Political Deepfakes Are as Credible as Other Fake Media and (Sometimes) Real Media

There is widespread concern that political “deepfakes”—fabricated videos synthesized by deep learning—pose an epistemic threat to democracy as a uniquely credible form of misinformation. To test this hypothesis, we created novel deepfakes in collaboration with industry partners and a professional actor. We then experimentally assess whether deepfakes are distinctly deceptive and find that deepfakes are approximately as credible as misinformation communicated through text or audio. However, in a follow-up discernment task, subjects often confuse authentic videos for deepfakes if the video depicts an elite in their political party in a scandal. Moreover, informational interventions and accuracy primes only sometimes (and somewhat) attenuate deepfakes’ effects. In sum, our results show that while deepfakes may not be uniquely deceptive, they may still erode trust in media and increase partisan polarization.