Social networking sites (SNSs) have provided many benefits for healthy and functioning democracies, such as the increased accessibility of information and opportunities for the interpersonal exchange of ideas. However, the quality of information on SNSs has also raised concerns because of the prevalence of uncivil messages. Drawing on theories from risk communication, social psychology, and collective actions, this study proposed an integrative model that incorporated communication and social cognitive variables to predict public support for COVID-19 policies in Taiwan. Based on a nationally representative sample, this study found that uncivil messages online increased policy support. The results also showed that uncivil messages were associated with heightened risk perception but only for people who frequently engaged in discussions. Furthermore, whereas risk perception increased policy support, collective efficacy decreased public agreement with policies, a potential free-rider effect. These relationships were also moderated by trust. Implications of the findings will be discussed.