State media articles from 136 countries reveal shifting global narratives
The World as Viewed by 136 Countries: An Open, Curated, Annotated, and Multilingual Dataset of 118 Million State Media Articles
Computation and LanguageComputers and Society
Summary
State-run news outlets often share their governments' views around the world, and this content is widely used to train large language models (LLMs). The authors created a huge collection of 118 million state media articles from 136 countries and 114 languages to study how much these sources shape online information. They found that state-controlled media coverage has grown a lot, especially in English, and that these media can influence the answers LLMs give by promoting their own country's perspective. Their dataset helps better understand how state media presents different viewpoints globally.
What this means in practice
- •For language model developers: Use the WorldStateMedia dataset to audit and balance training data for reducing state media bias in large language models.
- •For news monitoring teams: Track and compare how different countries’ state media report on international conflicts to assess information framing and propaganda trends.
Authors
Bryan Li, Samar Haider, Grace Kim, Fiona Luo, Jinqi Luo, Yuan Yuan, Duncan Watts, René Vidal, Chris Callison-Burch
Abstract
State media outlets often project their governments' narratives to target both domestic and foreign audiences. Because this content is plentiful and free to access, it appears widely in the web corpora that large language models (LLMs) are trained on and that retrieval systems draw from. However, the prevalence of state media in these corpora and its potential influence on LLM outputs remain largely unexplored. In this paper, we introduce WorldStateMedia, an open dataset of 118 million state media articles from 136 countries in 114 languages. We extract these articles from existing pretraining corpora and annotate each outlet with level of state control from State Media Monitor. Using WorldStateMedia to audit these corpora, we find that, in English web crawls, the share of state media pages from editorially controlled (vs. independent) outlets rose from 39% in 2013 to 93% in 2025. In several low-resource languages, state media make up a substantial share of all web data, such as 17.8% for Uyghur and 7.8% for Belarusian. To characterize the differing narratives of news outlets, we compare Russian and Ukrainian state media in English and in their native languages, revealing asymmetric topical focus and opposing framings of contested issues such as Crimea. Furthermore, in a retrieval-augmented generation (RAG) experiment on territorial disputes, retrieving a country's state media shifts model answers toward that country's claims, most strongly for English-language (foreign-targeted) media. We release WorldStateMedia to support multilingual, cross-national insights on state media at https://huggingface.co/collections/manestay/worldstatemedia