Lots of implications for future work wrt causality, effect magnitude of different actor dissemination, etc. LOTS more details in the paper, but please ask me stuff!

Overall, science communication is complex and involves multiple actors, even within the same platform. So there's potential for informational funnels to pass highly selective science through each pipeline stage (superspreaders -> media -> general public)

A figure showing a precedence network between contrarian superspreaders, conformist superspreaders, low-trust news outlets, and high-trust news outlets. Both types of superspreaders precede both types of news most of the time. News outlets typically only precede each other.

Peaks in superspreader activity tend to precede peaks in media activity; suggests that conversations take place on Twitter before they are picked up by the news, but this is by no means a strict causal pathway

Violin plot that shows the kinds of media outlet neighbors for conformist and contrarian superspreaders. Whereas conformists mostly align with high-trust, mainstream media outlets, contrarians mostly share similar papers with low-trust media outlets, including pseudoscientific and conspiratorial outlets.

There is some crossover in the activity of superspreaders and news media. Looking at superspreader-media outlet neighbors (where neighbors = share similar papers), conformist superspreaders align with mainstream media; contrarians align with pseudoscience/conspiratorial media

Two bar charts comparing who the coordinated network retweets, and who the rest of the network retweets. The coordinated network focuses its retweets around contrarian superspreaders (and some other contrarian users). The rest of the network predominantly retweets conformist (non-contrarian) users, with conformist superspreaders being the most-retweeted overall in this network.

Around ~20% of superspreaders are COVID contrarians. The coordinated network we detect retweets these contrarians almost exclusively (many are also MDs, scientists, etc). Further analyses show that the coordinated accounts are not bots, but difficult to ascribe intent beyond that

We also find "superspreaders": the most active paper disseminators whose posts receive the most engagement. These are mostly credentialed experts (scientists, MDs, etc.) who act as science communicators, with substantially larger followings than other accounts. BUT: ⬇️

A figure showing a coordinated retweet network. The network consists of six sub-clusters, with one being much larger than the rest.

Based on common retweets made close in time to each other, we find a coordinated network of accounts whose activity predominantly centers around vaccines/boosters, herd immunity, and excess mortality conversations. These accounts are overall younger than the rest.

We look at four main types of actors: coordinated accounts, superspreaders, bots, and news media. Our goal is to characterize their activity, but also to see how they interact with each other within Twitter and beyond.

Our new paper, "Information Pathways in Online Science Communication: The Role of Platform
Actors and News Media" (w/ Giuseppe Russo & @luceriluc.bsky.social) has just been accepted at ICWSM 2026!

Preprint: arxiv.org/pdf/2603.17249

Here are the main takeaways 🧵

University of Washington collaborates on study using AI to reduce political polarization on social media A University of Washington scientist explains how AI could be used to reduce political polarization on social media and perhaps give users greater control over their feeds.

@opb.org interviewed @msaveski.bsky.social about his research. "What we found when we surveyed [participants] at the end of the study was that, if polarizing content appeared lower in their feeds, they felt about two degrees warmer towards the other party," he said. www.opb.org/article/2026...

Lots more analyses and discussion in the paper. Any questions or comments, please get in touch!

Overall, this may create perverse incentives for gaining engagement on the platform.

Moreover, data access conversations often call out "permissible research" by platforms (h/t Ethan Zuckerman). Given owner's outsized influence, we may also need to start explicitly addressing "permissible reach"

A plot showing that accounts with increased visibility in the engagement feed are not legacy-verified, not left-leaning, get more attention from Elon Musk, are more agitating, less political, and post fewer links.

More visible accounts in engagement feed are more agitating, less political, get attention from Elon Musk, are not legacy-verified, and are not left-leaning. Other typical effects (e.g., lower exposure for links) also replicated. Right-leaning benefits disappear when controlling for these

Plots showing distributions of how centrality differs between engagement and chronological feeds. In the engagement feed, centrality is higher for right-leaning accounts and is much more skewed towards the platform's owner, Elon Musk, compared to the chronological feed.

Algorithm leads to more political cross-cutting exposure, but also much more centralization. Latter largely driven by outsized increased visibility of platform owner in engagement feed. Right-leaning accounts also enjoy increased visibility; explains increased x-cutting exposure (mostly one-sided)

Using existing data collected Feb 2023 (after Musk's Twitter acquisition, before X re-brand), we compare the accounts that users see between reverse-chronological and engagement (algorithmic) feeds. Time period aligns closely w/ Twitter's release of its recommendation system code

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

Recently accepted at ICWSM: arxiv.org/abs/2512.06129 (w/ @kayladuskin.bsky.social, @katestarbird.bsky.social, and @emmaspiro.bsky.social)

We look at how algorithmic feeds affect Twitter account visibility. More visible accounts:

-Are agitators
-Get attention from Elon Musk
-Are not legacy verified

Social media research tool can reduce polarization — it could also lead to more user control over algorithms A web-based method was shown to mitigate political polarization on X by nudging antidemocratic and extremely negative partisan posts lower in a user’s feed. The tool, which is independent of the...

🧵A new tool developed by researchers, including @cip.uw.edu faculty member @msaveski.bsky.social, shows it is possible to reduce partisan rancor in an X feed — without removing political posts and without the direct cooperation of the platform. (1/3) www.washington.edu/news/2025/12...

On YouTube Search API Use in Research | Proceedings of the 2025 ACM Internet Measurement Conference

My newest paper "On YouTube Search API Use in Research" is now published at dl.acm.org/doi/10.1145/...! I untangle some of the mechanisms with which the YouTube Search API samples results, and document limitations with some established research strategies in working with the API.

Come work with us!

⏰ Less than a week to apply to our PhD position on AI-driven Cybersecurity & Cybersafety!

Don't miss out—apply now and make the internet safer. 🚀👩‍💻

Job Link 👇

Huge news for science communication/metascience research!

CIP seeks applications for up to 2 postdoctoral fellows The University of Washington Center for an Informed Public (CIP) invites applications for up to two CIP Postdoctoral Fellows who will be hired as Postdoctoral Scholars.

We're currently seeking applications for up to two @cip.uw.edu postdoctoral fellows who will be hired as postdoctoral scholars.

Learn more: www.cip.uw.edu/2024/12/03/c...

For anyone who is interested but missed this, the recorded talk is now live at www.oii.ox.ac.uk/news-events/.... Please feel free to ping me if you have any thoughts or questions!

"Here's Your Evidence": False Consensus in Public Twitter Discussions of COVID-19 Science | Proceedings of the ACM on Human-Computer Interaction The COVID-19 pandemic brought about an extraordinary rate of scientific papers on the topic that were discussed among the general public, although often in biased or misinformed ways. In this paper, w...

For anyone interested in the full paper, it was published earlier this month at dl.acm.org/doi/abs/10.1... (open-access)

OII | From tribes to tropes: Network polarisation and false consensus around COVID-19 science In this talk, we cover recent work showing that COVID-19 science was vastly misrepresented on the Twitter platform, owing mostly to a large misinformation cluster that engaged in selective citation of...

This Friday (Nov 29th), I'll be speaking at the Oxford Internet Institute about our latest paper on how COVID-19 science was misrepresented on Twitter, leading to false consensus effects. The event is remote, and you can register at: www.oii.ox.ac.uk/news-events/...

UW campus, Drumheller Fountain and Rainier Vista.

[Please reshare]
I’m recruiting PhD students to work with me at UW!

I’m looking for students passionate about developing new *social media algorithms*, both broadly and within the scope of this NSF grant: tinyurl.com/395yfphd

More info: faculty.washington.edu/msaveski/

cc @uwischool.bsky.social

Misrepresenting Science Counter-consensus communities will frequently use selective and deceptive citation of the scientific literature as evidence for their views. How can we make ...

In case you missed our panel today on how counter-consensus communities (mis)use science, check out the recording: www.youtube.com/watch?v=fIWz...

Ft. @alextheefstra.bsky.social @beeeeeers.bsky.social and Rod Abhari

SciSky EpiSky MedSky

Excited to announce that my paper has just been accepted at The Web Conference (WWW) 2024! I analyze how users willing to engage with opposing views differ in terms of language they use and rewards they receive in their home communities. Preprint coming soon!

Great, thank you so much!