cib.pdf

cib.pdf.

View paper

Summary

This pre-registered observational study, conducted within the Meta–academic US 2020 Facebook and Instagram Election Study (FIES), measures the actual reach and effects of 49 “deceptive online networks” that Meta identified as targeting US users around the 2020 election. The authors argue for treating coordinated identity deception as a single conceptual category regardless of whether the motive is political (13 Coordinated Inauthentic Behavior networks) or financial (36 Financially-Motivated Operations). Using platform-level exposure data, individual survey/behavioral records, and network diffusion data, they find these networks reached roughly 37 million Facebook users and 3 million Instagram users, with reach dramatically concentrated in a handful of networks and heavily amplified by ordinary, unaffiliated users. Crucially, apparent associations between exposure and political outcomes largely vanish once pre-exposure characteristics are controlled, undermining naive causal claims.

Key Contributions

• Introduces and operationalizes “deceptive online networks” as a unifying category spanning politically- and financially-motivated coordinated deception in political discourse.
• First large-scale study to use platform-measured exposure (not engagement proxies) to deceptive networks across two platforms during a major election.
• Documents the underexamined prevalence of financially-motivated operations producing political content, challenging their dismissal as spam.
• Empirically foregrounds the role of ordinary non-network users in amplifying deceptive content, redirecting attention to user-level interventions.
• Releases curated dataset and code via SOMAR/ICPSR.
• Provides a methodological caution: exposure–outcome correlations are too sensitive to confounding to support causal interpretation.

Methods

Pre-registered (OSF, October 2020) analysis of 49 Meta-identified networks active during June 2020–February 2021. The authors combine aggregated platform exposure data (~250M FB, ~160M IG adult users, distinguishing direct vs. indirect/reshare exposure), linked survey-and-behavior data from ~73,000 consenting FIES participants across five waves, and network-level diffusion analysis (cascade size, depth, breadth, structural virality). Content was classified via Meta’s Topic and Civic classifiers. Outcome associations (truth discernment, election-legitimacy beliefs, partisan news clicks) were estimated using entropy balancing on pre-exposure covariates, with pre-registered sensitivity analyses for unobserved confounding and post-hoc exploration of FIES experimental feed conditions.

Findings

• Networks reached ≥36.8M unique US Facebook users (14.6%) and ~3M Instagram users (1.85%) over eight months.
• Origins were diverse: CIB from Russia (5), US (3), Iran (2), China, Romania; FMO concentrated in the Balkans (16) and South Asia (11).
• Politics/social issues were the top topic for both CIB (65% of direct FB content) and FMO (32%) networks—despite FMO’s financial motive.
• Reach was extremely skewed: three Facebook networks drove ~80% of Facebook reach; on Instagram, 1% of users accounted for 96% of content views.
• Indirect reshare exposure dwarfed direct exposure for top networks (e.g., CIB9: 1.3M direct vs. 13M indirect).
• Only 5.67% of FB viewers and 0.34% of IG viewers reshared network content, yet this small group drove amplification.
• Exposed users skewed older, more conservative, heavier platform users, and more exposed to untrustworthy sources—consistently across CIB/FMO and direct/indirect.
• Deceptive content was on average just 0.3% of participants’ pre-election political content views.
• Bivariate exposure–outcome associations attenuated to non-significance after adjustment and were not robust to plausible confounding.
• Only 12 networks ran any ads; paid promotion was 0.6% of posts.

Connections

This paper’s conceptual push to treat coordination-plus-deception as a unified phenomenon connects it to definitional and detection work on coordinated inauthentic behavior such as Giglietto2020-9d8acdd7, Giglietto2022-0e951ac5, and Nenno2025-xa. Its platform-scale exposure measurement and skepticism about effects sit alongside audit- and reach-oriented studies like Gonzalez-Bailon2024-rq and Budak2024-ef, while its finding that ordinary users amplify deceptive content relates to diffusion and untrustworthy-source exposure work such as DeVerna2025-dl and Pierri2025-hm.