Research about opinion dynamics is a thriving domain, that can help us understand pressing contemporary issues such as polarization. The theoretical behavior of the models is well understood, but confrontations with real data are scarce. We evaluate the empirical reliability of the well-known Voter Model. Its macroscopical properties were previously shown to match with vote-shares in U.S. elections. We take a fine-grained approach and investigate the reliability of the model on a microscopical level. We uncover a tight correspondence between the equilibrium state of the model and individual ground-truth opinions of users in a highly heterogeneous French Twitter dataset.