Marcus, 34

Detected — Earring shape and position differ between photos 2 and 3, despite same outfit and lighting context.

Why it matters — Geometry inconsistencies across photos of the same person are a common signature of AI face generation or face-swapping.

Detected — Background blur in photo 1 shows smearing patterns typical of diffusion-model generation rather than optical depth-of-field.

Why it matters — Real cameras blur backgrounds with smooth optical falloff. Diffusion smearing suggests the photo was generated, not taken.

Detected — No detectable AI-generation or splice indicators in photo 3.

Why it matters — One clean photo isn't proof of identity, but it raises the value of insisting on a live video.

Detected — Bio uses generic phrasing matching 14 known scam-script templates (e.g. "looking for someone real in a fake world", "lost my partner two years ago").

Why it matters — Bios that pattern-match known scam scripts are reused across many fake profiles, often by the same operator.

Detected — Stated job title ("Offshore engineer, Aberdeen") is not present in any LinkedIn footprint for the named person.

Why it matters — Roles like offshore engineer, military deployed, or oil-rig contractor are commonly used to explain away the inability to meet or video call.

Detected — Urgency around moving the conversation to WhatsApp within 48 hours of matching.

Why it matters — Scammers move off-platform quickly to escape moderation and screenshot review by the dating app.

Detected — Financial topic (a stuck wire transfer, mention of crypto investing) introduced on day 4.

Why it matters — Genuine connections rarely surface money in the first week. Early financial framing is a precursor to the ask.

Detected — Live video declined three times with shifting reasons ("bad signal offshore", "camera broken", "company policy").

Why it matters — Pattern refusal of video is the single strongest behavioural indicator of catfishing in our dataset.