With advances in information and communication technology, a relatively new form of disinformation based on deepfake technology appeared. Deepfakes- hyper-realistic fake videos using face swaps-have become more and more widespread and easier to create, challenging the old notion of “seeing is believing”.So far, little empirical evidence exists analyzing the factors of people’s ability to counter the harm and spread of deepfakes. Additionally, what kind of video is more deceptive is also a question worth exploring. Therefore, this study empirically tests the factors related with people’s skepticism towards deepfakes and clarifies the relationship between perceived credibility and sharing intention. In practice, this understanding is essential for discussing and programming media literacy interventions as well as strategies of combating the negative effects of deepfakes.

We employed an offline survey of Chinese college students in December 2021 (N=191). To evaluate people’s ability to identify deepfakes and the influence of information cue (brief text introduction of video content), we presented respondents with six stories, three of them were deepfakes, and three of them had information cue. Participants were asked to watch all videos and complete measures. The survey includes three sections: (1) Demographic information, including age, gender, and education. (2) Independent variables, including media literacy, media trust, short video use and knowledge of deepfake. (3) Dependent variable, including perceived credibility and sharing intention of video.

Results of two-way analysis of variances showed that there were significant main effects of video type (deepfake vs. real video) [F= 219.75, p < 0.001, = 0.16] and information cue [F = 36.28, p < 0.001, = 0.03]. For one thing, the perceived credibility of deepfakes (M = 2.71, SD = 0.68) was significantly lower than that of real videos (M = 3.40, SD = 0.66). For another, the information-cue-present condition (M = 3.28, SD = 0.70) resulted in higher perceived credibility than the information-cue-absent condition (M = 2.84, SD = 0.75). In addition, the significant interaction effect indicated that information cue had a greater impact on deepfakes [F = 36.85, p < 0.001, = 0.03]. To examined the predictors of identifying deepfakes, hierarchical multiple regressions were conducted. The regression analysis suggested that those who had higher education (β = -0.156, p < 0.05) were likely to have lower perceived credibility which exhibited skepticism of deepfakes. And media trust (β = 0.18, p < 0.05) was positively associated with perceived credibility. Unexpectedly, media literacy was not a significant predictor. What’s more, perceived credibility positively affected sharing intention (β = 0.29, p < 0.001) which means that individuals are more likely to share videos when they consider the content to be credible.

The findings of this study suggest that people to some extents are skeptical of the credibility of deepfakes, which in turn negatively impact their sharing intention thus may limiting the potential harm of deepfakes at the very root. However, the inference that the more information cues provided by deepfakes, the more deceptive they might be, raises concern regarding the deceptive nature of well-made deepfakes. In addition, evidence has confirmed that people who trust the media are more likely to accept the claim of videos without making independent and rational analysis. According to a recent survey, Chinese people's trust in media is much higher than that of people in western countries, which deserves vigilance. Last but not least, no protective effect of media literacy was found in this study, possibly due to the fact that we put at the forefront the literacy of critically evaluating information. Theoretically, the critical understanding of media should be important in identifying disinformation, however, that may not be enough to discern real from fake stories in practice. As other studies have suggested, it is more crucial that individuals are equipped with the skills and competencies to sustain and update their access to rapidly changing information systems. Considering media literacy is a multifaceted concept, future research could explore the impact of different aspects of media literacy.