Every guy on Tinder has had the same thought at 1am: is the app hiding me? You swipe, you wait, nothing happens, and it feels like there's a machine somewhere deciding you're not worth showing to anyone.

This week, a video went viral claiming to show what that machine actually looks at. It's 79 seconds long, it's racked up 1.5 million views, and it's worth your time. Here's the source:

A
Ayo O
@TheDonAyo
"I reverse engineered Tinder"
Video, 79 seconds · Posted Oct 20, 2026 · 1.5M views
Watch it on X →

Ayo is the co-founder of a0.dev (Y Combinator, W25 batch). Full credit to him for the digging. Everything below in the "what he found" section is his claim, presented the way he presents it in the video. After that, I'll tell you what I'd actually do with it, and where I'd take it with a grain of salt.

What he found, minute by minute

0:00The setup

Ayo opens by saying he reverse engineered Tinder to figure out how its algorithm works and how it gets 8.5 million people to pay every month. For that subscriber number, he cites Match Group's Q2 2026 shareholder letter.

0:05Two numbers: what you want, and how much you're wanted

According to the video, Tinder calculates two things about you. The first is your type. Every photo you swipe right on gets turned into an embedding (basically a list of numbers a computer uses to describe what's in an image), and Tinder averages those together into a picture of what you like.

It doesn't just watch what you swipe on, either. Ayo shows code for something called media_view_details_card, which logs a photo_id, a view_count and a total_duration_ms. In plain English: how long you look at each photo.

0:15RSRR, the number that matters

The second thing is how much you're wanted. Ayo shows decompiled code from Tinder's Android app (version 13.0.0, from 2022) referencing RSRR, or "right swipe received ratio." It's exactly what it sounds like: of everyone who sees your profile, how many swipe right.

0:23An AI model on your phone scores your photos

Here's where it gets interesting. According to the video, there's a TensorFlow Lite model built into the Tinder app itself that predicts RSRR for each individual photo. In other words, it guesses how many right swipes a given photo of you will get.

Ayo says this powers Tinder's Photo Selector feature, which scans up to 20,000 photos from your camera roll, only from the last two years, matches them to your face (using a similarity threshold of 0.7), and picks the best 27.

0:35The privacy question

Tinder's iOS copy for this feature says everything stays on your device and that Tinder doesn't collect your selfie, your biometric data or the photos in your library.

Ayo then shows a data payload called PhotoSelectorPhotosCollectionDetails, sent to POST /v3/publish/app/binary, which includes fields named rsrr_scores, explicit_scores, text_scores, underage_scores and similarity_scores. His claim: even if the photos themselves never leave your phone, Tinder still collects the scores for each photo, including how explicit it is and how young you look.

0:42He ran 800+ test photos through the model

Ayo says he pulled the model out of the app and ran it on more than 800 synthetic (AI-generated) photos, changing one thing at a time to see how the predicted score moved. These are his results:

What changed in the photo Score change How consistent
Mirror selfieAbout 3.5x16 of 16 went up
Men, outdoors+49%8 of 8 went up
Men, shirtless+23%25 of 25 went up
Men, sunglasses+15%8 of 8 went up
Men, heavierNo change23 up, 15 down
Men, 15 years olderNo change19 up, 19 down
Women, low neckline+63%15 of 15 went up
Women, outdoorsNo change4 up, 4 down
Women, sunglasses-26%8 of 8 went down
Women, 15 years older-16%28 of 38 went down
Women, heavier-54%6 of 8 went down
Darker skin-15%12 of 12 scenes went down

Source: Ayo O's video on X. These are his numbers from AI-generated test photos run through a model he extracted from the app. They are not results from the live Tinder app, and several rows are based on small samples (as few as 8 photos).

1:03The Super Like upsell

The last part of the video shows text from Tinder's iOS app: "You just liked a popular user. Send a Super Like to %@ instead for a better chance to match!" (the %@ is where the person's name goes). He also shows an analytics event named upsell_superlikes_high_rsrr, tagged as revenue.

His claim: if your profile is popular, people who swipe right on you get nudged toward paying for a Super Like, and Tinder tracks how much money your popularity brings in.

What this actually means for your profile

First, the most important thing to understand. That score is a model's prediction of how many right swipes a photo will get. It is not the person on the other side of the screen. Real humans still look at your profile and decide. The model is guessing what they'll do. Your job is to give the humans something worth swiping on, and the rest tends to follow.

With that in mind, here's how I'd use this.

Get at least one good outdoor shot

For men, outdoors was the biggest clean win in his test: +49%, and every single test photo went up. That lines up with what this blog has been saying all along. Natural light, some background that says you have a life outside your apartment. If your whole profile is indoors, fix that first.

A good mirror selfie is fine

Guys have been told for years that mirror selfies are cringe. In Ayo's test, the model loved them. I wouldn't build a profile out of them, but one clean mirror shot (good lighting, tidy room, mirror actually wiped) is not the dealbreaker people make it out to be.

Sunglasses are OK, just not in every photo

For men, sunglasses scored +15% in his test. I still wouldn't lead with them, because your eyes are how someone decides whether you seem trustworthy. One sunglasses shot in the mix is fine. Five is hiding.

Shirtless is a judgment call

Yes, shirtless photos scored +23% for men. But a right swipe isn't the goal. A match you actually want to talk to is the goal. A shirtless photo tends to attract a certain kind of attention, and it may not be the kind you're looking for. If it's natural (beach, surfing, a hike) and it fits who you are, fine. A bathroom flex, probably not.

Recent photos matter more than you think

According to the video, the Photo Selector only looks at the last two years of your camera roll. Whatever the reason, that's a good rule for you too. If your best photos are from three years and fifteen pounds ago, she will notice on the first date. Recent photos build trust. Old photos burn it.

First impressions compound

If RSRR works the way Ayo describes, it's a ratio: right swipes out of everyone who sees you. That means a weak lead photo doesn't just cost you one swipe. It drags down the number that (at least in theory) shapes how you're shown going forward. Your first photo is not the place to experiment. Put your strongest, clearest shot there.

Don't panic-buy Super Likes

The Super Like upsell he found is aimed at the people swiping on popular profiles. If the app is nudging you toward paying to stand out, that's a business decision, not a sign you're doing something wrong. A better profile does more for you than a stack of Super Likes ever will.

The uncomfortable parts

Some of the numbers in this video are hard to look at, and they deserve to be said plainly.

In Ayo's test, darker skin scored 15% lower, and it went down in all 12 scenes he tried. For women, looking 15 years older scored 16% lower, and looking heavier scored 54% lower. For men, being older or heavier made no consistent difference at all.

If those patterns hold up, that's a model reflecting (and potentially amplifying) bias, not a verdict on anyone's worth. And it's not something you should try to "fix" about yourself. Nobody should change their skin tone, age or body to please an algorithm. What I'd take from it is simpler: the score is not the person, and plenty of real humans swipe very differently from a model's average.

The privacy piece matters too. If Tinder says photos stay on your device but per-photo scores (including how explicit a photo is and how young you look) still get sent back, that's something users deserve a clear answer on. Based on what's in the video, it rests on one payload snippet, so treat it as a serious question rather than a proven fact. If it bothers you, you don't have to use Photo Selector. You can pick your own photos.

Take it with a grain of salt

This is a fascinating video, but it's one person's investigation, and some important things can't be confirmed from the outside:

  • We can't verify the code is what's running today. Some of what he shows comes from a 2022 version of the Android app. Tinder may have changed things since.
  • The privacy claim rests on a payload snippet. It shows field names, not the full picture of what's sent, when, or how it's used.
  • The photo results come from his own setup. AI-generated photos, run through a model he pulled out of the app. That's not the same as testing real profiles on the live app.
  • Some samples are small. Several rows are based on just 8 photos. That's enough to be interesting, not enough to be the final word.
  • Tinder hasn't confirmed any of this. Everything here is Ayo's claim.

So don't rebuild your whole profile around a table of percentages. Use it as a nudge in the same direction good profile advice already points: clear, recent, well-lit photos of you actually living your life, with your best one up front.

Want a human to look at your photos?

An algorithm guesses. I actually look. If you want to know how your current profile reads to a real person, the $19 Profile Grade gets you your score and your biggest issues within 24 hours. If you want the whole thing redone (bio, prompts, photo order and opening strategy), that's the $229 Profile Rewrite. Either way, reviewed by me, no AI grading.