fg-zettelkasten

Prigozhin’s propaganda team: The st Petersburg internet research agency (2013–2021)

3 min read

Prigozhin’s propaganda team: The st Petersburg internet research agency (2013–2021)

Summary

This paper offers the first systematic organizational study of the St Petersburg Internet Research Agency (IRA), Yevgeny Prigozhin’s notorious “troll factory,” covering its full lifespan from 2013 to its dissolution in 2023. Rather than analyzing IRA content output — the dominant approach in the field — Poliakoff and Toepfl use an open-source intelligence (OSINT) methodology built on 350 self-published CVs of former employees scraped from Russian job platforms HeadHunter and SuperJob. They argue that the IRA is best understood not as a state-funded intelligence appendage but as a Kremlin-aligned, privately operated “outsourced troll” enterprise structurally indistinguishable from a mid-sized PR or media company, deeply integrated into Russia’s legitimate media labor market.

Key Contributions

  • First systematic empirical account of the IRA’s workforce, hierarchy, and organizational evolution rather than its messaging.
  • A novel, replicable OSINT methodology — using employee CVs from public job platforms — for studying covert disinformation organizations.
  • Reliable demographic baseline that resolves contradictions in prior journalistic reporting on IRA staff.
  • A detailed five-level hierarchy and six-category functional taxonomy of IRA work.
  • Conceptual reframing of the IRA as “outsourced” rather than state-funded, refining typologies of state–private disinformation arrangements.
  • Empirical support for broadening “disinformation-for-hire” scholarship to encompass propaganda, PR, marketing, and promotional culture.

Methods

The authors manually collected 393 CVs (deduplicated to 350) by searching HeadHunter and SuperJob for known IRA-affiliated entity names (GlavSet, MediaSintez, Internet-issledovaniya, etc.). Structured fields were extracted via JavaScript and hand-coded for education, location, and job category (Krippendorff’s α = 0.947); free-text job descriptions were analyzed in MaxQDA using grounded-theory open and axial coding. The study was IRB-approved at Passau, with pseudonymization, restricted data sharing, and explicit assessment of self-selection, temporal, and social-desirability biases.

Findings

  • Workforce was 52% male / 48% female; median age 26; 98.7% Russian citizens; median tenure 2.3 years.
  • 78.3% held university degrees, predominantly from St Petersburg institutions; top fields were economics/management (22.2%) and journalism/PR/advertising (19.2%).
  • 27.4% had no prior work experience; only 14.3% reported advanced English, suggesting limited foreign-language operational capacity.
  • Recruitment spiked sharply in Q1 2014 (51 hires) around the Crimea annexation, and declined in summer 2016 before the US election — undermining a US-election-centric origin story.
  • Six functional task areas: mimicking citizens, mimicking journalism, advertising/marketing, monitoring/analysis, visual propaganda, and admin/tech support.
  • Five-level hierarchy from specialists up to CEOs, with both stable and project-based subunits.
  • Activity spanned Twitter, Facebook, Instagram, Telegram, YouTube, TikTok, Tumblr, VK, OK, LiveJournal, Yandex, and Pikabu; several VK groups and YouTube channels (Meet Bob, Chto esli, Kratkaya istoriya) remained active into 2025.
  • Use of mainstream ad tools (Google AdWords, Yandex.Direct, CPA/teaser networks) alongside in-house Russian-language synonymizers and Telegram/VK bots.
  • 32 CVs document seamless transitions to Prigozhin’s Patriot Media Group, often with identical job descriptions; shell companies kept staffing even after formal closure.

Connections

This paper provides organizational ground-truth that complements content- and behavior-focused IRA studies, particularly large-scale platform-data analyses such as DeVerna2025-dl and account-behavior work like Luceri2025-tr and Minici2024-tf. Its production-studies framing of trolling-as-labor connects to broader interrogations of professionalized influence operations and platform–state entanglements in Starbird2025-jj and Kuznetsova2025-nu, while the OSINT-on-personnel approach is methodologically distinct from the content-classification orientation that dominates the rest of this topic cluster.

Podcast

A research-radio episode discusses this paper: Listen

Graph View

Backlinks