How do you know when you're looking at a deepfake?
You often can't be sure by sight alone anymore. The best signals are context and behavior: a shocking claim, urgency, a request for money or secrecy, or a source you can't confirm. Combine visual checks (odd edges, lighting, lip-sync) with out-of-band verification, contacting the person or source directly, before you believe or act.
Why — the first-principles explanation
The honest starting point is that realism is no longer proof. As models improved through 2025-2026, many deepfakes pass a casual glance, so 'it looks real' can't be your test. Detection has shifted from how it looks to whether it makes sense and whether you can confirm it.
The strongest signals are contextual. Deepfakes are usually made to move you emotionally or financially: they show something outrageous, arrive with urgency, ask for money, credentials, or secrecy, or surface from an unverifiable account. Any media pushing you to act fast is exactly when to slow down.
Visual and audio cues still help as a second layer. Look at the face's edges and hair, check that lighting and eye reflections are consistent, and see whether lips track the words. These tells are shrinking, so treat them as hints that raise suspicion, not verdicts.
The decisive step is verification you control. If a video shows your boss demanding a transfer, call your boss on a known number. If a clip claims a public event, check trusted news outlets. Provenance tools, like C2PA Content Credentials, and reverse image or video searches can also reveal whether media was altered or lifted from elsewhere. In short, you know it's a deepfake less by staring harder and more by confirming the story through a channel the faker doesn't control.
An example that makes it click
Imagine getting a text from a friend's number saying 'I'm stranded, wire me $500 now.' Even if it sounds like them, you'd call your friend to check, because the situation is suspicious, not just the message. Deepfakes work the same way: the giveaway is often the urgent, money-shaped ask, not a pixel.
So the reliable move is the phone call, not the magnifying glass. A faker can make a face look perfect, but they can't answer your friend's real phone or match the story your trusted news source tells.
How to do it
- Pause on emotion: treat shocking, urgent, or money-related media as suspect by default.
- Check the source: is it a verifiable account or outlet, or something you can't trace?
- Scan visual tells: face edges, hair, lighting, eye reflections, and lip-sync.
- Verify out of band: contact the person or confirm the event through a trusted, separate channel.
- Use tools: reverse image/video search and check for C2PA Content Credentials or provenance.
Key facts
- Modern deepfakes can pass a casual visual inspection, so realism is not proof.
- Context clues, urgency, money requests, secrecy, unverifiable sources, are the strongest warning signs.
- Out-of-band verification (a known phone number or trusted outlet) is the most reliable check.
- Reverse image/video search can reveal reused or manipulated media.
- C2PA Content Credentials attach tamper-evident provenance to legitimate media.
▶ The 60-second explainer (script)
How do you know you're looking at a deepfake? The uncomfortable answer: you often can't tell by sight anymore. Today's fakes can pass a quick glance, so 'it looks real' isn't a test. Instead, read the situation. Deepfakes are built to push you, they show something shocking, arrive with urgency, ask for money or secrecy, or come from an account you can't verify. That pressure to act fast is your biggest red flag. Visual cues still help as a backup: check the face edges, the hair, the lighting, the eye reflections, and whether the lips match the words. But treat those as hints, not proof. The move that actually works is verification you control. If a video shows your boss demanding a wire transfer, hang up and call your boss on a number you already have. Confirm the story through a channel the faker can't touch. That's how you really know.
What authoritative sources say
People also ask
Can experts always tell if something is a deepfake?
Not reliably by eye. Even specialists lean on detection software, provenance data, and independent verification rather than visual inspection alone.
What's the single best habit against deepfakes?
Verify out of band. For anything urgent or high-stakes, confirm through a known phone number or a trusted outlet before believing or acting.
Does a video being high quality mean it's real?
No. High quality can just mean a good faker. Judge the source and context, not the resolution.
How can provenance tools help?
Standards like C2PA Content Credentials attach a tamper-evident record to media, so you can check whether it was AI-generated or edited.