Preventing the spread of COVID-19 requires persuading the vast majority of the public to significantly change their behavior in numerous, costly ways. Many efforts to encourage behavior change – public service announcements, social media posts, speeches, billboards – involve relatively short, persuasive messages. Here, we report results of five experimental tests (N = 5,351) of persuasive short messages conducted in the US from March – July 2020. In our first two studies participants rated the persuasiveness of 56 unique messages (31 drawn from the social science literature, 25 crowdsourced from online respondents). We then conducted three well-powered, pre-registered experiments testing whether the four top-rated messages would increase intentions to comply with public health guidelines. We compare messages to both a null control condition and an “active control” message that included a reminder of the virus and suggested behaviors with no persuasive frame. Five messages in the initial studies were rated as more persuasive than a control, and four messages in the later studies increased behavioral compliance intentions relative to a null control. However, none of these messages had consistent effects when compared to the active control message. We conclude that it may not be practically possible to identify short messages that reliably out-perform a simple reminder of the virus and recommended behaviors during the advanced stages of the pandemic. The most persuasive message studied was one emphasizing people’s civic responsibility to reciprocate healthcare workers’ sacrifices, which performed best in three of five studies.