{"id":4136,"date":"2025-09-30T11:36:00","date_gmt":"2025-09-30T08:36:00","guid":{"rendered":"https:\/\/factcheck.lt\/?p=4136"},"modified":"2025-10-01T02:13:26","modified_gmt":"2025-09-30T23:13:26","slug":"","status":"publish","type":"post","link":"https:\/\/factcheck.lt\/eng\/news\/zapad2025_youtube\/","title":{"rendered":"YouTube as a propaganda platform: analysis of Belarusian state media coverage of the Zapad 2025 military exercises","raw":"YouTube as a propaganda platform: analysis of Belarusian state media coverage of the Zapad 2025 military exercises"},"content":{"rendered":"

Key findings of the study \/ TL;DR<\/strong>
\nThis publication is a continuation of our previous analysis.A comparative study of coverage of the Zapad-2021 and Zapad-2025 exercises in Belarusian Telegram channels<\/a>This analysis is based on our regular monitoring of key state YouTube channels with over 1 million subscribers. Three of these five are YouTube versions of television channels. The other two are the main news agency and a major publishing house.
\nOur dataset for this year contains over 13,500 videos, 271 of which (2%) mentioned the joint Belarusian-Russian military exercises “Zapad-2025.” All transcribed versions of these videos were processed using LLM models for risk level, disinformation narratives, and coordinated campaigns.
\n50 claims require additional verification (classical fact-checking). Three coordinated campaigns were detected. The predominant narrative technique is military_rhetoric. About a third of the materials are classified as high-risk.<\/p>\n

Methodology and limitations<\/strong>
\nBasic Analysis: Statistical analysis of video metadata and temporal patterns. AI Analysis: Using gpt-4o-mini to detect disinformation techniques. Campaigns: Finding coordinated activity based on temporal and content patterns.
\nStrategic Sampling: Uniform time distribution for AI analysis.<\/p>\n

Limitations:
\nAnalysis is based on automatically extracted fragments of mentions. AI assessments require additional expert verification. Coordination patterns may have innocent explanations. Sample size for AI analysis is limited by API quotas. Focus on quantitative metrics; qualitative analysis requires expertise.<\/p>\n

A detailed analysis of disinformation techniques<\/strong><\/p>\n

\"\"