When artificial intelligence learns Russian propaganda on your behalf
150 coordinated sites, 3.6 million articles in 2024, 33 percent of leading AI chatbots' responses contaminated by its narratives: the Pravda network does not target human readers, it targets the AI models that answer on your behalf.
« Pravda Network Infects AI Chatbots (LLM Grooming) »
More than 150 sites coordinated by the Pravda Network continuously publish content aligned with Kremlin positions on Ukraine, NATO, Western elections and health policy. The network, identified by Viginum in February 2024, published 3.6 million articles in 2024 according to NewsGuard. It targets generative AI models, which index and then repeat this content in 33 percent of their responses on sensitive topics according to a NewsGuard March 2025 audit of 10 chatbots and 15 narratives.
Executive summary
Identified by Viginum in April 2024 under the name Portal Kombat, the Pravda network now operates 190 country-specific sites and 140 sub-domains centralized on news-pravda.com, targeting 83 countries with 3.6 million pro-Russian articles per year. NewsGuard documented in 2025 that 10 major AI chatbots cite these articles as legitimate sources in 33 percent of tested responses. The network has also inserted more than 1900 links into 44 Wikipedia language versions.
What is observed
The network uses explicit domain naming conventions: pravda-fr.com, pravda-de.com, mali.news-pravda.com, serbia.news-pravda.com, burkina-faso.news-pravda.com. Over 224 sites identified by Viginum in April 2024, expansion documented by DFRLab to 190 sites and 140 sub-domains in 2025. Editorial volume estimated at 3.6 million articles per year. NewsGuard audit of 10 major AI chatbots (ChatGPT, Gemini, Claude, Copilot, Mistral, Perplexity and others) shows 33 percent of responses on sensitive topics include Pravda Network sources. CheckFirst identified more than 1900 Pravda links inserted in 44 Wikipedia language versions.
What this does not prove
The Pravda network has not formally demonstrated a chain of command with a specific Russian intelligence service. Domain centralization and naming patterns suggest coordination, not state attribution directly signed by internal documents. The cited AI chatbots may apply their own anti-disinformation filters and the 33 percent rate reflects a sample of sensitive questions, not an average response across all topics. Operator Yevgeny Shevchenko is identified as a Ukrainian national from Crimea, his direct cooperation with Moscow is inferred but not documented by public internal sources.
Confidence level
Five reference organizations converge on the figures (Viginum, DFRLab, NewsGuard, BISI, CheckFirst). The NewsGuard audit is reproducible and has been reconfirmed by multiple publications. Domain patterns are public and automatable.
Methodological limits
The Pravda Network ecosystem is volatile: sub-domains appear and disappear within weeks. The 190-site list is a snapshot from early 2025, not a fixed reality. The 33 percent NewsGuard figure comes from a targeted question set chosen by auditors, not a representative average usage.
How to cite this investigation
DisInfo Monitor (2026), "When artificial intelligence learns Russian propaganda on your behalf", independent publication, disinfo-monitor.com/en/narrative/pravda-portal-kombat-llm-grooming, first detected February 12, 2024, last updated May 18, 2026, accessed May 19, 2026.