Abstract: On social media platforms some messages spread much further than others, even if they concern the same topic. This fact is not reflected in models of social learning (or opinion formation) in networks. Our model fills this gap and shows that asymmetries in signal sharing determine learning of a society on a given issue. More “shareable” information fully dominates in the long run. This yields a substantial probability of misinformation, in contrast to the special case of symmetry covered by the literature. Asymptotic learning would require a perfect balance between two types of asymmetry: the product of decay factor and largest eigenvalue in the respective signal sharing networks must coincide. Approaching this balance reduces speed of convergence and enables social learning in the shorter term. Our analysis suggests that asymmetries in signal sharing should be mitigated, e.g. by weakening echo chambers or by fostering the shareability of cumbersome, boring messages.