ONE MAN “REVERSE ENGINEERED TINDER.” THE INTERNET REVERSE ENGINEERED WOMEN.
A viral photo-scoring video has people drawing conclusions about what women want. Tinder uses actual user behavior to match people. So whose model produced the scores everyone is debating?
AI NEWS PAGE · PUBLISHED October 21, 2026
On October 20, Ayo O posted a video claiming to have reverse-engineered Tinder. The clip shows photos being processed into an embedding. The next day, menswear writer Derek Guy observed that heavier men did not seem to score lower and suggested the scores reflected women's preferences. A demo number had turned into a claim about an entire dating population.
There is a missing step. A score produced in a Tinder-themed demonstration is not necessarily a score produced by Tinder. To establish that, we would need to know where the scoring model came from, what labels trained or validated it, whether its outputs were compared with real Tinder behavior, and whether the exercise measured photo quality, individual attractiveness or the probability of a match. Those are different things.
Tinder's own September 2026 matching explanation says its recommendations account for activity, proximity, profile details, Likes and Nopes, and anonymized similarities between photos people have liked. It says the old Elo score is no longer used. Nothing in that description supports interpreting an unexplained single photo score as a universal referendum by women on male body size.
Tinder really does use AI to rank photos, which makes the viral claim more plausible at a glance. But its engineering description is of Smart Photos: comparing two photos of the SAME user and predicting which will perform better, using training examples drawn from actual impressions and likes. Selecting the best picture of one person is not ranking different people by desirability. Tinder also documents AI photo suggestions that run partly on the device, so saying no Tinder algorithm could live in the mobile app would itself go too far.
The supported conclusion is narrower and stranger than the viral takeaway: people are arguing over supposed Tinder findings without a verified bridge between the demonstrated scores and Tinder's real recommendation outcomes. The claim that heavier men do not score lower on this demonstration may describe the demonstration accurately. It does not, by itself, reveal how Tinder or heterosexual women rank heavier men.
WHAT WE DON'T KNOW
- Whether Ayo O obtained any actual Tinder server response, proprietary model output or measured interaction data; the full methodology was not independently verifiable from accessible public material.
- What model produced the displayed scores, and whether it was trained on real Tinder outcomes, synthetic preferences, or general image features.
- Whether the compared people and photos were controlled for age, presentation, photo quality, sampling and who was shown each image.
- Whether the author has published source code, a dataset, or a reproducible comparison with live Tinder recommendations.
SOURCES
- PUBLIC TRACE · ORIGINATING SIGNALTHE SCORE IS VIRAL. WHOSE SCORE IS IT?Viral Tinder reverse-engineering video and heavier-men score discussionDerek Guy quoting Ayo O on X
- PRIMARY · CORROBORATIONTINDER SAYS LIKES, NOPES, PHOTOS AND ACTIVITY MATTER — NOT AN OLD ELO SCORE.Powering Tinder — The Method Behind Our Matching (updated September 1, 2026)Tinder Help Center
- PRIMARY · CONTRADICTIONITS OWN AI COMPARES TWO PHOTOS OF THE SAME PERSON.How Tinder Uses AI to Rank Profile PhotosTinder Engineering
- PRIMARY · BACKGROUNDSOME TINDER PHOTO AI DOES RUN ON-DEVICE.Photo-Based SuggestionsTinder Help Center