In this paper we point to a surprising phenomenon in Online Social Networks (OSNs), which we call "The Million Tweets Fallacy". Our hypothesis is that the measurements of activity of a user in an OSN are not correlated with the measurements of feedback that the user receives on that activity. For example, the number of tweets is uncorrelated with the number of retweets or likes. In other words, a voluminously tweeting user is not necessarily fuelled by the attention he gets from his followers. An innovative aspect of this work is that we treat "activity" and "feedback" as multidimensional axes, and do not reduce the problem to a one-dimensional pairwise correlation problem. We apply our methodology to six OSNs. For Twitter, Instagram, LinkedIn and Steam we gathered the data ourselves, collecting features that cover both users' activity and feedback in the OSN. For YouTube and Flickr we used existing data from the literature. In all OSNs, with the only exception of Steam, we confirmed our hypothesis.