Only 1 in 1,000 People Can Spot a Deepfake — Here's the 30-Second Habit That Actually Protects You

Here's a number that should stop you mid-scroll: in a 2025 study where 2,000 people were specifically told to look for deepfakes, only 0.1% of them got it right every time. That's one person in a thousand. Not one person in a hundred. One in a thousand — under controlled test conditions, with the challenge spelled out in advance.

Think about that for a second. These weren't distracted people skimming their phones at midnight. They were primed, focused, and looking hard. And nearly all of them still couldn't tell the real from the fake.

So what does that mean for the rest of us? It means the advice you've probably heard — "look for weird blinking, bad teeth, glitchy hands" — is becoming dangerously out of date. And the replacement skill isn't harder looking. It's smarter pausing.

Why Your Brain Was Never Really Built for This

Here's the uncomfortable part. The same study, published by digital identity company iProov, found that over 60% of participants felt confident in their deepfake-spotting abilities — regardless of whether they were actually correct. Read that again. People felt sure of themselves whether they got it right or wrong. Confidence and accuracy had almost no connection to each other.

Psychologists have a name for this gap. When someone lacks skill in a specific area, they also tend to lack the awareness to know they lack it. You can't see what you're missing. The result is that the people most likely to share a deepfake video without question are often the ones most certain they'd never fall for one.

That 55.54% number comes from a peer-reviewed meta-analysis of 56 separate studies, published in the IACIS Journal of Information Systems. Across every format tested — photos, audio clips, videos — humans hover just above random guessing. A coin toss gets you to 50%. Human judgment, apparently, gets you to 55%. That's not a superpower. That's barely a margin of error. This article is part of a series — start with Your Kids Birthday Photo Is All A Stranger Needs And It Take.

The Tech Didn't Stay Still While You Were Learning the Old Rules

To understand why detection got so much harder, you need to understand how fast the tools for creating deepfakes have improved — and how cheap they've become.

Between 2023 and 2024 alone, the number of deepfakes detected in fraud cases increased fourfold, according to iProov's research. Fourfold in one year. That's not a trend. That's acceleration. And the reason it's happening isn't that some secret lab cracked a hard problem. It's that the tools are now cheap, fast, and widely available. Someone with no technical background can download software and produce a convincing synthetic face in the time it takes to watch a TV episode.

Voice cloning has crossed a genuinely unsettling threshold, too. According to threat intelligence firm ZeroFox, modern voice cloning systems need as little as 3 to 5 seconds of audio to replicate someone's voice with 85% accuracy. Three to five seconds. That's a voicemail greeting. That's a clip from a YouTube video. That's a three-second clip someone posted on social media two years ago.

And video? Video is genuinely harder to judge than photos — for a specific, teachable reason. The iProov study found participants were 36% less likely to correctly identify a fake video compared to a fake image. Why? Because a single still photo only needs to fool you once, in one frame. A video plays out across hundreds of frames, and each tiny inconsistency — a weird flicker around the jaw, a slightly unnatural blink — gets buried in the motion. Your brain smooths over small errors when everything is moving. Video deepfakes exploit exactly that tendency.

"The widespread availability of low-cost deepfake tools allows individuals with minimal technical skills to create highly convincing synthetic content, while the sophistication of generative AI systems continues to advance, producing outputs that are often indistinguishable from content created by humans." — IACIS Journal of Information Systems, 2025

Why the Old Advice Isn't Protecting You Anymore

For a few years — roughly 2017 to 2019 — the "look for glitches" advice actually worked. Early deepfakes were visibly rough. The eyes didn't blink right. Teeth looked like a bad video game render. Hair at the edges of the face seemed to melt. Tech journalists and security researchers hammered the message: look for artifacts (which just means visual errors, the digital equivalent of a smudge on a painting). And for a while, it helped.

But here's the problem. That message created a mental model — deepfakes have visible tells — and that model stuck around long after it stopped being reliable. The AI generating synthetic faces has now cleared the bar where those tells mostly disappear. The smudges got fixed. The teeth look real. The eyes blink at natural intervals. Looking harder at pixels won't save you from a well-made fake, because there's genuinely nothing to see. Previously in this series: That Hot Stranger Sliding Into Your Dms Probably 40 000 Line.

It's a bit like spotting wildlife at dusk. In full daylight with binoculars, you'd never confuse a deer and a coyote. But in dim light, moving fast, your eyes just can't grab the details — the ear angle, the gait, the tail. Staring harder doesn't give you information that isn't there. The smarter move is to check the tracks, listen for surrounding sounds, ask whether this is even an area where coyotes have been spotted. You stop relying on a sense that's been compromised and switch to a system that still works.

Nobody should feel bad for falling behind on this. The advice changed because the technology changed — not because people got careless.

The Habit That Actually Works Now

The researchers who study this keep landing on the same uncomfortable conclusion: the real vulnerability isn't that people can't spot deepfakes. It's that even when people suspect something is fake, the vast majority do nothing about it. They feel uncertain, shrug, and scroll on — which is functionally the same as believing it.

So the fix isn't better eyes. It's a pause with a purpose.

At CaraComp, we work in facial recognition and digital identity, which means we spend a lot of time thinking about how synthetic media exploits the gap between what looks real and what is real. And the pattern we see consistently is this: deepfake-based scams almost always use urgency as a weapon. Someone who looks and sounds like your boss asks you to wire money before the end of the day. A video of a family member says they're in trouble and need you to act now. The pressure to move fast is almost always part of the design — because the longer you pause, the more likely the whole thing falls apart.

That's actually useful information. Here's the 3-step pause that works: Up next: App Store Age Verification Scotus 28 States.

None of these steps require technical knowledge. They don't require you to spot a pixel anomaly or understand how generative AI (software that creates new content from scratch, rather than editing existing content) works. They just require you to slow down for thirty seconds before you react.

Age matters here in two different directions, by the way. The iProov study found that 30% of people aged 55-64, and 39% of those 65 and older, had never even heard the word "deepfake." They don't know the threat category exists. Meanwhile, younger age groups have the opposite problem — high confidence, low accuracy. One group doesn't know to be cautious; the other thinks caution doesn't apply to them. Both groups end up in the same place.

The final thing worth sitting with: in a world where a video of someone's face can be faked in an afternoon using tools anyone can download — and where human accuracy at detecting that fake sits at roughly the same level as a coin toss — the question is no longer "can I spot it?" It's "have I built a habit that works even when I can't?"

Because the scam that gets you won't look fake. That's kind of the whole point.