How to Spot a Deepfake: Why Confident Voters Get Fooled Most
Quick answer
How can you spot a deepfake video of a political candidate?
Slow down and run a few quick checks before reacting. Look at blinking and eye movement, whether shadows fall in one consistent direction, the edges around the hairline and jaw, and reflections. No single check is reliable alone, so combine them, and treat a sudden strong emotion as a warning sign.
A video shows up in your feed two days before an election. A candidate, someone you've seen speak a dozen times, is saying something that makes your stomach drop. You watch it. Maybe you share it. Maybe you don't, but something shifts in the back of your mind. That feeling stays.
Here's the part that should keep you up: it doesn't matter if the video was real.
New research from Utah Valley University confirms that a fake political video can change how you think and feel about a candidate just as powerfully as a real one, and the people most convinced they'd catch a deepfake are the ones most likely to miss it.
The Study Nobody in Politics Wants to Talk About
Researchers at Utah Valley University (UVU) just published findings that should make every voter, and every election official, genuinely uncomfortable. TechBuzz News covered the release, and the headline is exactly as blunt as it sounds: fake videos and real videos moved voter opinion by the same amount. Full stop.
Detect Deepfakes Before They Detect You
Knowing how to spot a deepfake starts with understanding why the task is so hard in the first place. The researchers tested participants across the political spectrum, different ages, different backgrounds, different levels of media experience. They showed some people real political footage. They showed others deepfakes, AI-generated videos (meaning video created by software to look exactly like a real person saying or doing something they never actually said or did). Then they measured opinion change.
According to Fox 13 Now, opinion shifts landed between 15 and 19 percent across tested groups, and that number held whether the footage was authentic or completely fabricated. That's not a rounding error. That's a result that says, functionally, your brain cannot tell the difference. This article is part of a series, start with 1 In 3 Teens Now Hit By Fake Ai Nudes Heres What To Do Tonig.
No demographic group performed better than any other. Young people, older voters, politically engaged people, casual news consumers, everyone was equally susceptible. Brandon Amacher, the UVU researcher leading the study, has pointed to this finding as the piece that should reshape how we talk about deepfakes in elections. It's not a fringe problem. It's not a "media literacy" gap that education can simply fix. It's a feature of how human brains process video.
The Confidence Trap Is Real, and It's Set for You
Pay Attention to How You React, Not Just What You See
Here's where it gets genuinely uncomfortable. The UVU research didn't just show that deepfakes work. It revealed something about who they work on most effectively, and the answer is maddening.
The people most confident they could identify a fake were the most likely to be fooled by one.
Think about that for a second. Your skepticism, that "I'm too smart for this" feeling, is not a shield. It may actually be the thing that lowers your guard. You stop looking for problems because you're already sure you'd spot them. The video plays. The emotional response fires. And your brain has already moved on before your critical thinking has a chance to catch up.
This is not about intelligence. It's not about political affiliation. It's biology. Our brains are wired to process moving images and human faces as real and meaningful. Deepfakes are purpose-built to exploit exactly that. They don't need to fool you completely, they only need to plant a feeling, and feelings are faster than fact-checking.
"AI deepfake media can sway public opinion as effectively as real media." UVU Research Finding, as reported by KSL.com
Amacher has also pointed to something even more sobering: we currently have no institution, no rapid-response authority, no public body whose job it is to flag political deepfakes in real time during an election cycle. When a fake video goes viral on a Tuesday night before polls open Wednesday morning, there is no official voice saying "that's not real" fast enough to matter. According to KSL's coverage of the research, Amacher has pushed directly for that kind of infrastructure, and so far, it doesn't exist. Previously in this series: Your Id Has Secret Ink And Its Why Fake Faces Dont Fool The .
This Isn't a Problem for Later. It's Already Here.
Voice Cloning Makes the Video Deepfake Problem Worse
You might be tempted to file this under "something to worry about eventually." Don't. The 2026 election cycle is already showing the damage. It's also not just video anymore. Voice cloning now lets bad actors fake a candidate's voice on a robocall or a leaked audio clip, no video deepfake required, which means the same emotional trap works through your ears alone.
RoboRhythms tracked real-world deployment of AI-generated political content in the 2026 midterms and found something that should make you set down your phone for a moment: nearly half of voters said deepfakes had influenced their election decisions, even though most of those same voters said they don't trust the technology. Read that again. They knew deepfakes exist. They didn't trust them. And they were still influenced by them.
That's the mechanism. You don't have to believe a fake video. You just have to see it. The doubt it plants, the emotional temperature it shifts, that happens before your rational brain gets a vote. A synthetic video can make you feel less certain about a candidate, more angry, more anxious. Those feelings then shape how you interpret everything else you read and hear afterward. The fake doesn't have to win an argument. It just has to color the room.
Meanwhile, TrueScreen has documented something critical about the technology itself: the tools that create deepfakes are improving far faster than the tools that detect them. Early deepfakes had tells, a weird blink pattern, odd lighting around the hairline, a slight audio lag. Those glitches are largely gone now. Generating a convincing deepfake takes a smartphone and a few minutes. Detecting one reliably? That's still an open problem for researchers with serious computing resources. The arms race is not a fair fight.
Why This Matters Right Now
- 🧠Your brain is the targetDeepfakes don't need to fool your intellect. They only need to trigger an emotional response, and they're engineered to do exactly that with moving images of human faces.
- âš¡ Confidence makes you more vulnerable, not lessUVU's own research shows that people who think they can spot fakes are statistically the worst at detecting them.
- 📊 There's no rapid-response systemNo official body currently exists to debunk political deepfakes in real time during elections. By the time fact-checkers catch up, the video has already done its work.
- 🔮 The creation-detection gap is wideningMaking a convincing deepfake now requires only a phone. Detecting one reliably still requires tools most institutions don't have.
So What Do You Actually Do With This?
Check the Eyes, the Lighting, the Shadows, and the Reflections
Look, nobody's saying you need to stop watching video entirely or treat every clip you see as suspect until proven innocent. That's not a realistic way to live. But there are a few genuinely useful things you can do right now, before the next election cycle heats up. If you want a practical way to spot deepfake video, start with a slow check of the eyes, the lighting, and the shadows: unnatural blinking, flat or mismatched lighting across the face, and shadows that don't line up with the light source are still among the more reliable tells, even as the technology improves.
The World Economic Forum has written about the psychological mechanics at play here, specifically how the emotional impact of disinformation lands before critical thinking has a chance to engage. Their guidance echoes what the UVU research implies: the most important habit isn't spotting fakes. It's pausing before sharing anything that makes you feel a strong, sudden emotion. Anger. Disgust. Shock. Those are exactly the feelings that deepfake political content is designed to produce, because emotionally activated people share first and think second. Up next: Government Login Identity Verification Malta What It Means F.
The pause is the whole game. Not a long pause. Just enough of one to ask: "Where did this come from, and is that source accountable for what they publish?" That single question filters out an enormous percentage of manipulative content, deepfake or otherwise.
If you've ever looked at a photo, a video, or even a profile online and thought "I'm not sure this is actually who it claims to be", that instinct is exactly right. Verifying whether a face, a voice, or an identity is authentic used to be something only investigators and journalists needed to worry about. Now it's a skill every voter, every employee, every person scrolling a feed at 11pm needs some version of. Tools built to certify the authenticity of visual media, to give you a real answer instead of a gut feeling, are no longer a luxury for specialists. They're the logical next step when your eyes alone can no longer do the job.
A deepfake doesn't have to fool you completely to work. It only has to make you feel something, and then your brain does the rest. The UVU study didn't just prove fakes are convincing. It proved that our instinct to trust ourselves is the exact vulnerability being exploited.
If a realistic video showed up in your feed right before an election tomorrow, something that showed a candidate doing or saying something damning, what would your first instinct be? To believe it? Question it? Share it to ask what others think?
Whatever your answer, the UVU researchers would tell you this: the group that answered "question it" with the most confidence also turned out to be the group that got it wrong the most often. The trap isn't fooling people who are asleep. The trap is set specifically for people who think they're wide awake.
Learning how to spot a deepfake is less about becoming a forensic expert and more about slowing down at the exact moment a video wants you to speed up. The strongest deepfake detector most people have is not software, it is a five-second pause paired with a habit of checking eye movements, lighting, and edges before reacting. That pause is free, it works on video deepfakes and audio clips alike, and it does not require any special equipment.
Reliable detection of video deepfakes is still hard for machines, which is exactly why human habits matter so much right now. When you check a clip, look at eye movements first. Real eyes drift, blink irregularly, and refocus in small, imperfect ways, while some deepfake systems still render eyes that track too smoothly or blink on an unnatural rhythm.
Next, check the edges of the face against the background, especially around the hairline and the jaw. Lighting is another useful check. In an authentic video, shadows fall in one consistent direction based on the actual light source in the room, but many deepfakes struggle to keep lighting and shadows physically consistent across a moving face.
Reflections offer a smaller but still useful check. If someone is wearing glasses, sitting near a window, or standing in front of a mirror, look at what the reflections show. A real reflection matches the room around the person, while a synthetic one can look blurry, delayed, or slightly wrong compared to the rest of the scene.
None of these checks are foolproof on their own, and detect deepfake tools built by researchers still outperform the human eye in controlled tests. But stacking two or three of these checks, eyes, lighting, shadows, reflections, gives you a meaningfully better shot at catching a fake than relying on gut instinct alone, which is exactly the instinct the UVU research says fails the most confident viewers first.
Look closely at the skin too, since it often gives away more than the eyes do. Some deepfake artifacts show up as texture problems: skin appear too smooth in one frame and too wrinkly in the next, especially around the eyebrows and hairline, right where the software has to blend two different faces together.
Lighting checks work best when you look for clear spatial cues, like where a shadow should fall based on a window or lamp actually in the room. Visual inconsistencies in that consistency, a shadow bending the wrong way or a highlight that jumps between shots, are still one of the more reliable signs that a clip has been manipulated.
Not every manipulated clip is a video at all. Synthetic media now includes audio-only fakes, and ai-generated deepfakes of a candidate's voice can be built from just a few minutes of public speech pulled from old interviews or campaign footage.
A deepfake voice clip is often harder to catch than a video, because you lose the visual tells like eyes and lighting and have to rely on tone, pacing, and the small, uneven breathing sounds that a real human voice naturally has.
Learning to detect deepfakes is not a one-time skill, it is a habit you build clip by clip. Pay attention to your own reaction first, since a sudden jolt of anger or fear is often the biggest clue that something was built specifically to manipulate you.
None of this means you need special software to get started. Most people can improve their odds of spotting deepfakes just by slowing down and running through the same handful of checks every time a suspicious clip lands in their feed.
Detection tools built by researchers are improving, but they are not something most voters can access in the middle of an election cycle. That gap is exactly why personal habits around spotting deepfakes matter so much right now, and why they will keep mattering as the technology gets better.
Local news outlets and fact-checking groups are often the fastest way to confirm whether a viral clip is one of these deepfakes or genuine footage. Waiting a few hours for that kind of confirmation costs you very little and protects you from spreading something false to everyone you know.
Some people assume deepfakes only target national elections, but the same techniques show up in local races, school board contests, and even workplace disputes. The tell-tale signs, mismatched lighting, odd blinking, a voice that sounds slightly off, are the same no matter the size of the race or the size of the audience.
It also helps to remember that not every low-quality video is a deepfake. Bad lighting, a cheap webcam, or a heavily compressed upload can all create the same kind of visual noise that deepfake detection tools are trained to flag, so context still matters more than any single visual clue on its own.
The goal is not paranoia. The goal is a short mental checklist: check the eyes, check the lighting, check the edges of the face, and check your own emotional reaction before you share anything. That checklist takes seconds, it works whether the fake is a full video or just a manipulated voice clip, and it is the same habit the UVU research suggests confident viewers skip first.
Frequently asked questions
How to spot a deepfake when the video looks completely convincing?
New research from Utah Valley University found that opinion shifts of 15 to 19 percent happened whether footage was real or fake, meaning visual inconsistencies people rely on have largely disappeared. Early deepfakes had tells like odd blinking or lighting around the hairline, but those deepfake artifacts are mostly gone now, so pay attention to your emotional reaction rather than assuming your eyes will catch something wrong.
Are confident people better at detecting deepfakes?
No. The UVU study found the opposite: people most confident they could identify a fake were the most likely to be fooled by one. This isn't about intelligence or political affiliation, it's how brains process moving images and faces. Assuming you'd notice manipulation actually lowers your guard, since you stop watching for problems once you're sure you'd spot them.
Can voice cloning make deepfakes harder to detect than video?
Yes. Voice cloning lets bad actors fake a candidate's voice through a robocall or audio clip with no video needed, triggering the same emotional response through sound alone. Research tracking the 2026 midterms found nearly half of voters said deepfakes influenced their decisions even though they distrusted the technology, showing that seeing or hearing synthetic media is enough to shift feelings before facts catch up.