What visual clues reveal an AI-generated dating profile photo?

Look for unnatural symmetry, blurred or melting edges behind ears and hair, inconsistent lighting where the face is lit differently than the background, misplaced pupils or mismatched eye reflections, and skin that looks waxy or airbrushed rather than natural.

How do I test for a deepfake during a video call?

Ask the person to touch their nose or cover their mouth with their hand. Deepfakes struggle when a hand occludes the face — you will see blurring, pixelation, or misalignment between the hand and face. Also ask them to hold a specific object and turn around slowly.

Why do automated liveness checkers fail against modern deepfakes?

Advanced deepfakes are trained to pass standard liveness tests such as blinking or simple head movements. Human analysts use unpredictable, real-time requests — such as holding a specific object to the camera or responding to a sudden lighting change — that AI cannot reliably replicate on the fly.

Modern Threat Detection

How to Detect AI‑Generated Dating Profiles and Deepfake Video Calls

By Michael Muinov, Founder and Lead Investigator · Active casework since 1999 · Updated 17 May 2026

The dating landscape has been flooded by AI‑generated faces and video calls that look incredibly real. Here’s how human experts – not just algorithms – can tell the difference.

Quick answer

How can you detect AI-generated dating profiles and deepfake video calls?

Detection is split between profile-level and live-video checks. For profiles: an AI face usually fails on small details — misaligned earrings, irregular pupils, text in the background that resolves to gibberish, and hands rendered with wrong finger counts or impossible joints. Photo sets are too consistent in lighting, age, and styling because they were all generated in one session. For deepfake video calls: ask for sudden head turns past 90 degrees (real-time face-swap degrades at extreme angles), have her place a hand flat against her cheek and rotate it (the algorithm cannot track occlusion well), or request she pick up and read a random book from her room (combines unscripted speech, varied lighting, and a moving object the model was not trained against).

Important limit: these tests work today but degrade as AI tools improve. The detection arms race favours the attacker over time. Cross-referencing the claimed identity against state records is the one check that AI cannot defeat — a synthetic person does not exist in birth registries, employment databases, or residency records, regardless of how convincing the photo or video output is.

Signs of an AI‑Generated Face

Unnatural symmetry: GAN‑generated faces are often unnaturally symmetrical.
Background artefacts: Blurred or melting edges behind ears and hair.
Inconsistent lighting: Shadows don’t match the scene.
Strange eyes/pupils: AI often misplaces pupils or creates mismatched eye reflections.

Our AI Image & Deepfake Review manually inspects these clues.

Deepfake Video Call Indicators

Lip‑sync issues: Audio lags behind mouth movements.
Blinking irregularities: AI sometimes forgets to blink naturally.
Warping: The face may distort when moving quickly.
Low quality: Scammers intentionally degrade video to hide flaws.

If a video call feels off, trust your instinct and verify with a professional Catfish Investigation.

The Human Edge in Liveness Detection

Automated “liveness” checkers are fooled by advanced deepfakes. Our analysts use heuristic tests: requesting a real‑time gesture, a specific object held to the camera, or a sudden change in lighting. No AI can replicate these on‑the‑fly requests reliably. See also AI in Russian/Ukrainian Romance Scams and our Russia / Ukraine hubs.

Submit for AI Review

How to Detect AI‑Generated Dating Profiles and Deepfakes

Examine profile photos for AI-typical artifacts. Zoom in on eyes – AI often generates two different reflections. Look at ears (asymmetrical or deformed), teeth (blurry or too uniform), and hair strands (merge unnaturally into skin).
Use AI detection tools on images. Upload suspicious photos to free detectors: Hugging Face’s “AI or Not” (best for realistic faces), Sensity, or Illuminarty. No tool is 100% accurate, but a high confidence score is a strong red flag.
Analyze skin texture and lighting inconsistencies. Real human skin has pores, fine lines, and uneven tones. AI faces look “waxy” or airbrushed. Also check lighting: if the face is lit from one angle but the background from another, it is a composite.
Check for deepfake video indicators. On a video call, ask the person to touch their nose or cover their mouth. Deepfakes struggle with hands occluding the face – you will see blurring, pixelation, or hand-face misalignment.
Listen for voice synthesis artifacts. If they send voice notes, listen for robotic intonation, unnatural pacing, lack of ambient noise, or missing breaths. Ask them to say a number sequence – AI voices often mispronounce digits or sound flat.
Check for duplicate metadata across multiple profiles. Use reverse image search on the photo. If the same AI-generated face appears on multiple dating profiles with different names, it is a fake. Also, AI images often have no EXIF data (no camera, no date).
Demand a live action that cannot be deepfaked in real time. Ask the person to hold a specific object (e.g., a spoon, a book) and turn around slowly while on a live video call. Real‑time deepfake generation cannot produce consistent object interaction or 360‑degree head rotation.