Deepfake technology has moved from research labs to consumer apps. The 2024 World Economic Forum Global Risks Report ranked AI-generated misinformation among the top short-term risks, citing the speed at which both video and voice clones have become accessible. The good news: detection has improved too, and the obvious tells — though shrinking — still catch most low-effort fakes.
This guide covers what to look for visually, what to listen for in audio, and which free tools are worth your time.
Video deepfake — visual signs
Lip sync drift
Mouth shapes that don't quite match consonants — especially 'p', 'b', 'm' sounds, where lips visibly close on real video.
Unnatural blinking
Either too few blinks (older face-swap models) or perfectly metronomic blinks. Real blinking is irregular.
Lighting that doesn't match
Face lit from a different angle than the room behind it. A common artifact of face-swap pipelines that don't relight the inserted face.
Edges around the hairline and jaw
Faint blur, color shift, or warping where the synthetic face meets the original head — visible if you pause and zoom.
Teeth and eye reflections
Teeth that look like a single fused row; eyes whose reflections don't match the actual lighting environment.
Voice deepfake — audio cues
Flat affect
Voice clones often miss the micro-pauses, breaths, and pitch variation of real speech, producing an oddly smooth delivery.
Background room mismatch
The voice sounds 'dry' (studio-clean) while the video shows an outdoor or echoey environment.
Inconsistent breathing
Real speakers breathe between phrases. Many voice clones omit or misplace breath sounds.
Detection tools worth trying
Intel FakeCatcher
Detects blood-flow signals in pixels (PPG) that real video has and deepfakes typically don't reproduce.
Sensity AI
Commercial deepfake monitoring used by media organizations and platforms. Free demo available for individual checks.
Microsoft Video Authenticator
Released alongside the 2020 US election, focused on detecting subtle fading and grayscale artifacts in face-swap video.
Deepware Scanner
Free, mobile-friendly tool that scans video uploads or URLs for deepfake indicators.
AI or Not (audio)
Focused on voice-clone detection. Useful for short suspicious audio clips circulating on messaging apps.
What to do if you suspect a deepfake
Report it to the hosting platform first — every major platform (YouTube, TikTok, X, Meta, WhatsApp) has a synthetic-media policy and a reporting flow. For non-consensual intimate deepfakes, the StopNCII.org pilot accepts reports and works with platforms to remove content. For political or election-related deepfakes, IFCN-signatory fact-checkers in your country are the right second stop.
And: don't reshare to debunk. Re-uploading a deepfake to point out it's fake amplifies it. Describe, link to the fact-check, and let the original lose reach.
See deepfakes already in the wild
FAXTR's AI Fakes feed catalogs confirmed AI-generated video and voice misinformation tracked by fact-checkers worldwide.