Self-Assessed Detection Ability for Identifying AI-Generated Visual Content: Evidence from Saudi Arabia

Abstract

This study examines individuals’ ability to distinguish between genuine and generative AI (Gen-AI) content across visual media, addressing the growing erosion of trust in digital information. Using a two-part questionnaire, participants were assessed on classification accuracy and  self-assessed detection ability used to identify Gen-AI images and videos. The cross-sectional perceptual classification study design employed varied content types and randomized presentations to reduce perceptual bias. Results showed that participants demonstrated limited ability to accurately distinguish AI-generated from genuine content, with overall performance only slightly above chance level. Signal Detection Theory analysis revealed low perceptual sensitivity and a minimal authenticity bias toward classifying content as real. One-way ANOVA and MANOVA results indicated no significant differences in actual performance or self-assessed detection ability according to AI experience level. Regression analyses further showed that self-assessed detection ability generally did not predict actual performance. The findings highlight the need for stronger visual literacy, specialized training, and improved verification approaches for AI-generated media.