Supersharers of fake news on Twitter

Sahar Baribi-Bartov; Briony Swire-Thompson; Nir Grinberg

doi:10.1126/science.adl4435

In brief

A small fraction of Twitter users—0.3% of 664,391 US voters tracked during the 2020 election—account for 80% of all fake news shared on the platform. These "supersharers" reach 5.2% of registered voters, are disproportionately older women and Republicans, and amplify misinformation through manual and persistent retweeting rather than automation. Their volume exposes a structural vulnerability in social media where a tiny population distorts political discourse at scale.

Contributions

Identification of a large, meaningful sample of supersharers using linked voter and Twitter data, enabling empirical study of a population previously documented only in aggregate
Measurement of supersharers' network influence beyond raw volume: 5.2% of voters follow a supersharer; followers are 2.5× more likely to encounter fake news; supersharers supply ~24% of fake news in their followers' feeds
First detailed sociodemographic profile of supersharers: overrepresented among women (59% vs. 50% baseline), older adults (median 58.2 vs. 41 years), and Republicans (64% vs. 29%); concentrated in Florida, Arizona, Texas
Refutation of automation hypothesis: fewer than 7.1% of supersharers are bots; posting patterns show no automation signals; the dominant mechanism is high-volume manual retweeting (74.7% of their tweets)

Method

Leveraged a panel of 664,391 registered US voters matched to their Twitter activity during the 2020 presidential election (August–November 2020). Defined supersharers as the 2,107 users accounting for 80% of tweets linking to fake news sources (using manually curated and NewsGuard-rated domain lists). Compared supersharers against two reference groups: top nonfake-news sharers (SS-NF, N=11,199) and a random sample (N=11,199). Used logistic regression to test sociodemographic differences, network-influence metrics, engagement rates, and bot-detection tools (Botometer) to assess automation. Examined posting patterns, app metadata, and retweeting rates to triangulate on the mechanism of high-volume sharing.

Results

Prevalence and dominance: On an average day, 7.0% of political news shared by the panel linked to fake sources, but this was concentrated: 0.3% of users generated 80% of fake-news tweets. Supersharers posted 15.9 political news links per day vs. 5.0 for SS-NF and 0.3 for the panel; posted 2.8 fake-news links per day vs. 0.1 and 0.01 respectively.

Network reach and influence: 5.2% of voters (N=641,144 with network data) directly follow at least one supersharer. Median supersharer ranked in the 86th percentile of network influence. Supersharers had significantly higher engagement (replies, retweets, quotes) per follower than all comparison groups (P<0.001). About 22% of the heaviest fake-news consumers follow a supersharer.

Exposure profiles: Average follower of a supersharer encounters fake news at 4.11% vs. 1.66% for average panelist (2.5× higher rate). Supersharers account for 24.4% of fake news available to their average follower and are the only source for 11.3% of their followers.

Sociodemographics: Logistic regression revealed supersharers significantly more likely to be:

Female: 59% vs. 49–52% in other groups
Older: median 58.2 years vs. 53.2 (average fake-news sharer) and 41.4 (panel)
Republican: 64% vs. 36–39% in other groups
Caucasian: higher percentage than panel and SS-NF (P<0.001)
Geographically concentrated: overrepresented in Florida, Arizona, Texas (P<0.05)
Slightly lower education (−0.3 years relative to panel/SS-NF) but higher relative income (+$2,500 USD vs. SS-NF)

Technical affordances: No evidence of widespread automation. Botometer analysis: ≤7.1% (95% CI: 2.0–12.2%) of supersharers classified as bots—no significant difference from SS-NF (P=0.35). Posting times, session lengths, inter-post intervals showed no automation signatures. Tweet metadata showed no elevated use of automation-supporting apps (P=0.75). Dominant mechanism: 74.7% of supersharer tweets were retweets vs. 59.9% (SS-NF) and 32.7% (panel)—high-volume manual and persistent retweeting.

Notes

Strengths: Exceptional dataset (664k voter-matched accounts); large, statistically meaningful sample of supersharers; multiple triangulated approaches to test automation hypothesis; replicated findings across robustness checks. The scale and precision of network-influence measurement (followers, engagement, feed composition) goes well beyond prior work's aggregate findings.

Limitations acknowledged by authors: Sample may differ from unregistered voters; findings specific to Twitter in 2020; unclear whether findings generalize to other platforms or post-Elon Musk era. Causal mechanisms and motivations remain open (activism vs. incidental vs. intentional misinformation).

Key surprise: Supersharers are not the stereotype of "young, alt-right, tech-savvy hackers"—they're older, predominantly female (among Republicans), and use low-tech retweeting rather than bots or sophisticated automation. This inverts assumptions about misinformation at scale and has direct implications for intervention design (e.g., retweet-limiting, account-monitoring, or community-level nudges may be more effective than bot-purging).

Significance: First empirical profile of supersharers at scale; establishes that information flooding by ordinary citizens is a structural feature of social media, not an edge case. The mechanism (manual retweeting) is deceptively simple and hard for platforms to police without infringing on ordinary users. Bridges the gap between rare prior observations of supersharers and testable, actionable mitigation strategies.