🚨Out in Science!🚨 Conspiracy beliefs famously resist correction, ya? WRONG: We show brief convos w GPT4 reduce conspiracy beliefs by ~20%! -Lasts over 2mo -Works on entrenched beliefs -Tailored AI response rebuts specific evidence offered by believers https://t.co/3Rg79Cx5id 1/

🚨Out in PNAS🚨 Examining news on 7 platforms: 1)Right-leaning platforms=lower quality news 2)Echo-platforms: Right-leaning news gets more engagement on right-leaning platforms, vice-versa for left-leaning 3)Low-quality news gets more engagement EVERYWHERE https://t.co/kRzHK4YP9Q

NEW: @oiioxford researchers analysed 10 million posts across 7 social media platforms. They found: 🗳️ Lower-quality news is shared more on right-leaning platforms 📈 Posts aligning with a platform’s politics get more engagement ⚡ Across all platforms, low-quality posts

New video! Watch Dr Mohsen Mosleh, @oiioxford sharing his insights on his new co-authored research with Jennifer Allen @NYUStern and @David_Rand_ @Cornell, analysing news-sharing trends across seven major social media platforms.   Download the full study:

Huge credit to lead author Mohsen Mosleh and always-valuable collaborator Jennifer Allen. Full paper in PNAS: https://t.co/kRzHK4YP9Q And explore more of our group’s misinformation research here:

Conclusion: Some social-media dynamics are consistent across ecosystems (eg high-quality news getting less engagement), but others differ substantially (eg partisan engagement). ⚡️Studying behavior across multiple platforms is crucial for understanding news consumption online⚡️

Result 3b: Surprisingly, the pattern isn’t driven by low-quality outlets overperforming, but by the underperformance of top high-quality outlets (NYT, WSJ, WaPo, Reuters, USA Today). Additional analysis shows paywalls don’t explain the difference.

Result 3: Low-quality news gets more engagement on every platform, including left-leaning BlueSky and algorithm-free Mastodon. Models include user fixed effects, accounting for all possible user-level variation. Within a given user, their lower quality links get more engagement.

Result 2: Instead of a universal right-wing engagement advantage, we see *echo-platforms* where right-leaning domains earn more engagement on right-leaning platforms, left-leaning domains earn more on left-leaning ones.

Result 1b: Neutral news is shared more often than partisan news, and high-quality domains appear more frequently than low-quality ones. These plots are normalized within each platform; the dominance of high-quality news is even stronger without normalization across platforms.

Result 1a: Across platforms, average political lean strongly predicts average quality (r = –0.93). More right-leaning domains tend to be lower quality. Mirrors, atthe platform level, past findings about conservative users sharing lower-quality news.

Research usually looks at only 1 platform - here we look *across* 7 platforms (X, BlueSky, TruthSocial, Gab, GETTR, Mastodon, LinkedIn) and collect *every* post from Jan 2024 with a link to a news domain. Then we estimate domain quality (expert ratings) + political lean (GPT-4)

🚨Out in PNAS🚨 Examining news on 7 platforms: 1)Right-leaning platforms=lower quality news 2)Echo-platforms: Right-leaning news gets more engagement on right-leaning platforms, vice-versa for left-leaning 3)Low-quality news gets more engagement EVERYWHERE https://t.co/kRzHK4YP9Q

Back to X just to post new PNAS paper for @elonmusk: we find @CommunityNotes flags 2.3x more Republicans than Democrats for misleading posts! The issue is Reps sharing misinformation, not fact-checker bias... https://t.co/qnVNWCUHPt

Many thanks to lead author @captaineco_fr who does amazing work on Community Notes and other topics; and the always-wonderful coauthor Mohsen Mosleh For more of my group's work on misinformation, check out this doc:

CONCLUSION: *Clear partisan diff in misinfo sharing not due to political bias on the part of fact-checkers or academics *Undercuts logic offered by Musk+Zuckerberg for eliminating fact-checkers *Platforms should expect more Rep-sanctioning even when using Community Notes!

We also address two poss confounds: *Is it just that more Reps are using X than Dems, throwing off base rates? No, it's the opposite *It is just that more Dems use CN? No, b/c rated-helpful notes + bridging algorithm require people from both sides rating the notes as helpful

Across all English notes 1/23-6/24: 1.5x more notes proposed on tweets written by Reps than Dems. The partisan diff is MUCH bigger when restricting to "helpful" (ie ~unbiased) notes: 70% on Reps, 30% on Dems. The "vox populi" has concluded that Reps share more misinfo than Dems!

Many studies find Reps do share more misinfo. But those studies could be biased in evaluating what is misinfo. So we analyze Community Notes, where users flag posts & X decides what is misleading using a “bridging algorithm” requiring agreement from users who typically disagree

Accusations of bias against Reps (eg Trump+others suspensions) drove @elonmusk to buy Twitter & gut fact-checking in favor of @CommunityNotes. @finkd did same at @Meta. But greater sanctioning of Reps could just be the result of Reps sharing more misinfo

Back to X just to post new PNAS paper for @elonmusk: we find @CommunityNotes flags 2.3x more Republicans than Democrats for misleading posts! The issue is Reps sharing misinformation, not fact-checker bias... https://t.co/qnVNWCUHPt