In contemporary online journalism, photographs play a central role in convincing users of the veracity of news. This can push forward "fake news", and eclipse real information. Although prior research on textual misinformation shows that directing attention to specific textual elements influences perceived authenticity, comparable research on images remains scarce.
We investigated which types of visual content influence the subjective perception of photographs as real or fake, and how visual attention to these elements affects authenticity judgments.
All stimuli were genuine press photographs obtained from the Polish Press Agency/European Press Agency. A large, representative Polish sample(N = 327) provided authenticity judgments along with narrative justifications. Qualitative analysis of these justifications yielded 15 content categories (e.g., Poverty, Protest, Child, Natural Disaster, Medical, Military, Fire, Wounds). Through another study, we defined Regions of Interest (ROIs) in the images. Machine learning classifiers were trained to predict authenticity judgments based on attributed content. A free-viewing eye-tracking study (N = 50) examined how visual attention predicted authenticity decisions. Participants were unaware that all photographs were real and that they would later evaluate authenticity. Mixed-effects models were applied using multiple gaze measures.
Classifiers achieved a minimum accuracy of 75%. Feature analysis identified Fire, Protest, and Natural Disaster as the strongest predictors of fake judgments. Eye-tracking analyses showed that sustained attention to Medical and Wounds content increased the likelihood of judging images as real, whereas attention to Child, Military, and Natural Disaster decreased it.
These findings demonstrate that specific visual contents, and sustained attention to them, can shape perceived authenticity and provide a basis for automated systems capable of identifying images likely to be perceived as misleading.