Twitter, as a popular social networking tool that allows its users to conveniently propagate information, has been widely used by politicians and political campaigners worldwide. In the past years, Twitter has come under scrutiny due to its lack of filtering mechanisms, which lead to the propagation of trolling, bullying, and other unsocial behaviors. Rumors can also be easily created on Twitter, e.g., by extreme political campaigners, and widely spread by readers who cannot judge their truthfulness. Current work on Twitter message assessment, however, focuses on credibility, which is subjective and can be affected by assessor’s bias. In this paper, we focus on the actual message truthfulness, and propose a rule-based method for detecting political rumors on Twitter based on identifying extreme users. We employ clustering methods to identify news tweets. In contrast with other methods that focus on the content of tweets, our unsupervised classification method employs five structural and timeline features for the detection of extreme users. We show with extensive experiments that certain rules in our rule set provide accurate rumor detection with precision and recall both above 80 %, while some other rules provide 100% precision, although with lower recalls.