Global Claims: A Multilingual Dataset of Fact-Checked Claims with Veracity, Topic, and Salience Annotations

Tackling misinformation is a growing challenge, and collecting false claims is fundamental to support fact-checking activities, improve media literacy, understand how narratives spread and, increasingly, for training AI models to detect misinformation across platforms. Global Claims is a large-scale annotated dataset comprising 67,000 fact-checked claims, collected from more than 200 fact-checking websites in 50 languages. Each claim remains linked to their original web sources, and is classified along three dimensions: veracity (standardised into ''true'', ''false,'' ''satire,'' and ''other''), topic (mapped using a detailed codebook with 22 subtopics grouped into five broad categories: politics, health, society, science, and environment), and proxy salience (based on the claim's detection on Twitter/X). Instead of assigning a single topic per claim, the dataset provides probability distributions across topics, using a combination of expert, non-expert, and GPT model annotations. This multifaceted approach allows for a more nuanced and realistic analysis of misinformation, enabling researchers to examine claims that span multiple thematic areas, such as political actors discussed in a health context. The dataset has been carefully validated and curated to support diverse research needs, particularly in misinformation studies and fake news detection across multiple languages. It also offers a detailed comparison between human and automatic classifications, and is especially suited for machine learning applications, including automatic fake news identification and advanced topification tasks. By combining linguistic diversity, topical richness, and media salience indicators, this dataset offers a valuable resource for studying misinformation at scale.