How to Spot a Deepfake

Why your eyes aren’t enough anymore

A few years ago, deepfakes gave themselves away with extra fingers, melting ears and garbled text. Today’s image, video and voice generators have fixed most of those flaws. In a bank study, 89% of people believed they could identify a deepfake scam — only 42% actually did when tested (Vectra AI).

So treat visual clues as a bonus, not a test. The strongest defence is verification. Try it yourself: our game Spot the Deepfake mixes real photos with AI images — most people do worse than they expect.

Step 1: Check the source before anything else

• Who posted it first? An unknown account, a brand-new page, or a forwarded message with no origin is a warning sign.
• Is it reported anywhere reliable? A genuinely shocking video of a public figure will be covered by major news outlets within hours. If it only exists on social media, be very careful.
• Does the timing feel engineered? Fakes often appear just before an election, a market move or a crisis — when there is no time to check.

Step 2: Use free tools

• Reverse image search — Google Lens or TinEye can show where an image appeared before, and whether it was altered from an older photo.
• Content Credentials — some cameras, apps and AI generators now attach tamper-evident information about how an image was made. If an image carries it, you can inspect its history.
• Be careful with “AI detectors” — online deepfake detectors are often wrong in both directions. Never treat one as proof.

Step 3: Look for clues (they still catch lazy fakes)

• Text and signs: words that almost make sense, or letters that melt together.
• Hands, ears, teeth, jewellery: extra or merged fingers, mismatched earrings, too many teeth.
• Lighting and shadows: light that comes from different directions on different objects.
• Backgrounds: repeating faces in crowds, warped lines, objects that blend into each other.
• Too perfect: flawless skin, cinematic lighting and a composition that tells the story a little too well.
• In video: blurring at the edge of the face, lips slightly out of sync, unnatural blinking, or a face that glitches when the head turns.
• In audio: flat emotion, odd pauses, no breathing sounds, or a voice that won’t answer unexpected questions naturally.

Step 4: For calls and video calls — verify, don’t analyse

If a call, voice message or video call asks you to send money, share a code or keep a secret, don’t try to judge whether it looks real. Hang up and call the person back on a number you already know, or check through a second channel. Families can agree on a safe word for emergencies. Read our full guide to AI voice-cloning scams.

If you find a deepfake

• Don’t share it — even to mock it. Sharing spreads it.
• Report it to the platform where you found it.
• If it shows you or someone you know in a sexual or harmful way, save evidence and report it to the platform and to the police. In many countries this is a crime.