We demonstrate that fabricated videos of public officials synthesized by deep learning(“deepfakes”) are credible to a large portion of the American public – up to 50% of a representative sample of 5,750 subjects – however no more than equivalent misinformation in extant modalities like text headlines or audio recordings. Moreover, there are no meaningful heterogeneities in these credibility perceptions nor greater affective responses relative to other mediums across subgroups. However, when asked to discern real videos from deepfakes, partisanship explains a large gap in viewers’ detection accuracy, but only for real videos, not deepfakes. Brief informational messages or accuracy primes only sometimes (and somewhat) attenuate deepfakes’ effects. Above all else, broader literacy in politics and digital technology increases discernment between deepfakes and authentic videos of political elites. Our findings come from two experiments testing exposure to a novel collection of deepfakes created in collaboration with tech industry partners.