AI outperforms people on deepfake photos but loses on deepfake videos
Mar 7th 2026
A University of Florida study finds AI is far better than humans at spotting fake faces in photos but struggles with video deepfakes, where people spot fakes more reliably.
- Algorithms reached up to 97% accuracy on still-image deepfake faces while human participants performed at chance.
- For deepfake videos, algorithms performed at chance while humans correctly identified real versus fake about two thirds of the time.
- Humans appeared to rely on motion, facial expression and timing cues that the algorithms struggled to interpret.
- Higher analytical thinking and internet skills improved human video detection while being in a better mood reduced accuracy.
- The study used hundreds of curated images and videos and thousands of participants and was published Jan. 7 in Cognitive Research: Principles and Implications by researchers at the University of Florida.