The paper touches upon a very up-to-date problem of fake news and reveals a new approach to define them. We analyse three types of coherence in mass media discourse: semantic, pragmatic and syntactic. The purpose of this research is to compare fake and actual news as a part of mass media discourse from the point of view of their coherence. Three types of coherence - semantic, pragmatic, and syntactic - are investigated together in a combined and complex method. According to this method, the term “coherence” is interpreted generally, as a whole, as the ability of media texts to detect semantic, thematic, connections, author-audience references and syntactic hypertextual and multimodal references to other sources within mass media discourse. To visualise the results, we compare three dimensions of coherence in the fake news story “45,000 Dead from COVID-19 Vaccines within 3 Days”, Sparks Lawsuit Against Federal Government” published by VAERS and in the parallel similar story of a reliable source “Coronavirus: Plans to End Free Rapid Tests in England ‘Reckless'” from The Guardian. As a result, the hypothesis is confirmed that the more coherent an article is the less fake it is. This research allows to improve the process of fake news detection and as a prospective; establishes a background for creating algorithms and automatic fake news detection.