We developed an algorithm to predict which version of the "same" announcement an author should choose to get more attention. The main idea is to automatically learn what wording works better by examining a large number of pairs of tweets posted by the same author containing the same url.