Deepfakes and the Crisis of Media Credibility

The deepfakes pose a threat to journalism and trust. This paper investigates the effect of unlabeled and disclosed synthetic media on credibility decisions and behaviors. The four-arm survey experiment (N=1,000) involved showing participants clips which were labeled as authentic, deepfake, deepfake with a label of an AI, and authentic with a deepfake warning. Some of the major findings are message and source credibility, perceived realism, trust in news, and sharing, verification, or reporting intentions. Results indicate that unlabeled deepfakes lowered credibility and slightly lowered news credibility but enhanced verification and reporting. Labeling that damages credibility minimized verification, whereas generic warnings on valid information created a liar-dividend effect, and this effect diminished trust. Vulnerability was improved by media literacy and numeracy. Policy implications imply more definite and narrow disclosures instead of vague warnings. Future studies are needed on multilingual deepfakes and longitudinal relationships of trust.