Deepfake Label Recall: Combating Disinformation with Labels is Especially Effective for Those Who Dislike the Speaker

Deepfake videos threaten to spread false information by depicting events that never happened, such as politicians making statements they never actually made. This research tested the effectiveness of accuracy labels to help people remember the difference between actual and deepfake videos of former US President Joe Biden. People accurately recalled 93.8% of deepfake videos and 84.2% of actual videos, suggesting that labeling videos can help users accurately recall information. Individuals who identify as Republican and had lower favorability ratings of Biden performed better in distinguishing between actual and deepfake videos. We use the elaboration likelihood model (ELM) as a theoretical framework to explain how distrusting Biden as a message source caused audiences to more critically evaluate messages. These findings support the practice of media companies labeling deepfake content to help users identify and recall information accurately.