The literature on China’s social media foreign propaganda mostly focuses on text-format contents in English, which may miss the real target and the tool for analysis. In this article, we traced 1256 Twitter accounts echoing China government’s #USAVirus propaganda before and after Twitter removed state-linked operations on June 12, 2020. The 3567 tweets with #USAVirus we collected, albeit many written in English, 74% of them attached with a lengthy simplified Chinese text-image. Distribution of the post-creation time fits the working-hour in China. Overall, 475 (37.8%) accounts we traced were later suspended after Twitter’s disclosure. Our dataset enables us to analyze
why and why not
Twitter suspends certain accounts. We apply the decision tree, random forest, and logit regression to explain the suspensions. All models suggest that the inclusion of a text-image is the most important predictor. The importance outweighs the number of followers, engagement, and the text content of the tweet. The prevalence of simplified Chinese text-images in the #USAVirus trend and their impact on Twitter account suspensions both evidence the importance of text-image in the study of state-led propaganda. Our result suggests the necessity of extracting and analyzing the content in the attached text-image.