How to Spot an AI Deepfake Before It Fools You
Slug: how-to-spot-an-ai-deepfakePillar: Technology > Online SafetyKeyword: how to spot an ai deepfakeExcerpt: Deepfakes in 2026 don't have weird teeth or bad lighting anymore. Here's what actually still gives them away, and the tools that can check for you.
The old advice for spotting a deepfake — weird teeth, flickering edges, bad lighting — doesn't reliably work anymore. Newer AI video models have mostly fixed those obvious tells. What still gives a deepfake away in 2026 is subtler: eyes, shadows, and the edges of physics the AI hasn't fully learned to fake yet.
Why the Old Tricks Stopped Working
Early deepfakes were crude enough that a careful look usually caught them — warped teeth, mismatched skin tone at the jawline, obvious blending seams around the face. Newer models have mostly solved those problems. So "check if the lighting looks off" is advice from a couple of years ago, built for tools that don't exist anymore. What you're looking for now is different: not obvious glitches, but places where the fake still has to guess at something the real world doesn't need to guess at.
What Actually Still Gives It Away
Start with the eyes. AI still struggles to simulate natural blinking patterns — either the blinking looks too regular, too rare, or the reflections in each eye don't match each other or the room's light sources the way real eyes do. If someone in a video has been talking for thirty seconds without blinking once, that's worth a second look.
Shadows and reflections are the next tell. Deepfake generation handles a face well but tends to fumble the surroundings — shadows that don't fall the direction the light source suggests, or reflective surfaces (glasses, windows, wet surfaces) that don't show what they logically should. And skin itself is a giveaway in a different way: real skin has texture that varies — pores, faint asymmetry, the occasional blemish or sun spot. Deepfake skin tends toward a strange uniformity, smoother and more even than any real face actually is up close.
Audio has its own tells, and they're often easier to catch than the video ones. Listen for a flat, slightly robotic tone, unnatural pauses between words, or audio that's oddly clean — missing the ambient room noise, echo, or background hum you'd expect from wherever the video claims to be filmed.
The Tools That Can Check For You
You don't have to do this by eye alone. Microsoft's Video Authenticator analyzes footage frame by frame and returns a confidence score on whether manipulation is likely, by detecting the blending boundary that deepfake generation leaves behind. Intel's FakeCatcher takes a genuinely different approach — it looks for subtle color changes in skin caused by blood flow with each heartbeat, something real faces show and deepfakes typically don't replicate. Neither tool is something the average person has installed, but they exist, and organizations dealing with verification at scale increasingly use them.
Where This Actually Matters Most
Video deepfakes get the headlines, but the more common real-world risk right now is audio-only: a cloned voice in a phone call, not a face on video at all. If a call sounds like a family member but something about the urgency or the request feels off — asking for money, gift cards, or account details under pressure — hang up and call them back on a known number. That single habit defeats almost every voice-cloning scam regardless of how good the audio gets.
The honest takeaway here: as these models keep improving, visual detection by eye is going to keep getting harder, not easier. The more durable defense isn't learning to spot every future generation of fake — it's building habits, like verifying unexpected requests through a second channel, that work no matter how convincing the fake becomes.
A Few Habits Worth Adopting Now
Agree on a family safe word nobody would guess from social media, and use it to verify any unexpected, urgent request over the phone. Be skeptical of videos that arrive with heavy pressure to act fast — urgency is a manipulation tactic regardless of whether the footage is real, and scammers lean on it precisely because it stops people from pausing to check. And if something feels slightly wrong about a video or call, trust that instinct enough to verify independently before you act on it.
For more on protecting your family from AI-driven scams, see our guide on online safety, and our technology hub for the rest of our how-to-tech coverage.
FAQ
Can you still spot a deepfake just by looking at the teeth or lighting?
Not reliably anymore. Those were tells for older, cruder AI models. Current tools handle those details well enough that they're no longer a dependable giveaway.
What's the single biggest tell in a modern deepfake video?
Eyes and reflections — inconsistent blinking patterns or eye reflections that don't match the room's actual light sources are still one of the harder things for AI video to fake convincingly.
Are deepfake detection tools available to regular people?
Some, though the most sophisticated ones like Intel FakeCatcher are mostly used by organizations rather than built into consumer apps yet. For most people, verification habits matter more than detection software.
How do I protect my family from an AI voice-cloning scam?
If an unexpected call asks for money or sensitive information under pressure, hang up and call the person back on a number you already have saved. Agreeing on a family safe word in advance also helps confirm it's really them.
