Gene conversion is the unidirectional transfer of genetic sequence from a “donor” region to an “acceptor”. In one of its modes, non-allelic gene conversion (NAGC, also known as interlocus gene conversion), the donor and the acceptor are homologous sequences on the same chromatid. Despite the implication of NAGC as the cause of various genetic diseases, and its role in the concerted evolution of many human gene families, the rates and contributing factors of NAGC are not well-characterized. Recent gene duplications are of focal interest in studying NAGC as NAGC is contingent on high sequence similarity between donor and acceptor. Notably, NAGC events are expected to distort the genealogy of a gene family at an affected region. Here, we develop a machine learning tools to survey duplicate gene families across primates in search of such local genealogy distortions, and identify converted regions in 44% of duplicate gene families surveyed. In addition, we estimate the parameters governing NAGC in humans. We estimate a tenfold higher rate of NAGC than point mutations in humans, with a median NAGC tract length of 525bp. Finally, we quantify the effects of genomic features which determine NAGC rates, including GC content, methylation levels and homology between donor and acceptor sequences. This work improves our understanding of the mechanisms behind NAGC and of the role NAGC plays in shaping sequence evolution in humans.