Schwurbelarchiv dataset: 63M Telegram messages for conspiracy research

Telegram has become a hotbed for conspiracy discourse, but systematic research has been hampered by messy, hard-to-process data. Enter the Schwurbelarchiv – a curated dataset named after the German term 'schwurbeln' (speaking nonsense). Built by Mathias Angermaier, Elisabeth Hoeldrich, Jana Lasser, and Joao Pinheiro Neto, it transforms a raw data hoard from the Internet Archive into a structured resource. The dataset covers over 5,800 groups and channels, totaling 63 million messages, with a unique feature: transcriptions of 3 million audio and video files amounting to roughly 126,000 hours of content. The team parsed, cleaned, validated, and pseudonymised all user data, ensuring privacy while preserving the conversational threads that often mix text, voice messages, and videos.

This contribution fills a critical gap in computational social science. Unlike existing English-language datasets, the Schwurbelarchiv is predominantly German, validated by linguistic and temporal markers. Researchers can now analyze multimodal text–audio–video interactions to trace how conspiracy theories evolve, how misinformation spreads through voice messages, and how opinion adaptation occurs in closed Telegram communities. The dataset also supports studies on political extremism, network structures, and the effectiveness of counter-narratives. By making this data accessible and structured, the team lowers the barrier for rigorous, large-scale analysis of German-language conspiracy discourse.