Appel, R. E., Kim, Y. M., Pan, J., Xu, Y., Nimmo, B., Thomas, D. R., Allcott, H., Barberá, P., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M., González-Bailón, S., Guess, A. M., Iyengar, S., Lazer, D., Malhotra, N., Moehler, D., Nyhan, B., Settle, J., Thorson, E., Tromble, R., Rivera, C. V., Wilkins, A., Wojcieszak, M., Xiong, B., de Jonge, C. K., Franco, A., Mason, W., Stroud, N. J., & Tucker, J. A. (2026). How deceptive online networks reached millions in the US 2020 elections. Nature Human Behaviour, 1–15. https://doi.org/10.1038/s41562-026-02435-2
Summary
This paper measures the actual reach of 49 deceptive online networks that Meta identified as targeting US users on Facebook and Instagram around the 2020 elections. Combining platform-level exposure data with the large-scale FIES survey (~73,000 consenting users), the authors argue for a broader conceptual category of “deceptive online networks” that unifies both politically motivated coordinated inauthentic behavior (CIB) and financially motivated operations (FMOs) under the common feature of identity deception. The central empirical picture is one of concentration and indirect amplification: a handful of networks account for the overwhelming majority of reach, and most exposure flows not from network accounts directly but from reshares by ordinary, unaffiliated users. The paper also delivers a methodological caution—correlations between exposure and downstream outcomes largely vanish once pre-exposure user characteristics are controlled.
Key Contributions
- Introduces the umbrella concept of deceptive online networks, spanning both politically and financially motivated identity-deception operations.
- First large-scale measurement of actual exposure (rather than engagement proxies) to deceptive networks across an entire major platform’s user base during a US election.
- Documents the pivotal role of non-network reshares in amplifying deceptive content, shifting focus from network actors to user amplification dynamics.
- Shows that financially motivated networks producing political content can out-reach politically motivated ones, arguing they deserve equal scrutiny.
- Releases a de-identified dataset via the Social Media Archive (SOMAR/ICPSR).
- Offers a methodological warning about causal inference from observational exposure data.
Methods
Pre-registered observational study drawing on Meta’s platform-level aggregated data for 49 networks (13 CIB, 36 FMO) active 26 June 2020–15 February 2021, distinguishing direct exposure (via network accounts) from indirect exposure (via reshares), and characterizing cascade size, depth, breadth, and structural virality. Content was classified with Meta’s Topic and Civic classifiers. Individual-level FIES survey and behavioral data (~73,000 users, five waves) were linked, with entropy balancing used to reweight non-exposed participants for pre-exposure covariate balance across nine model specifications, plus pre-registered sensitivity analyses for unobserved confounding and post hoc analysis of feed interventions.
Findings
- Networks reached ~37M unique Facebook users (14.85%) and ~3M Instagram users (1.86%), generating 175M and 70M views respectively.
- Reach was extremely concentrated: three networks accounted for ~80% of unique Facebook viewers; one network (CIB10) accounted for ~95% of Instagram viewers.
- CIB networks originated mostly from Russia, the US, and Iran; FMO networks concentrated in the Balkans (16) and South Asia (11).
- Politics and social issues made up 65% of CIB and 32% of FMO direct Facebook content—substantial political content even from financially motivated actors.
- Exposure was highly skewed: 1% of Facebook users received 55% of exposures; 1% of Instagram users received 96%.
- For the top network (Rally Forge/CIB9), 13M of 13.4M viewers were reached indirectly via reshares versus only 1.3M directly.
- Only 5.67% of exposed Facebook users (0.34% on Instagram) reshared content, yet these users drove disproportionate amplification.
- Network content averaged only 0.3% of exposed users’ political content views in the 41 days before the election.
- After entropy balancing, associations between exposure and factual discernment, election legitimacy, and partisan news clicks largely disappeared and proved vulnerable to unobserved confounding.
- Feed interventions (no reshares, chronological) reduced exposure but also engagement; effects were underpowered.
Connections
This is a flagship FIES-collaboration study and connects closely to other work in that program on platform effects, notably Allcott2025-jb and Gonzalez-Bailon2024-rq. Its focus on measuring and conceptualizing coordinated deceptive networks situates it alongside broader coordinated-inauthentic-behavior detection and analysis work such as Giglietto2020-9d8acdd7, Giglietto2022-0e951ac5, Giglietto2023-fa71a001, Luceri2025-tr, and Minici2024-tf, while its emphasis on ordinary-user amplification of deceptive content complements studies of resharing and influence dynamics in Kim2026-wg.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify · Apple Podcasts