Digital technologies offer new communicative affordances to fight corruption. Bottom-up efforts increasingly use algorithmic tools, i.e., bots, to automate corruption reporting on social media platforms. This study investigates how to design a bot to effectively and responsibly mobilize people for collective action against corruption. In a large (n=1,331) pre-registered choice-based conjoint survey, we test six message design features: type of injustice, degree of injustice, anger, political partisanship, gender, and efficacy cues. Our results show that calling out cases of severe corruption mobilized people against corruption effectively. We find no empirical support for in-group favoritism based on political affiliation and gender. Yet, some commonly used design features can have contrasting effects on different audiences. We call for more social science research accompanying the technical development of algorithmic tools to fight corruption.