Deepfake or Real? 7 Ways to Spot an AI-Generated Video
Deepfakes are getting harder to catch by eye, but they still leave tells. Here are seven practical signs that a video or voice is AI-generated, and where human judgment stops being enough.
You can often spot a deepfake by checking the eyes and blinking, the edges of the face, fine details like teeth and hands, inconsistent lighting, unnatural voice rhythm, the surrounding context, and the source. But as the technology improves, these visual tells are fading fast.
A few years ago, you could spot a deepfake from across the room. Today, the best ones can fool a careful viewer. But most deepfakes in the wild aren’t the best ones, and they still leave tells. Here’s what to look for, and the honest limit of what the human eye can do.
1. Watch the eyes and blinking
Early deepfakes barely blinked; modern ones over-correct. Look for blinking that feels mistimed, eyes that don’t quite track where the person is looking, or reflections that differ between the two eyes.
2. Check the edges of the face
Face-swap artifacts hide at the boundary. Watch where the face meets the hairline, ears, jaw, and neck, especially during fast head movement. Flickering, blurring, or a faint seam is a strong signal.
3. Look at teeth, hair, and hands
Generative models struggle with fine, high-frequency detail. Teeth that blur into a single white block, hair that smears at the strands, and distorted hands are classic giveaways.
4. Mind the lighting and shadows
Real light is consistent. If the lighting on the face doesn’t match the lighting on the body or background, or shadows fall in directions that don’t agree, be suspicious.
5. Listen to the voice carefully
Cloned voices often nail the timbre but miss the rhythm: flat emotion, odd pacing, missing breaths, or a subtle metallic artifact. If the audio and lip movements drift out of sync, that’s a major red flag.
6. Question the context, not just the clip
The most useful tell is often outside the video. Is a CEO suddenly promoting a crypto giveaway? Is the “urgent” request bypassing normal process? Deepfakes are usually wrapped in social engineering: manufactured urgency, unusual asks, off-platform payment. This is exactly the pattern behind executive impersonation scams on social platforms.
7. Reverse-search and verify the source
Check whether the clip appears on the person’s or company’s official channels. If a “company announcement” exists only as a paid ad from an unknown page, treat it as fake until proven otherwise.
The bottom line
These seven signs will help you catch many deepfakes, and you should train your team on them, because social engineering targets people first. But the uncomfortable truth is that the artifacts are disappearing as the technology improves.
That’s why brands don’t rely on the human eye. AI-based detection spots synthetic media even when it looks flawless, at a scale no team could match. Revelum scans 20M+ ads monthly at 99.8% accuracy to catch the deepfakes that get past everyone else and removes them in under 24 hours.
We will assess your situation and tell you what we are seeing, typically within 24 hours.
Frequently asked questions
- Can you always tell if a video is a deepfake?
- No. The visual tells are getting subtler every year, and the best deepfakes can pass human inspection. The signs above catch many fakes, but not all, which is why detection increasingly relies on AI, not eyes.
- What's the most reliable single sign?
- Context. Technical artifacts come and go as models improve, but deepfake fraud almost always pairs the fake with an unusual request: urgency, secrecy, or an off-platform payment. That pattern is the most durable tell.
- Why isn't manual checking enough for a business?
- Because the threat is automated and high-volume. A person can inspect one clip; they can't watch millions of ads across every platform, 24/7. At scale, detection has to be automated.