Similarity overestimation