It only took three days of swiping before he popped up. I froze, thumb hovering over the X. I scrolled through his photos and prompts, looking at what he had changed since I’d seen it the first time.
The first photo was the same: him holding a climbing rope somewhere remote, curly hair bursting from underneath a baseball cap. His simple pleasures were still “mountain roads, forests and alarm free mornings.” He had added a photo where he stood shirtless at the base of a cliff.
The twist of the knife was the note from Hinge at the top: “Most Compatible: We think you two should meet.”
In my head I could hear Hinge’s version of Microsoft Word’s ’90s-era paper clip helper, Clippy, squeaking at me: “It looks like you like hiking and concerts, would you like to be connected to this other person near you who likes hiking and concerts?”
The app couldn’t know two of its users had taxied down the runway of dating but never took off—a classic situationship, as the kids call it these days. The algorithm just saw a 31-year-old, outdoorsy local working in biotech and connected the dots to an outdoorsy 30-year-old a few miles away working in science media. And just like the Microsoft users of 1997, I hated it. I wanted to crush the computer program that thought it knew what I wanted, whether that was to write a letter or my perfect match.
The promise of dating apps is to show you all the romantic options in your city, but behind the scenes, the algorithm is cultivating a very specific, limited, at-least-somewhat-distinct dating landscape for each user.
The first big dating site was Match.com, which was founded in 1995 and followed by eHarmony and OkCupid in the early 2000s. These sites touted their surveys, compatibility scores, and science-backed approaches to pairing up couples as a better way to find long-lasting love. Such compatibility-based approaches to online dating dominated until 2009, when gay dating app Grindr hit the scene and changed online dating forever.
Grindr, as a mobile app, organized the romantic options not by compatibility but by distance—the top person was the one closest to you. This is still the default on Grindr today. When Tinder took Grindr’s idea to the straight world in 2012, it duplicated this perception of being distance-based, if not exactly in its code.
“When you think about platforms like OKCupid and eHarmony, it would be hard to use those and not know that there’s an algorithm, because it’s so much at the forefront of what they do,” said Liesel Sharabi, an Arizona State University scholar who studies dating apps. “But when I talk to people who use Tinder, they don’t always know that there’s an algorithm. A lot of people think it’s just showing the people around them, and it’s a lot more complicated than that.”
In 2016, Tinder confirmed it was using an Elo score, traditionally used to rank chess players, to rank users on desirability and match them accordingly. The media storm was quick and strong; by 2019 Tinder was claiming that it no longer used the Elo score, though it is still probably using some, if not many, algorithms. Since then, most dating app companies take a black box approach and don’t talk publicly about what factors into their algorithms.
So let’s get back to that Hinge recommendation of an ex that sent me down the rabbit hole of dating app algorithms as well as caused me to spiral about being single for a week. Hinge’s Most Compatible feature, Discover home screen, and paywalled Standouts feature are all subtle returns to the compatibility algorithms popular in the ’90s. The people in Discover are “members who both 1) match the preferences that you selected and 2) are interested in people like you,” per the company’s website. (The company exchanged a few emails with me before ghosting my requests for an interview; how fitting.) And the Most Compatible feature pairs you with people you are most likely to like and who are most likely to also like you.
All recommendation algorithms, including Hinge’s, work on a system called collaborative filtering, according to Sharabi. The same type of recommendation algorithms power Netflix, Facebook, Amazon, YouTube, and almost any social media platform. Unlike OkCupid and eHarmony, whose matches are based on survey answers, collaborative filtering works by recommending options based on what people similar to you have watched, clicked on, joined, bought, or swiped right on.
One of the big problems with this method is that the people or things who are liked by the most people get recommended to the most people—and then get liked by even more people. This causes a “rich get richer” effect that favors people whom the majority of users on the app already find attractive. (This is especially problematic for minority users, who are the victims of unconscious racism and thus lower desirability on the apps and end up getting ostracized. Dating apps have been skewered for this before.)
A big reason people turn to dating apps in the first place is because they are looking to expand their pool beyond their day-to-day offline bubble, in hopes of using the technology to find the other thousands of romantic options nearby. But that hopefulness can turn to despair when the filter bubbles kick in. According to Sharabi, if the majority of users are not swiping on you, you essentially become invisible.
As the internet has become ubiquitous in modern life, there has been no shortage of criticism of recommendation algorithms and social media filter bubbles. They have been accused of increasing polarization, and YouTube’s recommendation method was even the center of a New York Times podcast chronicling one man’s descent into the alt-right. But there’s been less introspection into the very similar algorithms used on dating apps, a service almost 30 percent of US adults have tried. Ten years ago, online dating was responsible for one third of marriages; that number is sure to be higher now.
“When you think about the decisionmaking happening on dating apps, if you’re looking for a serious relationship, you’re making one of the most important decisions that you’ll ever make,” Sharabi said. “People are letting algorithms make those decisions for them, or at least augment their decisionmaking. And I think it’s important to pay attention to what exactly those algorithms are doing.”
I’ve been on dating apps for almost a decade now, and I’ve started to see the patterns of the algorithms figuring out my type from my swiping. I’m seeing the same men over and over again. And not just the same type of man, but the exact same profiles. I’ve deleted and reinstalled Hinge a few times, and that curly-haired ex always seems to quickly pop up after the algorithm has gotten a hold on my swiping behaviors. I frequently see and match with the exact same man across multiple different dating apps, sometimes never exchanging a single message.
Just like how a pair of shoes you clicked on suddenly follows you around the internet from Facebook to Instagram to Google, the same can feel true on dating apps. I feel at the mercy of my dating app algorithms, just as I do my YouTube suggestions or my Twitter feed, but I can’t even pick channels to subscribe to, or people to follow.
Of course, these filtering algorithms are necessary in some ways. If you were bombarded with 500,000 options, you would be completely overwhelmed and frustrated. We all live in our own offline filter bubbles of our own making depending on the work we choose, the hobbies we pursue, the college we went to, and the friends we keep. Most people are attracted to partners who are similar to them, so on the surface, these filters seem like a good basis for a dating algorithm to operate off of. And I should give credit where credit is due: The app was recommending a person I did in fact really like. The relationship had just already run its course and the algorithmic recommendation felt like a painful confirmation that we would have made a good couple, sprung upon me during my search for someone new.
The problem starts when we become stuck in our algorithms and don’t even know what they are factoring in. “The issue is that you don’t have a lot of control,” Sharabi said, “You don’t know what the algorithms are doing in the background. And you also can’t opt out. So if you get trapped in this filter bubble where you’re seeing the same types of people over and over again, you’re gonna have to change your behavior and wait for the algorithm to essentially catch up.”
Unlike the surveys of last generation’s online dating, the apps aren’t even asking what I want, but inferring it from other people’s behaviors. I don’t have the power to explore something outside of my usual type or the type of people the algorithm thinks I like. “When you go about your day-to-day, you know you’re not seeing everybody, and I think it keeps people optimistic," Sharabi said. “If I try something new, if I go to a new location, I might meet somebody through a different network. But I think on dating apps, because they have so many users, you do start to feel like this is it. This is the entire pool. And that’s not the case.”
The filters put specific boundaries and constraints around the dating pool without ever telling you that, or how it’s doing it. When I’m seeing the same profiles again and again, it feels like I’ve seen it all, and I live in an area with almost 8 million people. The algorithm is limiting the dating pool—showing me how small it thinks it is.
Dating app companies remain opaque for both good and frustrating reasons—they don’t want people gaming the system, and they don’t want to give away their trade secrets. Still, Sharabi thinks users could benefit from the app companies specifically informing them how their swiping behaviors are influencing the pool.
There is the core question of what we want from our dating apps. Do we want them to mimic the offline world, or is there a missed opportunity to shake up some of the norms? Bumble’s “women make the first move” approach flipped the script on traditional dating conventions. Are there other changes we could make?
What about an Explore tab on dating apps, like the one on Instagram, that shows you people outside your normal type? Tinder actually has a version of this where you can explore people by interest—thrill-seeker, foodie, self-care, and so on—though these interests are still abstract and it isn’t clear how people end up in each group.
Sharabi wants a random component option, like Netflix’s Surprise Me feature. (However, I would guess there is an algorithm at play there as well.) Could dating apps incorporate more feedback about potential matches instead of just a “yes” or “no,” the way other platforms list possible reasons why you liked or disliked an ad? What about stealing a page from Spotify and creating a Tastebreakers pool of people you don’t normally see? And on the transparency side, wouldn’t we all love to know just what percentage of the pool we are actually seeing, instead of blindly swiping through as quickly as possible so we can feel like we’ve seen as many options as possible?
The way I can get sucked into doomscrolling on Twitter is the way I get sucked into a Desperation Scroll on dating apps, hoping that behind the next swipe there will be someone new, someone different, someone exciting, someone better. But instead it’s just the guy I already knew I liked, who doesn’t like me back, whom I’m trying to get over by swiping. Great job, Clippy.