In the internet age, the proliferation of fake news is a global concern (Lazer
et al., 2018). Despite the growing number of computational techniques for identifying digital news articles as untruthful based on their textual features, concerns about machine-based solutions have arisen. To complement the insufficiency of machine algorithms and gaining a clearer picture of the linguistic features in fake news, this study investigated the use of lexical bundles in illegitimate news by establishing a corpus of 100 false news texts obtained from the crackdown of stock promotion schemes in the United States in 2017. A total of 55 four-word lexical bundles were identified and analyzed through adapting and adopting the taxonomy devised by Biber et al. (1999). The findings reveal that (1) these identified lexical bundles are dominated by NP-based expressions and (2) referential bundles account for over half of the 55 instances, followed up an additional subcategory (i.e., subject-specific lexical bundles) of medical or financial vocabulary. It is hoped that these findings may shed some light on the recognition of misinformation and draw the attention of readers to some implicit features of untruthful journalistic language.