Earlier this week as I was logging entry number 3000 into my Tinder statistics I found myself going a bit up in levels of abstraction. The type of work I’m doing with the Tinder statistics is very detailed and looks into specific workings of the algorithms and systems that Tinder uses in order to make the app work. One of the recent developments is a little piece of legislation called the Global Data Protection Regulation (GDPR for short) that gives people the right to request the data that companies like Tinder and Facebook have collected on them. Thanks to this, a gentleman was able to request all the data Tinder had on him, and as a side-effect we figured out what data they track on every user.
The classical personal information such as name, email, age, and age bracket selected is not very interesting, what is interesting is that they track among others:
- Your swipes (Right and left)
- How many people who swipe left and right on you
- Messages sent and received
- Profile completeness
The reason why I found this interesting, is that if we start to reverse engineeer how Tinder works, those tracked data are the perfect tool for it. Odds are that Tinder also stores other “match related” things such as time spent looking at a profile before swiping, how many of the people you match with you actually have a conversation with, we know they track how often you get reported, and so on. This becomes important later.
If we think of Tinder and other apps as a bar, we all know what a bar has to do in order to do well (sportsbars and cigarbars excluded), and that is get a lot of hot girls in the door. If your bar has hot girls, the men will follow and buy them drinks. The girls are attracted to “how cool the bar is” the guys are attracted to “how hot the girls who think the bar is cool” are. From this perspective, we can thus outline the 3 major success criteria for Tinder:
A) Keep the girls happy
B) Maximize your user base
C) Keep the men around
Just like a bars revenue is based on a mixture of cover charges and drink sales, Tinder’s revenue is based on advertising, subscriptions to Plus and Gold, in addition to sales of consumables such as superlikes and boost. In the night club analogy, sending a girl a super-like is the equivalent of sending a girl a drink, a boost is equivalent to the club promoter shaking your hand and taking you to your reserved table with bottle service.
However, the most important man in the bar is the doorman, you see, his job is to:
A) Maintain a good mix of men and women
B) Keep the creeps out
C) Get rid of any troublemakers.
A bar without a solid doorman rapidly becomes a very unpleasant place to be. In the same way, Tinder has to walk a fine line between maintaining their female user-base, maximizing their user-base in general and maximizing the revenue from subscriptions and consumables (which I assume are mostly bought by men). How does Tinder do this?
The Algorithm and Pattern Recognition
Note that the following is my educated guess at how Tinder’s algorithm works, to begin with I think it’s a two, potential 3 component concept that is composed of your behavior, the other people on Tinder’s behavior and general pattern recognition based in a software that is constantly looking for “Red Flags”.
We can start with your behavior, I suspect that Tinder’s analysis starts with the following behaviors:
- Swiping Pattern
- Matching Pattern
- Open Pattern
- Response Pattern
- Conversation Pattern
- Unmatches and Reports
Your swiping pattern comes down to
A) How often do you swipe right or left
B) How quickly do you do it?
It used to be that you could just swipe right on everyone and it would work just fine, however this behavior demonstrates that you are not very discerning and that you’re hunting for matches rather than connections. In essence, you are going into the club, grabbing the DJ’s mic, tearing your pants off on the stage and screaming “Does ANYONE WANT TO FUCK??” Tinder doesn’t like that type of behavior. Secondly, the behavior can easily be mistaken for automated behavior and Tinder is not too fond of the various bots selling premium snaps, instagram and porn sites. Hell in some cases alternative dating sites. On the other hand, a person who purely swipes left is punished as well, because they are seen as not really wanting to match.
The Second is your matching pattern, how often do you match with people you swipe right on. What this comes down to is a simple question “Are you playing within your league”, we all date aspirationally, meaning we try to get someone who is 1 – 3 points above us in SMV, but are you going way out of your league? I think the reason why Tinder punishes this is that it’s often associated with the bots, but also because they want people to match.
The third thing is your open pattern, do you open most of your matches or do you leave most of them hanging? What Tinder is trying to encourage is that people should match, start conversations, go on dates and whatever else happens afterwards. If you are matching but hardly ever contacting your matches, then Tinder thinks you are simply match hunting for validation.
The Fourth thing is response pattern, how do your matches respond when you message them? Do they respond or do they leave you hanging, while I doubt that Tinder parses all the messages sent through the service and estimates a response, I do think they view a large amount of non-responders as you either being unable to come up with a good opening message, or you saying something that turned the other person off.
The fifth thing is conversation pattern, how often do you have what Tinder deems as a good conversation with your matches? Are all just you sending a few openers and getting no response, or do you have longer conversations that Tinder can interpret as being positive but “Decided we didn’t click well”?
The sixth and last variable is unmatching and being reported. Generally someone unmatching you right after an opener or reporting you is probably interpreted negatively, while being reported probably also makes them watch your account harder.
Why do I think these are the main things Tinder is doing with your data? Because it’s exactly what I would do if I was building a service that aimed to give people the chance to accurately match with other people. We have to keep in mind that Tinder’s business proposition to its users is “Match with people you are attracted to who you can build a relationship with” and yes I know it’s more or less a hook-up app, but Tinder wants the matches to be as accurate as possible.
How the Girls Respond
To go back to my club analogy, if a regular 6/10 dude starts making problems in the club, making girls cry, annoying the staff and so on, he will be kicked out, and probably have his ass kicked for good measure. However, if a 10/10 dude, probably a celebrity, or someone with a ton of cash starts making trouble in the club, he is given a rather wider berth, simply because he is a guy who gives the club a lot of income. All of us put up with some extra shit from great customers, whether that’s the guy who comes into the cafe every day at 11 am but orders off the menu or the client who calls you up at 11 PM on a Saturday to put an order through that nets you a huge commission.
The reason why I suspect Tinder implemented this is that we all know that there has to be two sets of rules, if Tinder bans a bunch of highly attractive men because they are behaving like bots or get reported a lot, that makes Tinder less attractive to the women. Secondly, on the girl side, if Tinder banned all the “bot-like” women who are using it to market their instagrams, by having a profile with 3 – 4 risque pictures of themselves and only having their insta-handle as their bio, Tinder would have a lot less hot girls. The club doesn’t give a shit if the 10/10s are acting like total cunts as long as men are buying them drinks all night.
So you can engage in most, if not all of the bad behaviors I outlined above, provided that many of the opposing sex swipe right on you and open. That seems to offer a certain degree of balance. I think this is simply a case of the algo being a weight of the former and the latter. For instance, lets say you score a -8 for your behavior according to points 1 – 6 in the earlier section, however you keep getting right-swiped by accounts Tinder views as highly desirable, you get away with it.
How Tinder Responds
Unlike the club, Tinder doesn’t have to kick out people who misbehave, it can simply reduce their visibility to 0, prevent their messages from being seen by anyone, block their super-likes and render their boosts ineffective. What this does, is that it effectively makes the app look like it’s still working you just get very few if any matches, and your response rate goes through the floor. This has the added benefit that it keeps people paying for subscriptions, buying consumables and swiping, instead of cancelling their subscription and deleting their account.
If you just simply got banned, then Tinder loses the revenue, and their user-base goes down.
Best Practice if I’m Right
In order to make this somewhat practical, I decided to write out a “Best Practice” if I’m right with an associated graph. This is based on two things:
A) Knowledge of which data Tinder records
B) Some knowledge I may or may not have about how to set up automated systems to analyze, categorize and respond.
The graph is pretty simple and is based on some very simple ideas. When it comes to the swipes, bots are going to be swiping 100% right, because their goal it to distribute their message to the highest number of matches possible. A lot of men are also going to swipe 100% right because they are simply looking to maximize their number of matches. Validation seeking girls with no intention of meeting up with, or even talking to their matches, or who are there to market their instagram feed to thirsty betas are also going to swipe right a lot more, but they will never respond.
The core principle here is simple, only swipe right on girls you would be happy to match with and that you have a 90%+ chance of opening if you match with them.
When it comes to matches you want to be in the 30% – 100% range to be honest. I wrote 30 – 70% because I suspect that 100% matches would trigger a few automated systems at Tinder HQ because it’s rare to have a 100% match rate, at least if you are swiping normally and not using Tinder Gold. If you are within 30 – 70%, or ideally 50 – 70% you are playing within your league enough that it won’t set of any Red flags.
For opens, it comes down to the fact that if you match with a lot of people but never talk to them, then Tinder assumes that they were not good matches. You have to remember, Tinder’s business idea is to match people up, if it fails to do that, it will eventually go out of business. If you do not open your match and your match does not open you, it assumes it was a poor match. Also keep in mind that Tinder does track the length of your messages, and whether you are sending the same message to multiple people at the same time, and this may set of “Bot flags”.
For responses it goes to the same thing as above, if you get very few responses, rarely get a conversation out of it, the girls unmatch or even worse, report you, that negatively impacts your account because Tinder assumes it was a poor match and that you are an asshole in the club. If you are the hot, rich, asshole in the club you are probably fine, but if you are not then you have a problem.
Summary and Conclusions
My idea behind writing this came from the fact that I’m recording a lot of Tinder data at the moment, and I want to have a baseline assumption of how everything works until I get colored by the data. In writing this, I tried to conceptualize based on what data Tinder saves about you, how I would go about setting up an automated detection system that acts as a bouncer in the club while maintaining for the best of the enterprise as a foundation principle. This is similar to why I tend to use mute on twitter rather than block, if I block someone they get that delightful screenshot that they can brag about, they know I’m no longer seeing their messages and they can create a second, third or fourth account to keep pestering me. If I mute them, they keep screaming into the blue nothing, doing nothing except wasting their energy.
I’m not sure that if I’m correct with my assumptions and guesses in writing this essay, however from the perspective of the fact that Tinder has to automate most of the processes that decides the visibility of users, by avoiding behaviors on the negative extremes (all right or all left swipes, being out of your league, never opening, never getting responses and being reported) you will stay under the radar of the automated systems that are designed to catch bot behavior. I would also recommend that instead of doing 1 massive swiping session every day, split it up into multiple smaller sessions.
A Final Note on Dating Apps and Text Game
Traditional text game has some rules:
A) Never respond immediately, ideally take [her time to respond+x] to give her an illusion that you’re busy.
B) Keep a 2:1 message ratio
C) Never write longer messages than her
D) Keep texting to a minimum only use it to set up a meet
I’ve followed these, and still do for my snap, text and whatsapp game, however for Tinder given the potential risks of the algorithm, I would recommend doing a longer, and on point opener, while sticking to the rest of the rules, with the exception being D. I suspect that occasionally having longer conversations in chat on Tinder may avoid triggering some detection flags. This is purely based on the fact that Tinder doesn’t know that the 3 messages that were exchanged resulted in a meet up, and a lot of short, but highly successful messages may be interpreted as something else.
The downside of most commercial grade behavioral analysis software is that it can only identify the “What” it cannot identify the why. For instance if you text a girl “Hey, meet up at 18 [insert location]” and she writes “OK” back to you, the software only sees that you sent 1 short message, then she responded with a short message back, and then no more conversation. This was a highly successful interaction for you, but for the software it looks like a mismatch.
Most software like this cannot make sense of the patterns that it sees, Tinder’s algo is similar to Facebooks in that it just locks you harder and harder into an echo-chamber, if you liked a story by Fox News, you get more Fox news, if you keep liking accounts who never swipe right, never respond to messages, have no bios except their insta-handle, and are ran by thirst-miners, then Tinder will keep giving you more or them.
For your information:
I will be on Red Mornings tomorrow with Rian Stone, catch us live at 9 AM EST for the only Red Pill Morning show: https://t.co/H2LSfrKIia
I will also be on the Red Man Group as usual at 11 AM EST: https://www.youtube.com/channel/UCY0fILilUr601SPy-wc0XUQ/videos