You may have seen the news this year about fake videos and photos of celebrities and politicians. They’re called deepfakes, and they’re made with artificial intelligence (AI).
They can look and sound remarkably real. That’s what makes them worrying, especially when one shows up in a text or a video call claiming to be a grandchild, a friend, or someone else you trust.
This guide explains what a deepfake actually is, why more of them are showing up now, and how to spot one. It also covers what to do if you ever see a suspicious video or hear a suspicious voice message of someone you know.
What is a deepfake?
A deepfake is a photo, video or audio clip created or changed by AI software so a real person appears to say or do something they never actually said or did.
The name comes from “deep learning,” the type of AI used to build them, combined with “fake.” Older photo editing needed a skilled person and hours of work. AI tools can now do a rough version in minutes, and a convincing one with more effort and a decent computer.
Plain talk: deep learning
Deep learning is a type of AI that learns patterns from huge numbers of examples, like thousands of photos of a face, rather than following rules a person typed in. That’s what lets it generate a new, fake image or voice that follows those same patterns.
Deepfakes come in a few common forms:
- Face swap videos, where one person’s face is placed onto another person’s body in an existing video.
- Fully AI-generated video or images, built from scratch by typing in a text description.
- Voice clones, where a short recording of someone’s real voice is used to make them “say” new sentences.
Why deepfakes are a growing concern right now
Two things changed recently. First, the AI tools that make deepfakes got easier to use and cheaper to access, so people without technical skill can now make one. Second, several high-profile examples have shown how much damage they can cause.
In January 2024, sexually explicit AI-generated images of singer Taylor Swift spread widely on social media before platforms removed them, according to reporting from NPR and CBS News. It became one of the most widely covered examples of nonconsensual deepfake images made of a real person.
That same month, a robocall using an AI-cloned voice of President Biden told New Hampshire voters not to vote in the state’s primary, NBC News reported. Digital forensics expert Hany Farid confirmed it was an AI fake, calling it a sign of “things to come” for elections, according to UC Berkeley’s School of Information.
The Federal Trade Commission (FTC) has been warning about this trend for longer. In a business guidance post, it warned that deepfakes and voice clones are increasingly used for “imposter scams, extortion, and financial fraud.”
Safety tip
You don’t need to become an expert in AI to protect yourself. You mainly need to slow down and check before you trust or act on a video, photo or voice message that surprises or alarms you.
Why families get targeted, not just celebrities
Most of the news coverage focuses on famous faces, but the same tools work on anyone. A scammer doesn’t need much: a short video clip posted online, a few seconds of someone’s voice from a social media post, or even a voicemail greeting can be enough raw material.
That’s part of why deepfakes overlap so closely with older, familiar scams. A “grandchild in trouble” phone call is not new. What’s new is that the voice on the line can now sound exactly right, because it was built from a real recording of that grandchild. The same idea shows up in romance scams, where a scammer might use an AI-altered photo or a short video call to seem more convincing. Our guide on romance scams and older adults covers that pattern in more detail.
How to spot a deepfake video or photo
No single sign proves a video is fake, but a combination of small details usually gives it away. Researchers at the MIT Media Lab’s Detect Fakes project recommend looking closely at the face, since that’s where most deepfakes struggle.
Watch the face closely
| Look for | What’s suspicious |
|---|---|
| Blinking | Too rare, too frequent, or oddly timed |
| Skin texture | Too smooth in patches, or doesn’t match the person’s age |
| Shadows around eyes | Shadows that don’t match the light in the rest of the scene |
| Glasses (if worn) | Glare that doesn’t shift naturally as the head moves |
| Facial hair or moles | Looks pasted on, or missing where it should be |
| Lip movement | Doesn’t quite match the words being spoken |
Check the whole picture too
- Edges and hair: AI often struggles with fine detail like flyaway hair or the border where a face meets the background.
- Lighting and reflections: Does the light on the face match the light in the room or the background?
- Audio quality: A cloned voice can sound slightly flat, rushed, or have odd pauses between words.
- Context: Would this person really say this, on this platform, right now? A shocking or urgent claim is a red flag on its own.
Try this
Pause the video on a single frame where the face is turned slightly, not straight at the camera. Deepfakes tend to fall apart most at odd angles, since the AI has less real footage to work from at that angle.
Use a second source to check
The most reliable check often has nothing to do with the video itself. Search for the same claim from a source you trust, such as a major news outlet or the person’s own verified account. If something this big were true, other reliable sources would be reporting it too.
What to do if you see a suspicious video of someone you know
This is the situation that matters most for families: a video, voice message or call that sounds like your spouse, child, grandchild or friend, asking for money, help, or personal information.
- Don’t act on it right away. Scammers rely on you panicking and responding fast.
- Don’t reply to the number or account the message came from. If it’s a fake, replying only confirms your number is active.
- Contact the real person directly, using a phone number you already had saved for them, not one given in the message. Call or text them on a different device if you can.
- Ask a question only they would know, or agree on a family safe word in advance. Our guide on setting up a family safe word walks through this in five minutes.
- If money is requested, treat it as a scam until you’ve confirmed otherwise through that direct contact. Legitimate emergencies rarely require wire transfers, gift cards or cryptocurrency.
- Report it. Our guide on where to report a scam lists the right agency for your country.
Say this
If a caller’s voice sounds right but something feels off, you can simply say: “I’m going to call you straight back on the number I already have for you.” Then hang up and do exactly that.
The FTC’s consumer alert on family emergency scams makes the same point: verify through a channel you control, not the one the caller gave you.
Plain talk: voice clone
A voice clone is an AI copy of a real person’s voice, built from a short recording. It can be used to make fake phone calls that sound like someone you know.
Practicing before you need it
You can also try asking an AI assistant to help you think through what a deepfake might look like, so you know what to watch for.
Type or say this
“Can you explain, in plain language, three signs that a video might be an AI deepfake?”
If you’re new to AI tools generally and want a gentler starting point, our guide on how to explain AI to an older person is a good place to begin. And if the concern is a voice on the phone rather than a video, our piece on AI voice-clone scams covers that specifically.
Deepfakes are getting better, and they’ll keep improving. But the basic defense doesn’t change: slow down, check the small details, and always verify anything urgent or emotional through a channel you control, not the one the message came from.
Where to go from here
Want more plain-English guides like this one, covering AI tools and how to use them safely? Our book, AI for Seniors, walks through the basics step by step, at a pace that makes sense.
