Here’s a finding that flips the usual AI story on its head: to spot deepfake videos, you’re actually better equipped than the algorithms built specifically to catch them. A recent University of Florida study found that while AI easily outperforms humans at spotting fake still photos, that advantage disappears almost completely once the fakes start moving.
AI detection tools caught fake photos with up to 97% accuracy, while human participants performed no better than chance on the same images. But for deepfake videos, the AI dropped to chance-level performance, while humans correctly identified real versus fake about two-thirds of the time.
The Study That Flipped the Script
Psychologists and computer scientists at the University of Florida ran thousands of participants through hundreds of real and fake images and videos, featuring both static faces and people talking. They then ran the exact same media through detection algorithms and compared the results side by side.
| Media Type | Human Accuracy | AI Accuracy |
|---|---|---|
| Still photos | No better than chance | Up to 97% |
| Videos | ~67% (two-thirds correct) | Chance level |
Why Humans Win on Video
According to the researchers, people appeared to pick up on subtle inconsistencies in movement, facial expressions, and timing — cues the algorithms simply struggled to interpret. Study co-author Brian Cahill put it plainly: the team was surprised to see humans outperform AI on video, since video offers a richer context with more signals for the human brain to catch than a single frozen image does.
Interestingly, the researchers also found that a person’s own state affected their accuracy. Participants who scored higher in analytical thinking and general internet skills detected fakes more reliably, while people who reported being in a better mood actually performed worse — possibly because a positive mood makes people more trusting by default.
The study tested specific, controlled types of faces and videos, which may not fully capture the chaos of real-world social media content. Both AI generation and detection technology are also improving quickly, so this balance between humans and machines could shift again soon.
How to Actually Spot a Deepfake Video Yourself
Based on what the research points to, here’s what to actually watch for:
- Timing mismatches: Pay attention to whether lip movements, blinks, or reactions land a fraction of a second off from what’s being said or shown.
- Unnatural facial expressions: Look for expressions that feel slightly “off” or don’t fully match the emotional tone of the speech.
- Movement inconsistencies: Watch how a person’s head, hands, and body move together — AI-generated video often handles these less fluidly than a real recording.
- Your own mood: If you’re in a great mood while scrolling, know that you may be more trusting than usual — it’s worth applying a bit more scrutiny, not less.
Frequently Asked Questions
Does this mean AI detection tools are useless for video?
Not useless, but currently weaker than human judgment specifically for video, according to this study. AI still holds a clear, large advantage for detecting fake still images.
Will this balance stay the same over time?
Unlikely. The researchers themselves note that both deepfake generation and detection technology are evolving rapidly, so today’s human advantage on video could shrink or disappear as detection tools improve.
What’s the single best habit for avoiding being fooled?
As one of the study’s researchers put it, you don’t need to personally verify everything you see — but staying alert, questioning surprising content, and looking for supporting evidence matters more than ever as deepfakes keep improving.
ChatGPT vs Gemini vs Claude: Which One Should You Use in 2026 — for more on how today’s leading AI systems actually compare.