Anti-AI Bias Toward Couple Images and Couple Counseling: Findings from Two Experiments

Generative artificial intelligence (AI) systems can produce text, images, videos, and audio in response to prompts. They are increasingly applied across various domains, including intimacy and sexuality-ranging from AI-generated pornography to sexual counseling via AI chatbots. While AI-generated content holds significant potential, it is also met with skepticism. Anti-AI bias is defined as a systematic tendency to evaluate AI-produced outputs more negatively than equivalent human-created content, regardless of actual quality. Following the experimental labeling paradigm, this study examined whether identical couple images (H_1a) and couple counseling excerpts (H_2a) were evaluated less favorably when labeled as AI-generated rather than human-created, and whether AI attitudes and AI literacy moderated these effects for images (H_1b) and counseling dialogues (H_2b). Two consecutive online experiments were conducted in 2024 with a national sample of adults in Germany (N = 2,658). In Experiment 1, identical romantic couple images received less positive evaluations when labeled as AI-generated images versus as human-generated photographs (d = .21; H_1a). In Experiment 2, identical sexuality-related couple counseling excerpts labeled as involving an AI counselor were rated less favorably than those labeled as involving a human counselor (d = .23; H_2a). AI attitudes and AI literacy combined moderated the labeling effect for images (η² = .01; H_1b) but not for counseling dialogues (η² = .003; H_2b). These findings extend the literature on anti-AI bias into intimate contexts. They also underscore the importance of considering user dispositions toward AI when designing and implementing generative AI systems in intimacy- and sexuality-related domains.