Work

The Deepfake Discount: Believing Erotic Images Are AI-Generated Reduces Affective Responses

Year: 2026
Type: preprint
Abstract: The proliferation of AI-generated content raises the question of whether beliefs about a stimulus’s origin alter emotional responses to it. Across two studies, we conceptually replicated and extended Marini et al. (2024) by presenting heterosexual participants (Study 1: N = 705; Study 2: N = 261) with erotic (Studies 1 and 2) and non-erotic (Study 1) images from validated databases (NAPS and NAPS ERO). Unbeknownst to participants, all stimuli were real photographs, each randomly labelled as “Photograph” or “AI-generated”. Participants rated sexual arousal, enticement, and valence for each image, and later judged their agreement with the labels. Images labelled as AI-generated received lower affective ratings, but this effect was small and driven primarily by male participants. Gender differences were robust yet context dependent: males yielded more robust differences, partially driven by a higher baseline arousal response. However, image of couples (Study 2), that females rated as more arousing, narrowed the gender gap found in Study 1. Critically, the effect of the labels interacted with participant’s own beliefs: photograph-labelled images were rated more arousing than AI-labelled only when participants’ subsequent belief ratings were congruent with the label, and the difference between labels reversed when opposite beliefs were reported, suggesting a complex relationship between emotion and reality beliefs. Individual-level differences in AI attitudes, pornography consumption, and sexual activity showed no reliable modulation of the labels effect. These findings suggest that the anti-AI bias is best understood as a belief-sensitive authenticity effect rather than a uniform response to AI labels, with implications for research using synthetic stimuli and for policies mandating disclosure of AI-generated content.
Language: English
Field: Psychology
Open Access status: green
OpenAlex ID: W7223502354
Cites:
Cited by:
Related to: 9