the consumption of potentially harmful content? Using paired behavioral and survey data provided by participants recruited from a representative sample in 2020 ( n = 1181), we show that exposure to alternative and extremist channel videos on YouTube is heavily concentrated among a small group of people with high prior levels of gender and racial resentment. These viewers often subscribe to these channels (prompting recommendations to their videos) and follow external links to them. In contrast, nonsubscribers rarely see or follow recommendations to videos from these channels. Our findings suggest that YouTube’s algorithms were not sending people down “rabbit holes” during our observation window in 2020, possibly due to changes that the company made to its recommender system in 2019. However, the platform continues to play a key role in facilitating exposure to content from alternative and extremist channels among dedicated audiences.

The Dynamics of (Not) Unfollowing Misinformation Spreaders

“…Some of these URLs had a disproportionate number of associated tweets. Consequently, in the third filter, we used a simple greedy algorithm to retrieve 5,600 misinformation spreaders 4 while minimizing the number of spreaders that came from any specific story (Appendix Algorithm 1). The motivation is that we did not want our entire sample and results to be dominated by a few misinformation stories.…”

“…Yet, the largest field experiment to date on this topic suggests algorithms do not promote misinformation on Facebook and Instagram [9]. In a different domain (extremist content on YouTube), [4] also questions the significance of algorithmic effects and points to demand effects as a more important mechanism.…”

“…PolitiFact categories: ['abortion', 'autism', 'coronavirus', 'drugs', 'disability', 'healthcare', 'health-check', 'public-health']3 PolitiFact truth values: ['pants on fire', 'false', 'mostly false']4 This number (5,600) was based on a power analysis conducted for a concurrent project that required detecting misinformation spreaders.…”

How Symbols Influence Social Media Discourse: An Embedding Regression Analysis of Trump’s Return to Twitter

“…A large body of scholarship has sought to explain how features of social media affect users. Some of this research investigates the activity or experience of using social media, such as the frequency of logging onto a platform (Jung and Lee 2023), the act of sharing messages or posts (Johnson et al 2020), and receiving algorithmic recommendations to content like videos and external news sites (Chen et al 2023; Levy 2021). Another portion of this literature focuses on social media content itself (Kubin and von Sikorski 2021).…”

“…Of course, there are other identities and groups we could have selected, but we limit ourselves to these few which have featured prominently in the discourse of Trump, the Trumpist movement, and the broader American conservative movement in recent years (Brooks and Harmon 2022; Chen et al 2023; Mutz 2018; Smirnova 2018). Our data are publicly available, and future research could examine discourse about other identities, groups, and topics.…”