Straight BS: an AI 'red flag detector' cannot read a date's character from faces, laughs and phone glances

Facebook · Oct 2026

The claim'AI Red Flag Detector: Real-Time Dating Behavior Analysis.' A restaurant date with face boxes and labels like 'Green Flag: Genuine Laugh 94%', 'Red Flag: Checked phone mid-sentence 98%', 'Overtalking Detected', 'MAJOR Red Flag: Still talking about his ex 99%', and a 'vibe score' dropping from 80 to 10.

This is a sketch with a code editor pasted underneath. The labels on screen are not things a camera can measure. 'Still talking about his ex, 99 percent confidence' requires understanding the conversation, not a face. 'Genuine laugh 94 percent' and a single 'vibe score' are judgments about character and intent, presented as sensor readings. The underlying science is the problem. Reading inner emotions from facial expressions is heavily disputed: a large 2019 review of the research found no reliable one-to-one link between expressions and emotional states, and expression varies by culture and person. The EU AI Act goes further and bans emotion-recognition systems in workplaces and schools, while requiring disclosure elsewhere. A system that turns a frown or a phone glance into 'red flag, 98 percent' is putting false precision on guesswork. The video itself is staged: polished stock-style footage of actors, overlays that change in sync with the edit, and the same 10-second template the page uses for construction, retail and traffic demos. The caption pivots to 'serious' uses like scoring sales calls and job interviews, which is exactly where emotion inference is most risky and, in the EU, restricted. If you want better sales calls, review the transcript for talk time and questions asked. That is measurable. 'Vibe' is not.

What holds up

Reel read frame by frame: labels include 'Green Flag: Genuine Laugh 94%', 'Red Flag: Checked phone mid-sentence 98%', 'Overtalking Detected', 'MAJOR Red Flag: Still talking about his ex 99%', 'Boundary Setting', and a vibe score falling 80 to 45 to 10.
Several labels depend on what is being said, which a face-and-gesture model cannot know; the reel has no audio analysis shown and the soundtrack is music only.
EU AI Act (Article 5) prohibits emotion-recognition systems in workplaces and education; scientific reviews cited by EPIC and EDRi report no reliable link between facial expressions and inner emotional states.
Same 10.08-second headline-overlay-code template as the page's other demo reels.

What doesn’t

Character judgments presented as percentage-confidence readings.
Staged footage with overlays timed to the edit.
Caption steers toward sales-call and interview scoring, a restricted and high-risk use.

The catch

Emotion and intent cannot be read reliably from faces, and 'talking about his ex' cannot be read from video at all. The percentages are decoration.

How to actually do it

For sales coaching, record calls (with consent) and review transcripts for talk-to-listen ratio, questions asked and next steps agreed.
Do not use any tool that scores job candidates or employees on facial 'emotion'; it is unreliable and banned in EU workplaces.
For dating, trust what someone says and does over time, not an overlay.

A camera can see that someone looked at a phone. It cannot see honesty, warmth, or whether they mentioned an ex, and the science on reading emotion from faces does not support these confidence numbers.

Confidence: High
Posted by: a computer-vision software page (about 91K followers) that posts 10-second AI demo reels; reel sent to Buddy Sun Oct 11, 2026, ~7:49 AM

See the original claim →

We test hype for free. We build the real thing for a living.

Thirty minutes, no pitch — and you'll leave with something useful either way.

Book a call with Todd or start with the free Business Checkup →

The Verdict Weekly

Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.