What Is a Deepfake?
Imagine watching a video of a friend announcing that they’ve won a trip to the moon. Their face moves, their voice sounds right, and they seem delighted. But your friend never recorded it. Someone used artificial intelligence (AI) to make the video.
That is a deepfake: AI-made or AI-altered video, audio, or imagery that makes a person appear to do or say something they did not. A deepfake might swap one face for another, change the words spoken in a video, or copy someone’s voice. Some are made for fun and clearly labeled. Others are made to fool people.
The important lesson is not “never trust a video again.” It is “a realistic video deserves a little checking before you believe or share it.”
How Does AI Make a Deepfake?
AI is computer technology that can find patterns in information. If that idea is new to you, our guide to what AI is and how it works is a friendly place to start.
To make a deepfake, a creator may give an AI system examples of a person’s face or voice. The system uses patterns in those examples to produce new pictures, movements, or sounds. It is not finding a secret recording of something that really happened. It is making new material that resembles a recording.
For example, imagine drawing a flipbook. Each page shows a face in a slightly different position. When you flip the pages, the face seems to move. A deepfake video works differently and is far more complex, but it also depends on many images shown one after another. AI can create or change those images so a face appears to speak or react.
Audio can be altered on its own. A voice clone is AI-generated speech designed to sound like a particular person. Video and cloned audio can also be combined, making it seem as though someone said words they never spoke. If you’re curious about the name, our explanation of AI, machine learning, and deep learning breaks down the technology behind many such tools.
Why Can Deepfakes Be So Convincing?
We use faces and voices as clues every day. If you see your teacher on a classroom video, you expect it to be your teacher. If your brother calls, you recognize his voice before he says his name.
A deepfake borrows those familiar clues. It may place a recognizable face into a believable setting, add natural-sounding speech, and arrive with a caption that tells you what to think. Even if the recording is imperfect, an exciting or frightening story can make you want to react quickly.
That is why context matters as much as appearance. A polished video can still carry a false claim. A shaky video can be real. Odd lip movements or strange lighting may be reasons to look closer, but their absence does not prove a clip is genuine. Researchers at the National Institute of Standards and Technology (NIST) study several ways to test media, including manipulation detection and source verification; they do not rely on one visual clue alone.
Are Deepfakes Always Bad?
No. The technology used to create realistic faces and voices can support storytelling and communication. With someone’s permission, a creator might make an imaginative scene for a film or use synthetic speech to help a person who has lost the ability to speak. The U.S. Federal Trade Commission (FTC) notes that voice cloning has potential benefits for people who need help communicating.
The difference is consent and honesty. Did the person agree to the use of their likeness or voice? Does the audience know what was made with AI? A clearly labeled, playful scene featuring a willing participant is very different from a fake message pretending to be a real emergency.
Deepfakes can cause harm when they impersonate people, damage reputations, spread false stories, or try to steal money. They can also make life harder for people whose real recordings are questioned. The best response is neither to believe everything nor to dismiss everything. It is to ask for evidence.
How Can You Check a Suspicious Video or Voice?
You do not need special software to make a useful first check. Try these steps before you share a surprising clip or act on a call:
- Pause. Be extra careful when a message demands immediate payment, secrecy, or a quick repost. A rush can keep you from checking.
- Find the original source. Who first posted the recording? Is it on the person’s official channel, or only in a cropped repost? Look for the full video and its original date.
- Look for independent confirmation. If a clip claims something important happened, see whether trustworthy sources or the person involved confirm it.
- Notice clues, but don’t treat them as a test. A voice that sounds odd or a mouth that seems out of step with speech may raise questions. A clip that looks perfect can still be fake.
- Contact the person another way. If someone appears to ask you for money or private information, use a phone number or account you already know is theirs—not one supplied in the suspicious message.
That last step is especially powerful. The FTC’s advice on family emergency scams warns that a scammer may use a voice clone that sounds like a loved one. The agency recommends calling that person back using a number you know is correct, or checking with another trusted person.
A Make-Believe Example: The Urgent Call
Suppose you get a call that sounds exactly like your cousin. The caller says they are in trouble and need you to send money right away. They insist that you must not tell anyone.
Your first feeling might be fear. That is understandable. But sending money is not the only way to help. You could hang up, call your cousin using their saved number, or contact another family member. If your cousin is safe, you have avoided a scam. If they really need help, you can still respond once you know what happened.
The key is to verify the story, not just recognize the voice. A familiar sound is a clue, but it is no longer enough on its own. The same rule works for an unexpected video call, a dramatic social media clip, or an audio message apparently sent by a friend.
Can Technology Tell Us What Is Real?
People are developing tools that look for signs of AI manipulation. Another approach records information about a file’s origin and changes. Such records are often called provenance—a word for a piece of media’s history.
One system, called Content Credentials, can carry information about how a file was made or edited. That history may be helpful, but it is not a magic “true or false” stamp. A file without credentials is not automatically fake, and knowing where a file came from does not prove that every claim it makes is true. The C2PA’s explanation of Content Credentials describes what this information can tell viewers.
Technology can assist us, but our questions still matter: Who made this? Where did it come from? Can I confirm the claim elsewhere? For more on why an impressive AI result still deserves a thoughtful check, read our guide to AI’s mistakes and limits.
Stay Curious, Not Scared
Deepfakes show just how creative AI can be—and why using that creativity responsibly matters. You do not have to become an AI expert to navigate them. A pause, a second source, and a direct call to someone you trust can make an enormous difference.
So enjoy imaginative AI-made stories when they are shared honestly. When a recording asks you to believe something surprising or do something urgent, give yourself time to check. Curiosity is exciting; careful curiosity is a superpower.
