I was shocked when a good friend of mine told me he had just bought a burner phone specifically to use with Tinder. What? Why would you do that? I asked.
My friend, like many other Tinder users, felt that there was a clear point in time where their matches, replies, etc all dramatically dropped, to the extent that it felt suspicious. Instead of being fully banned where they cannot open the app, many users feel that they are “shadow banned.” Despite being allowed to see and swipe on others’ profiles, they believe their profile is invisible to all other users so that they will never (or rarely) get matches.
To be clear, I’m not even sure “shadow banning” actually occurs on Tinder. I’ve yet to see concrete evidence proving its existence (just lots of anecdotes about people getting less matches 1), nor do I actually think it is something that is provable unless you had insider information from Tinder.
So while I do not work for Tinder, nor am affiliated in any way, I do have experience running software products that provides insight into what could be happening, if anything at all. While I am not Tinder, if Tinder were to hire me, I can tell you all of the things I would do (even the over-the-top ones) to try to effectively enforce shadow banning; therefore, I can also tell you all the ways in which you could circumvent any form of shadow banning measures I would implement.
Because I always want to help my friends, I gave him a thorough guide on how to avoid any potential shadow banning, and to probably even utilize Tinder’s matching process in your favor. He told me that there’s a lot of misinformation out there, and that this type of insight being publicly available would be very useful to other people.
This guide is long, because it is thorough. Feel free to skip around if needed using the section links below.
Also… for the love of god please don’t actually implement all of these things. If you care enough about Tinder to implement all of the below, you need to get a life.
I hope you enjoy my meticulousness.
Why your Tinder account could be getting less attention
Once upon a time, Tinder used an algorithm known as the Elo rating system to give its users a “desirability score” that assisted with ranking. Elo is typically a system used for competitive games, such as chess, American football, League of Legends, and more; in some sense Tinder is a competitive game as well, where your move set is simpler, like rock-paper-scissors. If you like another Tinder “player” but they haven’t liked you back, you’ve “lost the match,” and thereby your Elo score would go down. Your next “opponent” would be another profile closer to your new rank.
Tinder itself has publicly stated that they once used Elo scores, but now no longer rely on such a system. This makes perfect sense, as the tech industry has evolved in the past 5 years or so to have a thriving Artificial Intelligence and Machine Learning (AI/ML) employee base. While the Elo system is cool and smart, it is old news; Arpad Elo came up with the idea around 1940. In contrast, many people say we are in a machine learning “golden age,” with lots of new developments each few years that promise better results than older systems like Elo rankings. And when I say “better results,” what is it that I mean? Well, machine learning is a practice where you can generally tweak your model to optimize for essentially any parameter, but the thing Tinder cares about most (as all for-profit businesses do) is revenue. Often in large tech companies, teams will measure second-order metrics, such as user engagement, time spent in-app, or other factors, but ultimately the company leadership will pick these metrics because they correlate with revenue, but are easier ideas to conceptualize and implement.
This means that when an ML team is talking to product managers and executive leadership at Tinder, they are likely being told to optimize the matching and ranking system for things like “maximum number of in app purchases” or “most time spent in app.” Let’s be clear here — these outcomes are not at all the same as optimizing for “every single user meets a compatible date.” When looking at Tinder’s user base as a whole, whatever model is in place may increase revenue or engagement, but it also may easily have “unintended” side effects, such as certain profiles being shown less often for no specifically describable reason.
What are Tinder’s incentives for shadow banning?
I find it very unlikely that Tinder intentionally is trying to “shadow ban” or otherwise make certain users’ experiences bad. Instead, as I mention above, it is more likely a quirk in their matching process.
You can believe that Tinder does or doesn’t give a shit about its users — that’s not something I honestly have an opinion on. I will however note, that in the best case that they do care about their users, it is still an exceedingly hard problem to understand the quality of experience you are providing to your users.
A quick google search on the topic tells me that Tinder has 57 million users worldwide, of which almost 6 million are paying users (Tinder Gold/Plus/Premium/idk). Can you imagine asking 57 million people across 150 countries and 40 languages how they like your app?
Often times companies try doing this. It is typical that they will run standardized NPS surveys with supplemental questionaries in order to get a sense of how users feel about their service, but the sorts of feedback you can give are limited. That said, most people I know don’t answer in app surveys, and turn off all email newsletters they can. Even for the people who strongly believe that Tinder is shadow banning them, I highly doubt most of them they have successfully submitted their feedback through a channel where Tinder’s product teams would see it. At best, custom care might see the compliant, and the representatives are likely instructed to give a generic answer and advise that you try paying for a profile boost.
That is the “best case,” where Tinder cares. If instead you are just a subhuman money printing machine to them, then there is little incentive to make your free experience good if there is a high likelihood that you will pay instead.
How might Tinder’s matching algorithm actually work?
At a very high level, you can think of most machine learning as putting a few numbers into a complex mathematical simulation, then getting some output, and adjusting your mathematical model accordingly. What this likely means in terms of implementation details is that Tinder’s ML model takes in a few variables (one of which is likely ‘time spent on Tinder,’ which Tinder claims is the primary factor in matching) and outputs something like a rank or a list of matches. The exact way that the model goes from inputs to outputs is unknown to even the engineers; they are inputting variables into a complex math system to simulate what should cause the best outcome (which again, means most money for them).
“Time spent” in terms of raw hours doesn’t really make sense as an input though; if this was the sole deciding factor, new users with barely minutes on the app would get no matches.
Assuming Tinder is not completely lying about this being a parameter (which I doubt they are lying here), this is either a multivariable system, or “time spent” is further derived into some metric like “time spent in app since account creation.”
It is believable that is time spent is a metric use, taking a Tinder hiatus could mean your inactivity time may reach an unrecoverable amount, where even returning to the app frequently wouldn’t be enough to change the model’s output enough.
This is simply a theory on what might be used as an input, but the underlying concept is likely at play regardless of what Tinder is trying to analyze; there are some set of things that they are measuring that they at one point in time received such a strong signal, that further action will not overturn this input, and you are pigeon holed into where the ML model places you.
In contrast, new users are likely measured and matched in a very different way. For one, new user experience (often labeled NUX) is something tech companies often heavily focus on, as they believe it is critical to longer term user retention and profits. But secondly, from a technical perspective, new users have a cold start problem — with little to no data to use as an input to the ML matching system, you may be supplemented by being seen by “hotter” or higher ranked profiles so that the model can better learn your “value” as other users respond to your profile. Basically, new accounts probably get more attention so Tinder can quickly figure out how to categorize your profile.
From this perspective, when users complain about “shadow banning,” it is more likely that they are observing their Tinder return to normal behavior, since Tinder no longer needs to artificially boost your profile for the purpose of data collection.
tldr — there are many factors at play that a machine learning model might use to reduce the frequency your profile is shown. You may not be able to change Tinder’s ML ‘output’ to deem you popular given past ‘inputs’ on that account, but new accounts are a blank slate that might actually initially get lots of attention just because it is new.
Creating a new Tinder account and ‘resetting’ your rank
If in theory Tinder’s matching or ranking process has an unexpected consequence and you become “shadow banned,” there wouldn’t be a way to fix that for your account. All of the info Tinder uses to decide who sees your profile is tied to your account, and has no way to forcibly refresh. Technically speaking, Tinder may have some form of indirect process in place, such as queuing you for re-assessment and re-ranking once you take actions like uploading new photos that could theoretically have an impact on your desirability — that said, they may not, and this is by no means a reliable or consistent way to “reset” your rank.
The only surefire way to have Tinder try to reclassify you is by creating a new account that cannot be tied to any of your previous account’s data. That sounds like it could be very easy — and it likely is, because Tinder probably only employs a few of the below strategies to try to see if your new account is actually a remade profile. That said, this whole essay is being written to be meticulous, so I will help you understand all potential methods Tinder even has available to link your new account with the old one, so you can take your tinfoil hat off and rest easy at night.
How can Tinder identify you?
For the sake of our discussion, I’m going to assume they aren’t actually deleting all your data. This is actually plausible, as the EU recently began a review on how Tinder handles data. TechCrunch reports:
… the Tinder probe came about as a result of active monitoring of complaints received from individuals “both in Ireland and across the EU” in order to identify “thematic and possible systemic data protection issues”.
It’s not clear exactly which GDPR rights have been complained about by Tinder users at this stage. But some users have accused the company of not providing a copy of all the data it holds on them.
Either way, I suggest deleting your account if you want to make a new one, and we’ll pretend we still need to be proactive to prevent any account linking.
While the most obvious way for Tinder to connect various accounts you’ve made and assume they belong to one person is to use your login info (same phone number, email, Facebook login, etc), there are actually several techniques that they have at their disposal.
Identity via accounts (email, phone, and social media you link)
Any account that you link to Tinder could be used as a unique identifier. Tinder is interesting in this way, because unlike most apps that might ask for you to log in with Facebook or Google as soon as you open the app, Tinder will let you link other accounts like Instagram (which is owned by Facebook) or Spotify later on as a way to supplement your account. However, both of these logins can be used as an identifier. Even if you bought a brand new phone and downloaded Tinder anew, and made a new account with a new phone number, Tinder could use any of these other apps you authenticate with to link your newest account with an older account you also register with the same Instagram or Spotify account. Because of this, I wouldn’t recommend that you use these Tinder features, as they give away more signals about your identity.
When creating your new Tinder account, just make sure that you are using an email or phone number that you have not used before, and don’t use any other social logins, and you should be good to go (as far as logins are concerned anyway).
If you think Tinder doesn’t check too carefully, and you have a Gmail account, you can apply one of Gmail’s filtering features to effectively spoof having a new email.
Gmail has a feature where if your email is firstname.lastname@example.org, you could sign up for a service with email@example.com, where “random words” can be anything you want. This phrase after the ‘+’ sign then because the title of the filter that Gmail applies to emails coming in to that address, but they all still show up in firstname.lastname@example.org’s inbox.
This can be done indefinitely with most websites and apps that require accounts. You can keep making new accounts with example+1, example+2, etc… and they will all send emails to email@example.com.
Alternatively, you can just make a brand new email account. This is more secure, as if Tinder is serious about profile linking, they’ll take Gmail filters into account.
10 Minute Mail is another great service for these types of things, where you can make a temporary new email that only lasts 10 minutes. That’s likely plenty of time to make a new account and click the confirmation link in the email. Tinder may block the domains used by this email though. 10 Minute Mail tries to update their domains often, but they can only do so much.
Your most serious option is to make a Proton Mail account just for this Tinder profile. Proton Mail is the most secure email service I am aware of, and for our purposes, it is free.
If Tinder requires you to use a phone number (I’ve seen them play around with taking email login away), then you can likely get a Google Voice account for free that you can use as a secondary phone number. Tinder will not be able to tell that this number is owned by your specific Google account. Alternatively, there are services like BurnerApp (and many others) that you can use to get a temporary cell number.
If you really want you can go to Boost Mobile and buy a burner phone with cash, but this is unnecessary and provides no extra security from Tinder. They already won’t know if it is you using a Google Voice or BurnerApp number.
Identity via device (IDFA and other advertising IDs)
Some people online claim that you should use an entirely new phone because the App Store or Play Store give apps your iCloud or Google Account information. This is not true; it is not necessary to buy a new phone or create a new account to manage your device’s App Store.
Other device information, such as unique device ID (UDID) or International Mobile Equipment Identity (IMEI), are not things that an app can request directly (except maybe if you are on Android and the app has root access, which Tinder does not do). You do not need to buy a new phone.
There are however, some things that you can be tracked with, which are mostly designed for advertisers to track you across devices. On Apple, this is known as Identifier for Advertisers (IDFA), and on Androids there are Android Advertising ID (AAID) and Google Advertising ID (GAID). Apple is revamping IDFA entirely by 2021 so that tracking is more limited; people are speculating that Google may follow suit, but it is unconfirmed.
Doing a simple google search for “disable IDFA” (or whichever is relevant to your device), or more generally “disabling advertising identifiers” + [your phone name] will likely give you a guide relevant to your device.
tldr — don’t waste your money on buying a new device. The only hidden device IDs available to Tinder without your explicit consent are for advertising, and you can turn them off.
Identity via metadata (IP address, location)
While metadata (data that provides information about other data) is very indirect, at times it can still be very meaningful. In the semi-famous words of General Michael Hayden, former director of the NSA and the CIA, “We kill people based on metadata.”
Even if you were to use brand new accounts (email or phone) to create your Tinder account, and you were to turn off device level advertising IDs, it is theoretically possible that Tinder try to fingerprint you based on some metadata they receive from your app usage.
The most popular piece of metadata companies use is your IP address. This is associated with your internet connectivity; it is the address for your device so that websites and apps know where to deliver stuff. IP addresses are not permanent though. You can usually reset them per device, or even at a router level (sometimes as simply as by unplugging and replugging it in). You can also look into using a proxy or a VPN if you are really serious, but it’s probably easier to just reset it on the device settings of where ever you are using Tinder.
In theory, Tinder could also try building an assumption of who you are based on your location data (which is not on by default, but Tinder does request permission for). This can’t inherently identify you, but depending on how much you move around, it could be unique. For example, if 40% of your time is at work, 40% at your apartment, and 20% is taking the same route to-and-from those locations, collectively you may be the only person in the world who has that set of GPS coordinates recorded at those times.
Apple now has more robust location permissions, and you can disable apps from gathering information when they are not running. Soon Apple also claims they will allow for you to give approximate location instead of exact GPS data. Nonetheless, even with these nice phone permissions, a few rough data points could be enough to uniquely identify you.
I highly doubt Tinder does this, but if they did, you could either limit your app usage to certain locations and restrict the app from collecting data unless it is in the foreground, or you could invest in something like a VPN that can spoof location data.
VPNs typically cost money, but they are a nice security measure in general. Location spoofing has some nice unrelated perks too, like getting your Netflix catalogue to reflect the country you are pretending to be in, and thereby expanding the shows you can access.
Identity via profile data (photos and facial recognition)
Perhaps the deepest rooted identifier Tinder could potentially leverage is your actual appearance. This one is somewhat unfortunate, in that how you look is one of the few things you are (hopefully) trying to accurately represent on the app regardless of what account you use. (Please don’t catfish.) However, while there are different methods Tinder could approach photo matching from, it’s likely you can combat most tactics.
At the simplest level, it is most feasible that Tinder would try seeing if any photos in their database are literally the same photo. However, being the “same” photo could be interpreted to mean a lot of different things, and there are different ways you could implement such a check. For example, common image file formats like .JPG have information known as EXIF data. This would include metadata like the device the photo was taken on, date the photo was taken, geolocation, and more. If two photos were taken at same time and place on the same device, I’d say that could count as the “same” photo. For this reason, I’ve seen some guides online suggest that you can fake image comparison systems just by deleting EXIF data. Your computer’s file browser can probably edit some EXIF data natively, or you can use a more professional tool like Photoshop.
But we haven’t even gotten to the most important part of analyzing photos — the actual image! You can think about a photo of yourself as a grid of pixels (each dot of color) that a computer places in a specific order for it to appear like an image. Each colored dot (pixel) has a numerical value, which your computer just sees as a list of numbers. Comparing two lists of numbers is easy-peasy, and Tinder could very plausibly do this.
In its most naive version, if Tinder is simply looking for an exact match, then literally editing even a single pixel would thwart this kind of check. If this is the extent of the check, any filters (like the colorful Instagram ones), cropping, drawing on top of, rotating, or flipping done to the photo would be considered a different image.
This is the most naive version of such checking though. Pixel data represented as a matrix (middle school math coming back to haunt you!!) that has simple effects like the above applied to it are essentially simple operations (like multiplication) applied to the matrix. If that is over your head, just know that it’s not that sly of a trick if you truly want to fool a sophisticated machine.
You could try any of these tricks without much effort, as ones like flipping an image are especially easy and can be done on your phone. But if you think Tinder is really trying their hardest to cockblock your love life, you might need to apply more effort.
The simplest “more effort” option is to just pick entirely different photos that are not just transformations of an image you previously uploaded to any other Tinder account, but entirely new photos you’ve never used on Tinder before.
Depending on how heavily Tinder is investing in machine learning (ML), they could still have “facial recognition” style photo matching. Realistically speaking though, Tinder doesn’t have the infrastructure to do this, nor is the technology developed enough to use facial recognition accurately in this way.
For an example of what this could look like in a product, go check out Google Photos — if you upload enough pictures and initiate a new search through your photo library, you will see that Google lets you view only photos of a specific face. This has a lot of limitations though; Google is often wrong, or has to ask you for confirmation that this is indeed two of the same face. Facial recognition, even at a company like Google that has internal units like Google Brain and DeepMind that perform world-leading ML research, and even for a product like Google Photos that is a loss leader (it loses money because it keeps you using Google products overall), still has many issues. Tinder would at best have these same issues, and at worst, is not even suited to operationalize this level of facial recognition.
It’s very unlikely that even if they could do Google Photo’s level of facial recognition that they would use it due to the differences in the product’s use cases. For one, Google is only matching faces per photo library; that is, they are not trying to match a face in my photos to a face in your photos. The set of possible faces for Google Photos is much lower, (only ~100 possible options for faces with several examples each) and due to the lower volume compared to Tinder (millions of possible faces to compare to, with a profile defined by only ~5 photos), they aren’t even really comparable. Beyond that, the repercussions of Google being uncertain or simply wrong about two faces matching is also very low; usually this is resolved by asking the user to help confirm, whereas Tinder cannot reasonably have such prompts in the app. Everyone you see in Tinder is a stranger to you, and no user could help verify another user’s identity.
Let’s say that for some reason, Tinder is a hidden ML powerhouse, capable of richer image detection than even Google’s research labs have demonstrated, and can apply it on a more difficult facial recognition problem.
There are potentially some techniques you might have at your disposable, but none are particularly easy.
A Google ML tool called Tensorflow has a public example of adversarial ML techniques, such as introducing specifically designed subtle static (which Google refers to as “perturbations”) into the background of an image. To oversimplify the example a bit, by performing specific calculations to see how much each pixel contributes to the label given (in this example, the label is “panda”), you can add well chosen but nearly imperceivable changes to the most important pixels, and trick the machine entirely.
In this example, the Tensorflow documentation mentions that this is a white box attack. This means that you had full access to see the input and output of the ML model, so you can figure out which pixel changes to the original image have the biggest change to how the model categorizes the image. The box is “white” because it is clear what the output is. Our attempts to fool Tinder would be considered a black box attack, because while we can upload any image, Tinder doesn’t give us any information on how they tag the image, or if they’ve linked our accounts in the background.
That said, certain approaches to black box deception basically suggest that when lacking information about the real model, you should try to work with substitute models that you have deeper access to in order to “practice” coming up with clever input. With this in mind, it could be that static generated by Tensorflow to fool their own classifier may also fool Tinder’s model. If that is the case, we would want to introduce static into our own images. Luckily Google will let you run their adversarial example in their online editor Colab.
This will look very scary to most people, but you can functionally use this code without much idea of what is going on.
First, in the left side bar, click the file icon and then select the upload icon to put one of your own photos into Colab.
Then, under the Original Image step, delete the line where Google provides an example image:
image_path = tf.keras.utils.get_file('YellowLabradorLooking_new.jpg', 'https://storage.googleapis.com/download.tensorflow.org/example_images/YellowLabradorLooking_new.jpg')
And replace the line below it with:
image_raw = tf.io.read_file('YOUR_IMAGE_FILENAME_HERE.jpeg')
Replace my ALL_CAPS_TEXT with the name of the file you uploaded, which should be visible in the left side bar you used to upload it. Make sure you use a jpg/jpeg image type.
Then look up at the top of the screen where there is a navbar that says “File, Edit” etc. Click “Runtime” and then “Run All” (the first option in the dropdown). In a few seconds, you will see Tensorflow output the original image, the calculated static, and several different versions of altered images with different intensities of static applied in the background. Some may have noticeable static in the final image, but the lower epsilon valued output should look exactly like the original photo.
Again, the above steps would generate an image that would plausibly fool most photo detection Tinder may use to link accounts, but there is really no definitive verification tests you can run because this is a black box situation where what Tinder does with the uploaded photo data is a mystery.
While I myself have not tried using the above technique to fool Google Photo’s face detection (which if you recall, I am using as our “gold standard” for comparison), I have heard from those more knowledgeable on modern ML than I am that it doesn’t work. Because Google has a photo detection model, and has plenty of time to develop techniques to try fooling their own model, they then essentially just need to retrain the model and tell it “don’t be fooled by all of those images with static again, those images are actually the same thing.” Going back to the unlikely assumption that Tinder has actually got as much ML infrastructure and expertise as Google, maybe Tinder’s model also wouldn’t be fooled.
If you are concerned that entirely new photos that have never been uploaded to Tinder will be linked to your old account via facial recognition systems, even after you’ve applied common adversarial techniques, your remaining options without being a subject matter expert are limited.
At this point, my best advice is to take brand new photos of yourself at strange angles (such as mostly side profile shots), in either harsh or dim lighting, while wearing obfuscating face gear like sunglasses, hats, or masks, and potentially with substantial makeup on.
As far as I am aware, there’s no accurate abs or boob recognition software — depending on your goals and your physique, you could possibly both fake out Tinder and get more matches by showing off your body without your face in some of your photos.
If that still isn’t enough for you, I think you should probably not use Tinder. I suggest you instead try (1) other dating apps or (2) therapy.
How likely is it that Tinder does these things?
I would say it’s very very unlikely that Tinder goes through the effort to do most of the above. There are entire advertising driven companies that make billions of dollars trying to connect your laptop usage to your phone usage so that they can better serve ads, and even these places won’t bother with most of the above. Ultimately, Tinder is a business trying to make money, and it would be a misallocation of resources to invest EPD (engineering, product, and design) time on solving these non-essential problems so meticulously.
In the event that they do try to perform account linking to preserve shadow bans (or simply old ranking data that isn’t intended as a ban) for new accounts, and you do take all of the above precautious, you should be able to circumvent their attempts to stifle your love life.
If you take all of the above steps and still believe you are shadow banned… I’m sorry to be the one to break it to you, but you’re probably ugly.
Although you would assume that dating coaches like School of Attraction would be the most outspoken critics of Tinder’s shadow banning practices, I have yet to see any of these love gurus comment on the topic. That’s not evidence of anything of course, but maybe this implies that the problem isn’t very widespread (or existent)↩︎