AI can now put words in anyone's mouth. Here's what to look for — even if you're not a tech expert.
The word "deepfake" combines "deep learning" (the AI technique involved) with "fake." The technology works by training AI on thousands of images or video clips of a real person, then generating new footage of that person in any situation the creator invents.
You don't need to become a forensic analyst. Most deepfakes that you'll encounter in everyday life — shared on social media, sent via messaging apps — still have telltale signs you can catch with the naked eye and a little practice.
AI struggles most with the small details of real human faces, particularly at the edges and in areas that change rapidly with emotion. Here's a map of where to look:
When you see a video that seems shocking or too good to be true, run through these steps before you share or believe it.
Where did this video originate? A verified news account with a long track record is very different from a three-week-old account with no history. Pause on anonymous or new accounts sharing shocking content.
If a video shows something genuinely newsworthy — a politician confessing, a celebrity doing something scandalous — major news organizations would be all over it. A Google search for the person's name plus the claim takes 30 seconds and can save you from spreading false information.
Watch in slow motion if possible. Focus on the jawline, hair edges, and teeth. Look at the background for any text. One suspicious detail alone may not mean much, but two or three together are a strong signal.
Take a screenshot from the video and run it through Google Images (images.google.com). This often reveals whether the footage is borrowed from a legitimate older video and edited, or shows it appearing in other suspicious contexts.
Several AI tools can flag deepfake probability. Microsoft Video Authenticator, Deepware Scanner, and Intel's FakeCatcher are available to the public. None are perfect, but they add one more data point to your assessment.
AI assistants can be useful partners in thinking through suspicious content. Try these prompts:
AI assistants won't watch the video for you — but they can help you build a verification checklist, suggest fact-checking resources, and explain what detection tools exist and how they work.
While video deepfakes get the most attention, voice cloning has become just as sophisticated and is used just as often. AI can now clone someone's voice from as little as a few seconds of audio.
The key defense against audio deepfakes is establishing a verification word or phrase with people you care about — something only you two would know that you can use as a "prove it's you" check in suspicious situations.