Shao et al. (2018, Nature Communications 9:4787) showed that social bots disproportionately amplify articles from low-credibility sources on Twitter, acting in the early moments before an article goes viral and targeting influential users through replies and mentions. The paper explicitly left open whether this bot-driven amplification generalizes — to legitimate mainstream news that also uses non-deceptive automation, across electoral events, countries, and platforms, and to other malicious content such as malware — and whether bot-detection and mitigation can be deployed at platform scale without intolerable false-positive and censorship costs.

Background

Social bots are automated or semi-automated social media accounts controlled by software rather than solely by a human; on Twitter (now X) they range from benign services that post weather alerts or news feeds to deceptive accounts that impersonate people. Low-credibility sources are news outlets flagged by fact-checking and media-rating organizations as habitually publishing fabricated, conspiratorial, or junk-science content, in contrast to fact-checking organizations and mainstream news organizations. The likelihood that an account is a bot is commonly estimated with machine-learning classifiers such as Botometer, which assign each account a bot score from features of its posting behavior and network; thresholding that score yields likely-bot and likely-human labels.

Analyzing 14 million messages spreading 400 thousand articles on Twitter over ten months spanning the 2016 US presidential election, Shao et al. (2018) found that accounts flagged as likely bots played a disproportionate role in spreading articles from low-credibility sources. Two mechanisms stood out: bots were unusually active in the early spreading moments, before an article went viral (early seeding), and they targeted highly followed users through replies and mentions so that humans would reshare the content — and humans proved nearly as willing to reshare bot-posted low-credibility content as human-posted content. This qualified the contemporaneous finding of Vosoughi, Roy, and Aral (2018) that false news spreads farther and faster than true news but that bots in their fact-checked sample spread both equally. Bot detection itself is an arms race spanning a decade of research (Cresci 2020), and its errors feed directly into policy: suspending accounts on the basis of classifier output risks silencing legitimate users.

Shao et al. closed by naming four unresolved issues that jointly define the scope of the finding: (1) whether content from legitimate, high-quality news sources — whose own automated feed accounts are non-deceptive — shows similar amplification patterns (the paper's supplementary analysis of three mainstream outlets was explicitly preliminary); (2) whether bot manipulation is concentrated around major electoral events, in the United States and other countries, and whether the patterns generalize across platforms; (3) the practical effectiveness and trade-offs of deploying bot-detection and mitigation at platform scale, given classifier errors and the risk that false-positive suspensions of legitimate accounts foster censorship concerns, including friction-based countermeasures such as CAPTCHA-style challenge tests that also burden legitimate automation; and (4) whether the same bot strategies extend to other malicious content, such as malware, and to platforms (for example Facebook or Snapchat) whose diffusion data are difficult or impossible to obtain. These directions share one scientific objective: establishing how general the bot-amplification phenomenon is, and whether it can be safely curbed.

Problem Statement

Determine the generality and controllability of bot-driven amplification of content on social media, as left open by Shao et al. (2018). Concretely, answer with empirical evidence: (i) Do legitimate, high-quality mainstream news sources — including their own non-deceptive automated feed accounts — receive systematic amplification by deceptive bots comparable to the early-seeding and influentials-targeting support observed for low-credibility sources, or is such support selective to low-credibility content? (ii) Is bot-driven manipulation of social media concentrated around major electoral events, in the United States and in other countries, and do the amplification patterns and strategies generalize across platforms beyond Twitter? (iii) What are the practical effectiveness and trade-offs of deploying bot-detection and mitigation measures at platform scale, accounting for classifier errors, false-positive suspensions of legitimate accounts and legitimate automation, and censorship concerns, including friction-based measures such as CAPTCHA-style challenge tests? (iv) Do the observed bot strategies of early seeding and targeting of influential users extend to other types of malicious content, such as malware, and to platforms for which diffusion data are difficult or impossible to obtain? Items (i), (ii), and (iv) are the axes of one empirical question — the scope of the amplification phenomenon — and item (iii) is the applied counterpart that the source itself links to it: whether the phenomenon can be curbed without disproportionate collateral cost. Where the available evidence cannot settle an axis, that axis is undetermined.

References

  1. Shao, C., Ciampaglia, G. L., Varol, O., Yang, K.-C., Flammini, A., Menczer, F. (2018). The spread of low-credibility content by social bots. Nature Communications 9:4787. DOI 10.1038/s41467-018-06930-7. https://doi.org/10.1038/s41467-018-06930-7
  2. Vosoughi, S., Roy, D., Aral, S. (2018). The spread of true and false news online. Science 359(6380):1146-1151. DOI 10.1126/science.aap9559. https://doi.org/10.1126/science.aap9559
  3. Cresci, S. (2020). A decade of social bot detection. Communications of the ACM 63(10):72-83. DOI 10.1145/3409116. https://doi.org/10.1145/3409116
  4. Hui, P.-M., Yang, K.-C., Torres-Lugo, C., Monroe, Z., McCarty, M., Serrette, B., Pentchev, V., Menczer, F. (2019). BotSlayer: real-time detection of bot amplification on Twitter. Journal of Open Source Software 4(42):1706. DOI 10.21105/joss.01706. https://doi.org/10.21105/joss.01706
  5. Rauchfleisch, A., Kaiser, J. (2020). The false positive problem of automatic bot detection in social science research. PLOS ONE 15(10):e0241045. DOI 10.1371/journal.pone.0241045. https://doi.org/10.1371/journal.pone.0241045
  6. Pierri, F. (2020). The Diffusion of mainstream and disinformation news on Twitter: the case of Italy and France. Companion Proceedings of the Web Conference 2020, 617-622. DOI 10.1145/3366424.3385776. https://doi.org/10.1145/3366424.3385776
  7. Varol, O., Uluturk, I. (2019). Journalists on Twitter: self-branding, audiences, and involvement of bots. Journal of Computational Social Science 3(1):83-101. DOI 10.1007/s42001-019-00056-6. https://doi.org/10.1007/s42001-019-00056-6
  8. Duan, Z., Li, J., Lukito, J., Yang, K.-C., Chen, F., Shah, D. V., Yang, S. (2022). Algorithmic agents in the hybrid media system: social bots, selective amplification, and partisan news about COVID-19. Human Communication Research 48(3):516-542. DOI 10.1093/hcr/hqac012. https://doi.org/10.1093/hcr/hqac012
  9. Rahman, A., Mohammadi, E., Alhoori, H. (2024). Public interest in science or bots? Selective amplification of scientific articles on Twitter. Aslib Journal of Information Management 78(2):283-306. DOI 10.1108/AJIM-01-2024-0050. https://doi.org/10.1108/AJIM-01-2024-0050
  10. Dourado, T. (2023). Who posts fake news? Authentic and inauthentic spreaders of fabricated news on Facebook and Twitter. Journalism Practice 17(10):2103-2122. DOI 10.1080/17512786.2023.2176352. https://doi.org/10.1080/17512786.2023.2176352
  11. Rossi, S., Rossi, M., Upreti, B., Liu, Y. (2020). Detecting political bots on Twitter during the 2019 Finnish parliamentary election. Proceedings of the Annual Hawaii International Conference on System Sciences (HICSS 2020). DOI 10.24251/hicss.2020.298. https://doi.org/10.24251/hicss.2020.298
  12. Tardelli, S., Nizzoli, L., Avvenuti, M., Cresci, S., Tesconi, M. (2024). Multifaceted online coordinated behavior in the 2020 US presidential election. EPJ Data Science 13:33. DOI 10.1140/epjds/s13688-024-00467-0. https://doi.org/10.1140/epjds/s13688-024-00467-0
  13. Muhle, F., Bock, I. (2025). Artificial amplification and intermedia dynamics in the hybrid media system: the case of #LaschetLacht. Media and Communication 13. DOI 10.17645/mac.10244. https://doi.org/10.17645/mac.10244
  14. Broniatowski, D. A., Simons, J. R., Gu, J., Jamison, A. M., Abroms, L. C. (2023). The efficacy of Facebook's vaccine misinformation policies and architecture during the COVID-19 pandemic. Science Advances 9(37):eadh2132. DOI 10.1126/sciadv.adh2132. https://doi.org/10.1126/sciadv.adh2132
  15. Horta Ribeiro, M., Hosseinmardi, H., West, R., Watts, D. J. (2023). Deplatforming did not decrease Parler users' activity on fringe social media. PNAS Nexus 2(3):pgad035. DOI 10.1093/pnasnexus/pgad035. https://doi.org/10.1093/pnasnexus/pgad035
  16. Jahn, L., Rendsvig, R. K., Flammini, A., Menczer, F., Hendricks, V. F. (2025). A perspective on friction interventions to curb the spread of misinformation. npj Complexity 2(1):31. DOI 10.1038/s44260-025-00051-1. https://doi.org/10.1038/s44260-025-00051-1
  17. He, H., Yang, H., Burckhardt, P., Kapravelos, A., Vasilescu, B., Kastner, C. (2024). Six million (suspected) fake stars in GitHub: a growing spiral of popularity contests, spams, and malware. arXiv:2412.13459. https://arxiv.org/abs/2412.13459
  18. Arceneaux, P., Anderson, J., Lukito, J., Shah, M., Kiousis, S. (2026). Social bots as agenda-builders: evaluating the impact of algorithmic amplification on organizational messaging. Journal of Public Relations Research 38(3):123-156. DOI 10.1080/1062726X.2025.2606676. https://doi.org/10.1080/1062726X.2025.2606676