The rapid adoption of social media by billions of people from all over the world has unleashed unprecedented opportunities for marketers and cognitive scientists to better understand why some message become popular while other die quickly. We designed a novel technique for automatically learning to differentiate popular tweets from unpopular ones and to predict how popular a given tweet will become in a given target audience. To demonstrate the effectiveness of our approach, we applied it to real world data collected from six social media messaging campaigns run by a variety of marketing as well as non-profit organizations including Proctor and Gamble’s Always Campaign. The studies showed that our approach can be highly effective (achieving accuracy scores from 92% to 99%) for automatically learning what makes a message popular in any given group as well as for automatically predicting how popular a message will be in a given target audience.