Network ripple effects: How Twitter deplatforming flipped authority structure and discourse of the Arizona Election Review community

Summary

This paper examines what happened inside the Arizona Election Review (AER) Twitter community after Twitter’s July 27, 2021 suspension of eight accounts promoting state-level “audits” of the 2020 election. Combining network analysis with corpus linguistics on ~381,000 tweets bracketing the suspensions, Simeone and Corman argue that deplatforming produced ripple effects vastly exceeding the mechanical loss of banned accounts and their followers: the right-leaning cluster’s hub/authority structure collapsed, participation dropped sharply, and the surviving discourse pivoted from active ballot-surveillance recruiting to abstract defiance and meta-conspiracies about deplatforming itself. The authors frame this as evidence that targeted suspensions function as effective “network interventions,” and they question whether the visibility-reduction strategies favored under post-Musk X can replicate this disruption.

Key Contributions

• An in-situ case study of deplatforming effects on the community left behind, rather than the more common study of platform migration to alt-tech sites.
• A comparative analytic frame using before, after, and a counterfactual “before-minus-banned-accounts” network to separate direct removal from emergent ripple effects.
• Empirical demonstration that suspending a small number of central accounts can dismantle hub-and-authority structures sustaining election-denial mobilization.
• Connects deplatforming research to the “insurrectionist playbook” literature, situating moderation as an interruption point in election-conspiracy amplification.
• Raises a policy concern that visibility-reduction may underperform suspension when targets are already popular mobilizers.

Methods

Tweets matching (arizona AND audit) OR #ArizonaAudit OR #fraudit were collected via Netlytic for June 27–August 27, 2021 (264,412 before; 117,014 after). Directed, weighted edgelists were built from retweets, replies, quotes, and mentions, with conductance cutting (granularity 0.025) to identify left/right clusters. The authors compared degree distributions and Kleinberg hub/authority scores across the before, after, and counterfactual networks. An adapted CooRNet pipeline (10-second retweet threshold) tested for coordinated inauthentic behavior. Corpus linguistic analysis used log-likelihood keyness, KWIC, and collocation comparisons across the two corpora.

Findings

• The after network had ~46% fewer nodes and ~55% fewer links than the before network — far beyond what direct removal predicts.
• The right cluster collapsed from 6,380 to 1,395 nodes (7,676 → 1,437 links); the left cluster contracted only modestly (2,623 → 2,223 nodes).
• Hub concentration in the right cluster’s top 20 nodes collapsed, and authority concentration patterns reversed between camps — effectively flipping the network’s authority configuration.
• Only 10 accounts met the coordinated-retweeting threshold, indicating bot/CIB activity was not driving the network; disengagement was likely from real users.
• Pre-ban keywords (“ballots,” “forensic,” “retweet”) indexed surveillance and mobilization; post-ban keywords (“suspended,” “comply,” “distracted,” “Gab”) indexed defiance and platform migration.
• Surviving discourse depended more on a few high-profile verified accounts (Papadopoulos, Wendy Rogers, Arizona GOP) than on distributed grassroots recruiting.

Connections

This sits closely alongside other deplatforming and moderation-effects work in the register, particularly Copland2025-em and Donovan2025-ws on the politics and consequences of platform interventions, and Bak-Coleman2025-pm on post-moderation user behavior. It also speaks to the election-denial and insurrectionist-amplification literature represented by Starbird2025-jj and Moran2025-qn, and to broader work on coordinated and hyperpartisan mobilization such as Bastos2025-ya and Bastos2025-ol. Methodologically, its network-intervention framing complements platform-governance studies like Rieder2025-ju and Freelon2024-sc.