Recent studies, including a meta-analysis published in Computers in Human Behavior Reports and research led by the University of Florida, show that untrained humans generally perform near chance levels (around 50% to 55% accuracy) when distinguishing deepfakes from real content. While AI algorithms significantly outperform humans at detecting static deepfake photos, humans often retain an edge with deepfake videos by picking up on subtle inconsistencies in movement, facial expressions, and timing. Encouragingly, emerging research from institutions like The Australian National University demonstrates that short, targeted perceptual training can markedly improve human detection accuracy.

• ufl.edu
• truthscan.com
• ufl.edu
• anu.edu.au