Visual misinformation on Facebook.

This article presents the first large-scale study of image-based political misinformation on Facebook, analyzing over 13 million posts from more than 14,000 pages and 11,000 public groups during August to October 2020. Using perceptual hashing and computer vision to identify duplicate images and political figures, the study finds that approximately 23% of sampled political images and 20% of images featuring political figures contained misinformation. The research highlights significant partisan asymmetry, with right-leaning images being 5 to 8 times more likely to be misleading, though misleading images did not show higher engagement levels. This work emphasizes the importance of studying visual political content, which has been largely overlooked compared to link-based misinformation, and demonstrates the potential of computer-assisted methods to analyze large-scale image data.