Study of 3.1 Million Pieces of Content Reveals Coordinated Narrative Operations Across TikTok and Telegram in Germany

Large-scale analysis of German TikTok videos and 700+ Russian- and German-language Telegram channels reveals geopolitically aligned narratives amplified up to 30x, cross-platform coordination with a measurable 12-48 hour detection window, and extremist content invisible to keyword-based monitoring.

MUNICH – Repsense, the predictive narrative intelligence platform deployed by NATO institutions and EU structures, has published The Unseen Battlespace: Information Threats in Germany’s Algorithmic Media – among the largest open-source studies of political short-form video content in a European democracy. The research screened more than 1.5 million TikTok posts and analysed in depth 262,265 videos from over 3,000 accounts, cross-referenced with more than 400 Russian-language Telegram channels, 300 German Telegram channels, and 1.2 million mentions in freely available online media. Total content processed: 3.1 million pieces.

The findings will be presented as a case study at a roundtable on the sidelines of the Munich Cyber Security Conference on February 12, co-hosted by Security Network Munich, United Europe, and Centenary Policy Institute.

The study addresses a gap in Europe’s defence posture. Germany’s 2026 defence budget of €108.2 billion and its commitment to reach 3.5% of GDP by 2029 represent an unprecedented investment in conventional and cyber security. But information operations – which NATO’s Strategic Communications Centre of Excellence has identified as a distinct domain of warfare – remain largely unmonitored in the algorithmic video environment where a growing share of public opinion is now formed. According to the Reuters Institute (2025), 65% of people globally consume news via social video. TikTok’s role as a vector for coordinated influence was demonstrated in Romania, where a foreign-backed campaign contributed to the annulment of the 2024 presidential election. In Germany, five state elections in 2026 make the information environment a direct security concern.

The context is stark: the Allensbach Institute’s Sicherheitsreport 2026, published on February 10 and based on 1,077 representative interviews, found that only 55% of Germans still feel safe, that trust in NATO’s ability to repel a Russian attack has fallen from 56% to 42% in a single year, and that, at the same time, fewer people believe Russia will attack further beyond Ukraine, with that conviction falling from 46% to 35%.

Key Findings

30x Amplification of Geopolitically Aligned Narratives

Geopolitically aligned narratives dominate German TikTok at rates vastly exceeding online media. The claim that the Ukraine war is a U.S. proxy war – the most prevalent geopolitically aligned narrative detected – appeared in 25.5% of flagged TikTok posts versus 1.9% in online media, a 13.5x amplification. The framing that “the West caused the Ukraine invasion” was amplified 8.4x. The broader narrative that “media and the system lie and censor the truth” – a framing consistent with foreign information strategies aimed at undermining trust in democratic institutions – reached 26.2% on TikTok versus 0.9% in online media: a 30-fold difference.

Cross-Platform Coordination: Russian-Language Telegram to German TikTok

Nearly identical narratives appear simultaneously across Russian-language Telegram, German-language Telegram, and TikTok. Of tracked narratives, 97% show moderate-to-strong similarity across platforms and 48% peak on the same day. Median lag: one day. Russian-language Telegram channels show a slight timing lead, suggesting narratives are seeded in Russian-language networks before surfacing on German-language platforms and being algorithmically amplified on TikTok. The predictive model identified a consistent 12-48 hour window between early coordination signals and peak amplification – the effective window for counter-messaging.

Coordinated German Telegram Networks

Analysis of 67,904 German-language Telegram messages revealed a tightly coordinated network of channels systematically amplifying identical content. Dominant themes include, among others, pro-U.S. partisan narratives (28%), promotion of populist-right parties (17%), attacks on centrist and centre-left parties (24%), and conflict-related geopolitical framing (7%). Telegram content is roughly twice as negative in sentiment as equivalent TikTok content, consistent with a role as an intensification layer where aligned narratives are sharpened before wider distribution.

Algorithmic Asymmetry

Parties positioned furthest from the political centre are overrepresented on TikTok relative to polling. The most overrepresented party exceeds its polling share by 5 percentage points (28.8% visibility vs. 23.8% polls). Centrist and Green parties are underrepresented. Accounts at the extremes of the political spectrum produce the highest volume; alt-left accounts achieve the highest per-video reach.

Extremist and Geopolitically Aligned Content Evading Moderation

The study detected Nazi-sympathising content using coded references – blurred images, initials (“A.H.”), dog-whistle hashtags – alongside Soviet flags, St. George ribbons, propaganda songs as background audio, and deepfakes including fabricated conversations between world leaders. None of this content is detectable by keyword-based monitoring. Its presence on a platform used by millions of young Europeans represents a security concern that current regulatory frameworks have not addressed.

“Europe is investing hundreds of billions in conventional and cyber defence, and rightly so. But any actor that can shape public opinion through coordinated messaging networks and algorithmic amplification operates at a fraction of that cost. Information operations are not a communications problem - they are a security threat, and they require a security response. If we do not monitor the information battlespace with the same rigour we apply to air defence or cyber, we leave a critical flank exposed.”

- Mykolas Katkus, Founder, Repsense

Methodology

The study combined three analytical layers. Narrative identification and clustering grouped over 3.1 million pieces of content into coherent narrative threads across platforms and languages, scoring each for growth velocity, coordination level, and structural behaviour. Deep video and forensic analysis used multimodal AI to examine the visual, audio, and textual content inside each video – detecting symbols, coded references, propaganda audio, and AI-generated material that keyword-based tools miss entirely. Predictive modelling tracked the behavioural signatures of coordinated amplification, projecting narrative trajectories forward and identifying the 12-48 hour window between early coordination signals and peak amplification. The traditional media baseline of 1.2 million mentions was drawn from freely available online media sources.

This approach builds on validated methods: in a previous deployment tracking information operations targeting a planned military base near the Suwałki gap, the same predictive framework detected coordinated activity three days before mainstream media pickup and up to two weeks before conventional monitoring tools flagged it.

“The value of this research is not just what it found but when it could have found it. Our platform tracks thousands of narrative clusters simultaneously, scoring each for coordination and growth. When those signals converge, we can project a narrative’s trajectory before it peaks. The 12-48 hour window between early coordination signals and peak amplification is not just a finding – it is a window of opportunity. For defence and security institutions, predictive narrative intelligence turns that window into actionable lead time.”

- Alfredas Chmieliauskas, Co-founder, Repsense

About Repsense

Repsense is a predictive narrative intelligence company. Its Havel platform monitors the full information environment - broadcast, online media, social media, messaging platforms, and short-form video – and applies AI-driven narrative clustering, coordination forensics, and predictive modelling to detect information operations early. The platform is deployed by NATO institutions, EU structures, national governments, and global enterprises across six countries. Repsense has previously tracked information operations across the Baltic states, Romania, Moldova, Armenia, Spain, Mexico, the Philippines, and other countries, achieving 90% accuracy in narrative trajectory prediction based on internal testing across more than 35 million pieces of content in seven countries. The company has raised more than €4 million following its seed round. Headquartered in Vilnius, Lithuania.

Notes to Editors

1. The study screened 1.5 million+ TikTok posts; 262,265 passed filters for deep multimodal analysis. Total content across all platforms: 3.1 million pieces. The traditional media dataset of 1.2 million mentions was compiled from freely available online media.

2. Reuters Institute Digital News Report 2025 data cited under the Institute’s open attribution policy. Newman, N. et al. (2025), Reuters Institute for the Study of Journalism, University of Oxford.

3. Repsense representatives are available for briefings at MCSC, February 12–14, 2026.

4. Media contact: Mykolas Katkus, CEO, Repsense, mykolas.k@repsense.io +37068568092