Veracity and register in fake news analysis

Abstract This article addresses two fundamental methodological challenges in linguistic fake news research: how to reliably classify news texts as real versus fake and how to control for register variation when building datasets for comparative analysis. We call for a rethinking of the veracity labels produced by fact-checking services. While fact-checkers remain a valuable resource for identifying fake news texts, their labels are more productively seen as a proxy for the communicative intent of the author, rather than as an absolute measure of veracity. This perspective emphasizes communicative intent as the only viable explanation for systematic linguistic differences between real and fake news and sidesteps the politically charged notion of truth. To address potential confounds caused by register and topic variation, we propose a multipronged comparative approach. This method analyzes fake news alongside real news on the same topic from a variety of news registers, allowing us to isolate linguistic differences driven by intent. We exemplify this approach by building a comparative corpus focused on climate change, vaccinations, and COVID-19, which we make available upon request for other researchers to use.