'Pravda' Network: Worldwide Expansion and LLM/Wikipedia Pollution
By Valentin Châtelet and Amaury Lesplingart
With DFRLab, showed the Russian 'Pravda' network expanded globally and polluted LLMs and Wikipedia; X's Community Notes failed to address its spread.
About this research
This online publication draws on API/data analysis and an OSINT investigation.
- Publication
- Archived online publication
- Research approach
- API/data analysis and OSINT investigation
- Geographic focus
- Global
- Reported study scale
- 44 language · 83 countries · 153 posts
- Research materials
- Public-source data is described in the publication
Publication details prepared by Social Media Transparency from the original source.
Executive summary
AI-generatedThis summary was generated by AI from the original report to make it easier to scan and cite. It is not a substitute for the source. Read the original above.
CheckFirst, with the Atlantic Council's DFRLab, investigated the Russian-linked 'Pravda' disinformation network using Wikipedia API analysis, X monitoring via the CrossOver tool, a review of three years of Community Notes annotations, and direct queries to large language models including Gemini, Copilot, Perplexity, and ChatGPT to test whether they flagged network-sourced claims.
The research found more than 1,900 Pravda-network hyperlinks embedded across 44 language versions of Wikipedia, including 922 in Russian and 580 in Ukrainian, and that major LLMs generally failed to flag unverified claims traceable to these sources. By late 2024/early 2025 the network had expanded to roughly 140 subdomains targeting over 83 countries. On X, Community Notes proved largely ineffective, annotating only 153 posts over 36 months, while CrossOver identified over 2,000 tweets promoting one network domain in just three months.
The authors conclude the network is actively polluting Wikipedia and LLM training/retrieval pipelines and call for urgent action from platforms and the Wikimedia Foundation to curb its spread and the ongoing risk of AI-amplified disinformation.
Think this summary is wrong? Contact us.