In recent years, there has been a rise in incorrect information, or misinformation, being shared on social media. Such misinformation tends to be more eye-catching and misleading. Previous studies on detecting misinformation online have focused mainly on linguistic characteristics; however, the role of linguistic characteristics in misinformation dissemination has not yet been thoroughly researched. In this study, we propose a misinformation dissemination model that includes the direct effects of four novel linguistic characteristics on dissemination and the moderating effect of information richness. The model is tested by using 9,631 examples of misinformation collected from Sina-Weibo, the leading social media platform in China. The results indicate that compared with correct information, misinformation containing persuasive and uncertainty words is more likely to be disseminated than misinformation containing emotional and comparative words. Furthermore, in the case of information richness, the results indicate that when the misinformation includes images, the effects of persuasive, negative emotion, comparative, and uncertainty words are strengthened, while the inclusion of videos weakens the effects of the linguistic characteristics. Finally, the results of robustness check by using the Fakenewsnet data set are consistent with our hypothesis, indicating that the four linguistic characteristics proposed by this study are also suitable for the dissemination of misinformation in English. The robustness check further demonstrates our method with well generalization.