An Average Joe, a Laptop, and a Dream: Assessing the Potency of Homemade Political Deepfakes

Abstract

Academic and media commentary suggests that deepfake videos are problematic because they are both more easily created and more potent than previous forms of misinformation. Surprisingly, there is little research that experimentally tests these claims. In this study, we tasked a first-year undergraduate student with quickly creating political deepfakes using easily available online tools. We experimentally compared the effectiveness of misinformation delivered through those deepfake videos against misinformation delivered through text and synthetic audio format (N = 443). Deepfakes were effective at planting false memories for fabricated political scandals and, in some cases, reduced reported voting intention by up to 20%. However, they were not consistently more effective than simple text. In a follow-up study (N = 300), we confirmed that we effectively debriefed participants and caused no lasting measurable changes to their beliefs or memories. We encourage further critical study of the novel properties of deepfake technology.