How AI-generated images affect human preference and trust in websites

AI-generated images (AIGIs) are increasingly integrated into websites, yet their effects on user perception in contexts where trust matters remain underexplored. We report a within-subjects online experiment (N=43) in which participants evaluated six websites containing either photographic imagery or AIGIs, first without and then with disclosure of AI use. Without disclosure, participants tended to prefer websites using AIGIs, while trust judgments were more evenly distributed. When AI use was disclosed, trust shifted clearly toward websites using photographic images. These effects varied by website context, with higher tolerance in entertainment settings and stronger trust penalties in government and health settings. Open responses indicate a tension between visual appeal and perceived authenticity. While polished AI imagery enhanced aesthetic appeal, disclosure raised concerns about authenticity and credibility. We show that disclosure reshapes trust in website contexts and that acceptance of AIGIs depends on contextual expectations rather than image quality alone.