Image Quality, Not Disclosure Labels, Shapes the Credibility of Artificial Intelligence–Generated He

Generative artificial intelligence (AI) can produce photorealistic health-campaign images, and regulators ask for labels. This study tested how disclosure labels and image fidelity affect credibility and gaze. In a 2 × 3 mixed design, 155 adults viewed 12 vaccination advertisements during webcam eye tracking. Each appeared as the original photograph or a high- or low-quality AI-generated image, with or without a “Human-Made”/“AI-Generated” label (between subjects). Fidelity affected credibility (sr2 = .107). Low-quality AI images were rated less credible than original photographs (d = 0.82) and high-quality AI images (d = 0.63), whereas originals and high-quality AI images did not differ reliably. The label had no main effect and interacted with fidelity, but no simple effect survived correction. Positive attitudes towards AI predicted higher credibility only for low-quality AI images. Neither factor detectably changed gaze, and the label was looked at on 36% of trials. Image quality, not disclosure, shaped credibility.