High quality news sources consistently see lower engagement levels. (Credit: Tada Images on Shutterstock)

In A Nutshell

• Across seven platforms and ~11 million posts, links to lower-quality sites earned about 7% more engagement per post than links to higher-quality outlets.
• High-quality journalism still dominated by volume. It appeared more often and earned more total engagement overall, but it underperformed on a per-post basis.
• It’s not just algorithms. The same pattern showed up on Mastodon’s chronological feed and in exploratory Telegram data, which points to user preferences.
• Politics varies by platform. Conservative content performs best on conservative-leaning sites and liberal content on liberal-leaning ones; the quality gap persists in both.

You share a carefully reported story from The New York Times. Three likes. You post a link to some fringe site claiming the moon landing was faked. Twenty likes, a dozen shares, and your notifications won’t stop buzzing.

If this sounds familiar, a study analyzing over 10 million posts across seven social media platforms confirms you’re not imagining it. When researchers tracked what happened when the same person shared links to both high-quality and low-quality news sources, they found a consistent pattern: the junk outperformed the journalism. Every single time. On every platform.

The study, led by Mohsen Mosleh of the University of Oxford, examined all posts containing news links shared during January 2024 on X, BlueSky, TruthSocial, Gab, GETTR, Mastodon, and LinkedIn. The question was simple: does your low-quality content get more engagement than your high-quality content?

The answer: yes, consistently.

Posts linking to the lowest-quality sources received about 7% more engagement than posts linking to outlets like BBC, Reuters, and The Associated Press—the highest-rated sources in a database covering 11,520 news domains. That might sound modest until you consider the scale: across millions of posts and hundreds of thousands of users, the pattern never reversed.

The pattern held even on Mastodon, which displays posts chronologically without any engagement-based ranking. It showed up in exploratory analysis of Telegram, where content spreads through person-to-person sharing. Remove the algorithmic ranking systems that critics blame foramplifying misinformation, and people still engage more with lower-quality news.

The culprit appears to be us.

The Quality Paradox

High-quality journalism dominates social media by volume. Users share far more links to BBC, Reuters, and The Associated Press than to fringe sites peddling conspiracy theories. When researchers tallied total engagement across all posts, established news organizations came out on top.

But zoom in to what happens when you or I press “post,” and the picture inverts. Compare your own posts against each other—not against someone with more followers or better timing—and a pattern emerges. Your links to carefully reported journalism consistently underperform your links to lower-quality sources.

The study used ratings compiled from professional fact-checkers, journalists, and academics covering 11,520 news domains. Sites like BBC, The New York Times, and The Associated Press scored highest for quality. At the bottom sat outlets known for conspiracy theories and unverified claims, including Infowars and Natural News.

Posts linking to the lowest-quality sources received roughly 7% more engagement than posts linking to the highest-quality sources. This held true across all studied posts and users.

Findings remained consistent on social media platforms that don’t use algorithms like Mastodon and Telegram. (© Tada Images – stock.adobe.com)

Posts landing in the top 10% for engagement were more likely to contain links to lower-quality sources. On X (formerly Twitter), the only platform where impression data were available, low-quality posts also received more views, though the effect was about one-fifth the size of the engagement difference. This means people who saw questionable content were more likely to interact with it.

Why It’s Not Just the Algorithms

Many social media critics blame recommendation algorithms for amplifying misinformation. Adjust the ranking system, the thinking goes, and you fix the problem. The research suggests a more complicated picture.

Mastodon is a decentralized social network that displays posts in chronological order rather than using engagement-based ranking. No algorithm decides what you see first. No machine learning model predicts what will keep you scrolling. Posts appear in the order people publish them.

The pattern still appeared. Users’ low-quality posts received more engagement than their high-quality posts without any algorithm boosting them.

The pattern also showed up in exploratory analysis of Telegram, the messaging app, where content spreads through user-to-user sharing rather than algorithmic recommendation. Remove the algorithms entirely, and people still engage more with lower-quality news.

The study controlled for differences between users by comparing each person’s posts against their own. Someone with millions of followers will naturally get more engagement than someone with dozens. The analysis asked a different question: does your low-quality content outperform your high-quality content? Consistently, yes.

Something about low-quality news appeals to people. The researchers point to characteristics often associated with such content: emotional intensity, moral outrage, novelty. These traits may matter more for engagement than accuracy or careful reporting.

Why Major Outlets Struggle

Breaking down results by individual news outlets revealed that some of the most respected names in journalism struggle to generate social media engagement.

Posts linking to The New York Times, The Wall Street Journal, The Washington Post, USA Today, and Reuters showed comparatively low engagement rates relative to their quality and sharing frequency. These outlets produce most of the high-quality journalism dominating social media feeds by volume, but individual posts linking to them tend to underperform.

The researchers didn’t identify a single cause but offered several possibilities. Prestigious outlets often cover complex topics requiring careful reading, while lower-quality sites may package information in more immediately shareable formats. Many top-tier news organizations also use paywalls, though controlling for paywalls didn’t eliminate the quality-engagement gap.

Network effects may also play a role. People who frequently share low-quality news may have built audiences that expect and reward that type of content. When the same person shares something from a mainstream outlet, it may not resonate with their followers even if it reaches the same number of people.

This trend does not appear to be about individual articles being boring versus exciting. It’s about systematic differences inhow people respond to content from sources with different quality levels. The New York Times doesn’t struggle on social media because it publishes uninteresting stories. It struggles because social media engagement patterns favor different characteristics than journalistic quality.

Platform-Level Echo Chambers

While the relationship between news quality and engagement remained consistent across platforms, political orientation affected engagement in dramatically different ways depending on where people posted.

On right-leaning platforms like TruthSocial, Gab, and GETTR, conservative news sources received significantly more engagement than liberal ones. On left-leaning platforms like BlueSky and Mastodon, the pattern reversed. X and LinkedIn fell somewhere in the middle.

This contradicts the widely discussed theory of a “right-wing advantage” on social media, where conservative content supposedly outperforms liberal content regardless of platform. Instead, the research supports what scholars call an “echo platform” model, where entire social networks rather than individual users become ideologically isolated.

Conservative news posts receive more engagement on platforms where most content is conservative. Liberal news gets more engagement where most content is liberal. People engage more with news that matches their platform’s dominant political orientation.

This platform-level polarization may reflect broader changes in social media in recent years. As alternatives to established networks proliferated, users increasingly sorted themselves into ideologically homogeneous spaces. The study’s authors note that mass deactivations from X following the 2024 election, coinciding with growth in BlueSky memberships, suggest this trend may be accelerating.

Quality and Politics Connect

Platforms with more conservative user bases also featured lower-quality news overall. Comparing the seven platforms, researchers found a strong negative correlation between how conservative a platform’s content skewed and the average quality of news shared there.

Right-leaning platforms like TruthSocial and Gab had substantially more low-quality news sources in circulation than left-leaning platforms like BlueSky and Mastodon. This pattern mirrors individual-level research documented in prior studies showing that conservatives are more likely to share misinformation than liberals, though users on right-leaning platforms shared more high-quality conservative sources than users on other platforms did.

Despite this correlation at the platform level, the relationship between quality and engagement showed no statistically significant variation across platform ideology. Low-quality news outperformed high-quality news on left-leaning platforms just as it did on right-leaning ones. The preference for lower-quality content appears to transcend political orientation.

What This Means

These findings complicate efforts to combat misinformation. If the problem were primarily algorithmic, platforms could solve it by adjusting their ranking systems. But if user preferences drive engagement with low-quality news even without algorithms, solutions become more difficult.

Prior research has shown that prompting users to think about accuracy before sharing can reduce misinformation spread. Other studies demonstrated that adding context or warning labels can make people less likely to engage with false content. But those tactics don’t address why low-quality content appeals to people in the first place.

The study, published in Proceedings of the National Academy of Sciences, suggests that whatever makes junk news engaging—whether emotional intensity, moral outrage, novelty, or something else—consistently outweighs the appeal of carefully reported journalism. That pattern holds across different platforms, different ranking systems, and different political environments.

For news organizations, this presents a challenge. Quality journalism requires time, resources, and expertise. If that work consistently underperforms lower-quality alternatives in the metrics that matter for social media distribution, publishers face pressure to change their approach or risk being drowned out.

This research project examined only one month of data from early 2024, and social media platforms change quickly. The research also couldn’t include some of the largest platforms, including Facebook and Instagram, because their data policies don’t allow the kind of individual-level analysis the researchers needed. Short-form video platforms like TikTok and YouTube also weren’t included because they don’t center around sharing outside links.

Still, analyzing more than 10 million posts across seven different platforms provides a more complete picture of social media than most research, which typically focuses on a single network. That breadth revealed both surprising consistency in the quality-engagement relationship and striking variation in the political lean-engagement relationship across the social media ecosystem.

These findings suggest the pattern may reflect fundamental aspects of how humans interact with information online, rather than merely algorithmic artifacts. Whether that’s a problem platforms can solve, or one society has to address through media literacy and education, remains an open question.

---

Paper Summary

Methodology

Researchers collected every post containing a link to rated news domains published across seven social media platforms during January 2024: X (formerly Twitter), BlueSky, TruthSocial, Gab, GETTR, Mastodon, and LinkedIn. For GETTR, they included all available posts regardless of date due to low volume. For LinkedIn, they included all available posts from January 1, 2024, to the time of data collection. The dataset included over 10 million posts total.

News source quality came from a database of 11,520 domain ratings compiled by aggregating assessments from professional fact-checkers, journalists, and academics. Ratings ranged from 0 (lowest quality) to 1 (highest quality). Political lean came from GPT-4o ratings on a scale from -5 (extremely liberal) to +5 (extremely conservative), which the researchers validated against other established measures.

The analysis used linear regression models predicting the number of engagements (likes plus reshares) for each post based on the quality and political lean of the linked news source. All models included user fixed effects, meaning they compared the same person’s posts against each other rather than comparing different users. This approach controlled for all baseline differences between users, such as follower counts or posting frequency. Engagement counts were log-transformed due to heavy right skew.

For X, the researchers also analyzed impression data (views) using similar methods. An exploratory analysis of Telegram channels used a random sample of channels with relevant posts.

Results

At the platform level, there was a strong negative correlation between the average political lean of news shared on a platform and the average quality of that news. Platforms where more conservative content was shared (TruthSocial, Gab, GETTR) featured lower average news quality than platforms where more liberal content was shared (BlueSky, Mastodon, LinkedIn).

Looking at engagement patterns, political lean showed substantial heterogeneity across platforms. Conservative news received more engagement on conservative-leaning platforms, while liberal news received more engagement on liberal-leaning platforms. The correlation between a platform’s average political lean and the strength of the political lean-engagement relationship was 0.78.

In contrast, news quality showed a remarkably consistent pattern across all seven platforms. Within a given user’s posts, those linking to lower-quality sources received significantly more engagement than those linking to higher-quality sources. This relationship held even after controlling for political lean. The effect size corresponded to roughly 7% more engagement for posts linking to the lowest-quality sources compared to the highest-quality sources.

When analyzing which posts fell in the top 10% of engagement for each platform, lower-quality sources were overrepresented. However, looking at total volume, high-quality news was shared far more frequently and received far more total engagement.

On X, the only platform where impression data were available, lower-quality posts also received more views, though the magnitude was about one-fifth the size of the engagement difference. This suggests the quality-engagement relationship isn’t fully explained by algorithmic amplification of view counts.

The pattern persisted on Mastodon, which uses chronological rather than algorithmic ranking, and in exploratory analysis of Telegram. It also remained after controlling for whether sites used paywalls.

Disaggregating by individual domains revealed that lower engagement on high-quality posts was driven substantially by underperformance of major outlets including The New York Times, The Wall Street Journal, The Washington Post, USA Today, and Reuters.

Limitations

The study analyzed only one month of data (January 2024), and social media platforms evolve rapidly. Results may not generalize to other time periods or reflect current platform dynamics.

Data availability constraints prevented analysis of some major platforms, particularly Facebook and Instagram. The researchers focused on platforms allowing individual-level posting data collection and supporting link-based sharing, excluding photo and video-focused platforms.

The research was observational rather than experimental. While user fixed effects controlled for all baseline differences between users, the analysis couldn’t make causal claims about what drives engagement differences.

Quality ratings came from expert aggregations, which some critics might view as reflecting institutional biases. However, previous research has shown strong correlations between expert ratings and judgments from politically balanced groups of laypeople.

Political lean ratings came from a large language model (GPT-4o), though the researchers validated these against established measures.

The study analyzed posts and their engagement, not whether users actually clicked through to read articles or only reacted to headlines and preview text.

Because the analysis included user fixed effects and examined variation within users, it couldn’t make strong claims about between-user differences. Users who consistently share similar-quality content contributed less to the estimates than users who varied their sharing behavior.

Funding and Disclosures

The research received funding support from the Open Society Foundation. The authors declared no competing interests.

Publication Information

Mosleh, M., Allen, J., & Rand, D. G. (2025). Divergent patterns of engagement with partisan and low-quality news across seven social media platforms. Proceedings of the National Academy of Sciences, 122(44), e2425739122. doi:10.1073/pnas.2425739122 Published October 30, 2025.