How to Recognize Deepfakes: Video, Audio, Image
A Hong Kong finance worker transferred $25 million after a video call with what appeared to be his company's CFO and several colleagues — all of whom were deepfakes. In a separate incident, an AI-cloned voice of a CEO convinced a bank manager to authorize a $35 million transfer. These aren't science fiction. They happened in 2024-2025, and the technology has only improved since.
Deepfake generation tools that required specialized knowledge and expensive hardware three years ago now run on consumer laptops and smartphones. The number of deepfake videos online doubled every six months between 2023 and 2025, and detection is becoming an arms race between creators and identifiers.
Received a suspicious video or voice message? Paste the link into our free scanner →
Video Deepfake Detection
Current deepfake video technology is impressive but not perfect. Look for these artifacts:
Facial boundary issues: Where the face meets the hair, ears, and neck, deepfakes often show blurring, warping, or inconsistent lighting. The face may appear slightly "pasted on" — especially visible when the person turns their head to the side.
Unnatural blinking: Earlier deepfakes had issues with blinking patterns — either too frequent, too infrequent, or with both eyes not perfectly synchronized. While 2026 models have improved, extended video sequences still occasionally reveal abnormal blink patterns.
Inconsistent lighting: The lighting on the face may not match the lighting in the rest of the scene. Shadows may fall in the wrong direction, or skin reflections may not match the ambient lighting.
Teeth and mouth interiors: Teeth in deepfakes sometimes blur, shift, or appear inconsistently shaped between frames. The interior of the mouth during speech may show artifacts or lack the expected detail.
Background distortion: The area immediately around the subject's head may warp or shimmer slightly, especially during movement. This is caused by the AI model struggling to maintain consistency between the generated face and the original background.
Think it might be a scam?
Paste it here for a free, instant verdict.
Free · No signup required · Cmd+Enter to scan
Temporal inconsistencies: Watch the video at 0.25x speed. Deepfakes may show frame-to-frame flickering, momentary distortions, or skin texture that changes subtly between frames.
Audio Deepfake Detection
Voice cloning has progressed faster than video deepfakes and requires far less source material — some services need only 3 seconds of audio to produce a convincing clone.
Unnatural prosody: AI-generated speech may have slightly off rhythm, unusual pauses, or inconsistent emphasis patterns. It often sounds "too perfect" — lacking the small imperfections (breathing, filler words, self-corrections) that characterize natural speech.
Background audio inconsistencies: Cloned voices often have artificially clean backgrounds or backgrounds that don't match the claimed environment. If someone claims to be calling from an airport but there's zero ambient noise, that's suspicious.
Emotional flatness: While AI can replicate the basic tone of a voice, it often struggles with the subtle emotional variations present in genuine speech — the slight tremor when discussing something personal, the natural warmth when greeting a family member.
Phone call test: If you receive a suspicious call claiming to be from a family member or colleague, ask a question only the real person would know. Or better yet, hang up and call them back at their known number.
IsThisAScam's 6-layer detection system helps identify scam messages that may reference or link to deepfake content, such as fake celebrity endorsements, fraudulent video testimonials, and voice-cloned voicemail messages.
Image Deepfake Detection
AI-generated images (from tools like Midjourney, DALL-E, and Stable Diffusion) have become nearly indistinguishable from photographs. Current detection points:
Hands and fingers: AI still frequently produces hands with wrong finger counts, impossible joint angles, or blurred/fused fingers. Always check the hands in a suspected AI image.
Text in images: AI-generated images struggle with text — letters may be malformed, words may be gibberish, and text may appear blurred or inconsistent.
Asymmetry issues: Earrings that don't match, glasses frames that are slightly different shapes, or clothing patterns that don't mirror correctly across both sides.
Background anomalies: Look for objects that don't make physical sense — chairs with impossible geometry, buildings with inconsistent perspectives, or trees that fade into blurred masses.
Verification Strategies
Beyond visual and audio detection, use these verification approaches:
- Reverse image search: Use Google Images, TinEye, or Yandex to check if an image appears elsewhere or has been AI-generated
- Metadata inspection: Genuine photos contain EXIF data (camera model, date, location). AI images typically have no EXIF data or metadata from the generating tool
- AI detection tools: Services like Hive Moderation, Sensity, and Microsoft's Video Authenticator can analyze media for AI-generated characteristics
- Source verification: Check whether the content was posted by an official, verified account. Can the original source be independently confirmed?
- Cross-reference: If a video shows a public figure saying something extraordinary, check if any legitimate news outlet is reporting it
How Deepfakes Are Used in Scams
- CEO fraud: Deepfake video calls impersonating executives to authorize wire transfers
- Romance scams: AI-generated profile photos and video calls to maintain catfishing personas
- Celebrity endorsements: Fake videos of celebrities promoting cryptocurrency or investment scams
- Family emergency scams: Voice clones of family members calling to request urgent money transfers
- Fake news: Fabricated videos of public figures making inflammatory statements
For related protection strategies, see our guides on dating app scams and identity theft prevention.
Received something suspicious? Check it now for free →