Recent studies have found promising evidence that lightweight, scalable tips promoting digital media literacy can improve the overall accuracy of social media users’ sharing intentions and improve their ability to determine the accuracy of true versus false headlines. However, existing research is designed to test entire bundles of such tips, which limits our practical knowledge about whether some kinds of tips are more effective than others and hinders our ability to theorize about mechanisms. We address this limitation by designing experiments in which we randomly assign participants to receive one or more of 10 possible tips (or none, in a pure control group) and then indicate the extent to which they either believe or would share a series of social media posts. We find that assignment to nearly any of the tips improves sharing, but only tips drawing attention to the posts’ source improved accuracy discernment (because source was highly diagnostic of accuracy in our stimulus set). Sharing intent appears to be more malleable than belief, consistent with the idea that fickle processes like attention play an important role in driving this behavior.