How to Spot Deepfakes: Video, Voice and Photo Clues

Learning how to spot deepfakes means watching for small glitches in a face, a voice or a photo, and then checking any odd request another way. The glitches help, but they keep getting harder to see, so what holds up best is calling back on a number you saved before.

Deepfakes sit alongside other new terms, such as synthetic media and voice cloning, and our glossary of everyday AI words defines each in brief.

What makes something a deepfake?

Something is a deepfake when AI has been used to make a real person seem to say or do something they never did. It can be a video, a voice recording or a still photo. The point is impersonation, so the fake borrows a face or voice that people already trust.

A deepfake can look entirely real while every detail of it was changed or made with AI. For a small business, deepfake examples tend to look ordinary:

• Video. A video call where "the owner" asks the office to pay a rush invoice today.
• Voice. A voicemail from a known vendor asking you to update their bank details.
• Photo. A picture of the owner signing an order he never placed.

Voice fakes have their own playbook, which our guide to AI voice cloning scams walks through. For pictures that were made from scratch rather than borrowed from a real person, see how to tell if an image is AI-generated.

How to spot deepfakes by sight and sound

Spotting deepfakes by sight and sound means looking for things that do not quite match. Faces may blur at the edges, and lips may drift out of sync. A voice may sound flat or pause oddly. None of these clues proves a fake, and a good fake may show none.

When you need to detect a deepfake video call, ask the person to turn their head fully or wave a hand in front of their face. Those moves often make a fake flicker. In audio, listen for breathing that never happens and a tone that never changes. In photos, check hands, shadows and text in the background.

Tools fall short too. A GAO tech spotlight (opens in a new tab) found that deepfake detection has limited effectiveness in real-world use, so a detector can be wrong in both directions. NIST's report on synthetic content (opens in a new tab) adds that watermarks can often be removed and hidden file details stripped. So the lack of a mark tells you little. That is why the question of how to detect deepfakes is really about your process.

Which office habit beats a convincing fake?

The habit that beats a convincing fake is simple. When any call, voice note or message asks for money or private data, end it and reach the person on a number from your files. Pair that with one rule: nobody pays or changes payment details on a request alone.

That habit never asks anyone to judge the fake. A flawless video of the owner still fails when the bookkeeper hangs up and dials the owner's saved cell. It also gives staff a polite script. That matters when the face on screen looks like the boss.

A concrete contractor's Finance Agent holds every new payment for a person to release. A video call from "the owner," out on a job, asks the office manager to pay a new supplier before noon. The picture stutters when he turns his head, so she leaves the payment held and calls his saved number. The owner never called, and he briefs the team that afternoon.

Our Finance Agent holds every drafted payment until a person releases it, whatever a video call or message says. You can add it as one of our six AI Agents, built around your approval steps. Every payment and bank edit the Finance Agent prepares waits for a person who has confirmed it by phone.

Steps to take once a deepfake gets through

After a suspected deepfake, stop the conversation and verify through a channel you trust. Do not reply inside the same call or thread. If a payment already went out, ring your bank via the number on your statement. Then tell your team so nobody else falls for the next attempt.

Save the video link, the caller's number and the time, and report the account where it appeared. If the fake used your face or voice, warn your customers. Our post on keeping business data safe with AI covers the separate question of what your own tools share.

Frequently asked questions

What is a deepfake?

A deepfake is a video, audio clip or photo that AI has made or changed so a real person seems to say or do something they never did. The word joins deep learning, a kind of AI, with fake. Scammers use deepfakes to borrow trust from a familiar face or voice.

Can deepfakes be detected?

Sometimes, but not reliably. Careful viewers and detector tools catch some fakes, yet both miss others and sometimes flag real media. Treat any clue as a reason to verify, not as proof. Phoning the person on a saved number settles the question faster than any tool.

How do you protect against deepfakes?

Protect against deepfakes with a few firm habits. Verify any request for money or private data through a second channel. Use a code word with family and key staff, and let anyone pause a payment to check. Limit public video and voice clips of key people.

Next step

Book a free workflow consult and bring the steps your office follows before a payment goes out. We will mark the points where a call back belongs and where a person always approves. We work from the Kansas City metro and meet owners face to face or by video call.