New Study Reveals Universal Preference for Lower-Quality News on Social Media

A recent comprehensive analysis of millions of social media interactions across a diverse array of seven platforms has shed new light on the intricate relationship between political content, information quality, and user engagement. The research indicates that the digital environment significantly shapes how users interact with political news, with engagement often mirroring the dominant political leanings of a particular platform. Crucially, a striking and consistent pattern emerged: users across all examined platforms exhibited a preference for engaging with news from lower-quality sources, even when shared by the same individuals who also disseminated high-quality content. This study, published in the esteemed Proceedings of the National Academy of Sciences, underscores the increasingly fragmented landscape of contemporary online information dissemination.

This investigation was driven by a critical need to update the scientific understanding of social media dynamics, which for many years relied heavily on data from a single platform, primarily Twitter (now X), due to its accessible application programming interface. This narrow focus led to a potential misrepresentation of broader internet behaviors, influencing assumptions about misinformation and political biases. The research team aimed to rectify this by expanding their analytical scope to encompass a wider variety of modern digital platforms.

The collaborative effort involved researchers Mohsen Mosleh from the University of Oxford and the Massachusetts Institute of Technology, Jennifer Allen from New York University, and David G. Rand, also from the Massachusetts Institute of Technology and Cornell University. Their objective was to ascertain whether phenomena like the 'right-wing advantage' in engagement or the rapid spread of false information were universal online truths or merely artifacts of specific platform designs. Furthermore, they explored whether the proliferation of alternative social media sites has fostered 'echo platforms' where users predominantly interact with those sharing similar political ideologies.

Data collection took place in January 2024, focusing on seven distinct platforms known for public news sharing: X, BlueSky, Mastodon, LinkedIn, TruthSocial, Gab, and GETTR. This selection offered a balanced representation, including mainstream platforms, professional networking sites, decentralized networks, and those catering to specific political demographics. The resulting dataset comprised nearly 11 million posts containing links to external news domains, providing a rich foundation for analyzing online sharing behaviors.

To rigorously assess the content, researchers employed a 'wisdom of crowds' methodology for evaluating news quality. They utilized reliability ratings for over 11,520 news domains, derived from aggregated assessments by professional fact-checkers, journalists, and academics. This enabled the assignment of a quality score to each publisher, serving as a reliable indicator of content accuracy. Additionally, a sophisticated large language model was employed to gauge the political leanings of news sources, rating domains on a spectrum from strongly liberal to strongly conservative. The validity of these AI-generated estimates was confirmed through cross-referencing with established political benchmarks, ensuring high confidence in categorizing content as left-leaning, right-leaning, or neutral.

The study primarily utilized linear regression analysis with user fixed effects, a statistical method designed to isolate individual user behavior by comparing posts from the same user. This approach effectively controlled for user popularity, ensuring that the study measured whether specific types of content received disproportionately higher engagement from an individual user, irrespective of their follower count.

The findings regarding political polarization challenged the notion of a universal advantage for conservative content. Instead, the data revealed that the political orientation of highly engaging content generally mirrored the political leanings of the platform's user base. Platforms predominantly attracting conservative users, such as TruthSocial, Gab, and GETTR, saw significantly higher engagement with right-leaning news. Conversely, platforms with more liberal or neutral user populations, including BlueSky, Mastodon, and LinkedIn, showed greater engagement with left-leaning news. This observation supports the 'echo platforms' hypothesis, suggesting a trend where users gravitate towards entire platforms that align with their existing views, rather than forming echo chambers within a single site. This implies that the 'right-wing advantage' previously noted on platforms like Twitter and Facebook might have been more a reflection of those specific user demographics than an inherent characteristic of social media itself.

While political engagement varied across platforms, the results concerning news quality remained remarkably consistent. Across all seven platforms, posts featuring links to lower-quality news domains consistently received more engagement than those linking to high-quality sources. This pattern held true irrespective of the platform's political orientation or its algorithmic feed design; it was observed even on Mastodon, which displays posts chronologically. The magnitude of this effect was substantial, with posts linking to the lowest-quality sites garnering approximately seven percent more engagement than those linking to high-quality sites, a finding robust even after controlling for the article's political slant. This suggests that the engaging nature of low-quality news extends beyond mere partisanship, likely driven by factors such as novelty, negative emotional content, and sensationalism.

The study also clarified the relationship between content volume and engagement. Although high-quality news sources are shared far more frequently than low-quality ones, dominating the ecosystem in terms of prevalence, they generate less per-post interaction. The inclusion of Mastodon as a control condition was particularly insightful; its lack of an engagement-based ranking algorithm indicates that algorithms are not the sole driver behind the misinformation advantage. The fact that low-quality news still outperformed high-quality news on a chronological feed points towards fundamental human psychological tendencies as a primary factor, suggesting users are naturally more drawn to content characteristic of lower-quality outlets.

Despite its comprehensive nature, the research has certain limitations. Data collection was confined to a single month, which might not capture seasonal fluctuations or behaviors during significant political events. Furthermore, restrictions prevented the inclusion of data from Meta platforms (Facebook, Instagram) and video platforms like TikTok, meaning the findings are most applicable to text-heavy, link-sharing environments. The study's observational design identifies associations but does not definitively establish causation beyond the statistical controls applied. Future research could expand the analysis to include more platforms, investigate the specific psychological mechanisms driving engagement with low-quality news, and explore how user migration across platforms influences information dissemination as the social media landscape continues to evolve.