Domain-specific emotional effects of deepfakes on news-sharing behaviour

The rise of generative AI has raised societal concerns about the potential of synthetic media or deepfakes for misinformation. While previous research has established that negative emotions drive text-based misinformation sharing, their influence on deepfake dissemination remains poorly understood. Through two online experiments (N = 487; N = 479), we examined how emotional valence in deepfake images affects sharing intentions across political and entertainment news domains. Using GPT-3 and Stable Diffusion, we manipulated image emotional valence while maintaining consistent news contexts. Results revealed that negative images increased sharing intention only for political content, not entertainment content. Moreover, replacing negative images with less negative alternatives reduced sharing intention for political news, even among those initially inclined to share. These findings suggest domain-specific emotional dynamics in deepfake sharing and suggest novel content moderation strategies based on affective computing.