Swapped faces, swayed beliefs: How deepfakes affect trust in data videos with human narrators

With the rapid advancement of AI, deepfakes have gained widespread attention in visual content creation. This study focuses on the impact of deepfake manipulation of narrators’ faces in data videos and presents four progressive user studies. These studies examine two types of deception (deepfake narrators and deceptive visualizations) and two priming conditions (whether disclosing the possible existence of deception at the outset). Specifically, Study 1 swapped narrators’ faces and measured participants’ trust in narrators and visualizations. Study 2 mirrored Study 1, except that participants were informed about deepfakes from the start. Studies 3 and 4 added the deception of visualization design into consideration and respectively did not disclose, and disclosed the potential existence of deepfakes and deceptive visualizations in the beginning. In summary, we found that people can be confused by deepfakes, and their trust in narrators and visualizations shifted under different priming conditions.