Recently, it came to light that some national governments have been employing paid commentators (“trolls”) to interfere with online political conversations to express pro-government views and to challenge the narrative of political opposition. In this study, I explore the behavior and impact of 700 trolls allegedly employed in late 2014 and early 2015 to leave pro-government comments on the popular social media website LiveJournal.com. Using probabilistic topic modeling, I develop a method to estimate the causal effect of trolls intervention in online discussions under a set of assumptions. I find that trolls are more successful in diverting the discussions from politically charged topics than in promoting a pro-government agenda. I also find that trolls are successful in diverting discussions from purely political topics, but have no effect if users discuss problems of the national economy. Those who discuss poor economic growth, unemployment, and/or price inflation seem not to be responsive to troll interventions.