Content-based detection of misinformation expands its scope across politicians and platforms

Research on misinformation crucially depends on the methods to detect it. Prior studies have primarily identified content from untrustworthy news sources via their URLs. Yet, since this approach is limited to certain posts, content types, and platforms, new methods beyond URL matching are required. This study applies retrieval-augmented classification (RAC) to detect misinformation in free text, enabling broader platform coverage and more fine-grained content analysis. We analyze a unique dataset of over half a million posts from all German members of parliament and their official party accounts across four platforms – Facebook, X, Instagram, and TikTok. Using RAC, we match these posts with more than 5000 fact-checking articles and 1500 community notes. Our method reveals an absolute amount of misinformation nearly ten times higher than the URL-matching method would capture, while the relative share among all posts with text remains low. However, patterns differ across parties and policy issues, as misinformation is unevenly distributed among political actors. For some parties and specific issues, misinformation shares exceed ten percent. Our findings expand the scope of misinformation research through a text-based matching method, offering a more nuanced distinction between high- and low-prevalence segments, and highlighting the role of populist and conservative parties in spreading misinformation.