Dating apps show the same types of people because their ranking systems learn from swipes, and a person’s swipes tend to repeat a narrow pattern. Each like or pass is recorded, compared with the behavior of other users, and used to decide which profiles appear first the next time the app opens. A user who likes one kind of face or profile for a week trains the feed to supply more of it.
Ranking Systems Built on Swipe History
Many large apps use some form of collaborative filtering, the same family of methods that recommends films and products to shoppers. The system groups users whose likes resemble each other and then shows each person the profiles that similar users liked. If 2 users have liked the same 20 profiles, the next profile one of them likes becomes a strong candidate for the other, and the two feeds overlap more with each new like. New users present the opposite problem, known in computer science as a cold start. With no swipe history to compare, the system relies on basic settings such as age and distance and on profiles that other users already like.
A profile’s appeal to other users is the second input. In 2016, the chief executive of one of the largest swipe apps confirmed that it ranked users with an internal desirability score modeled on the chess rating system, in which a like from a highly rated user counted for more than a like from a low-rated one. The company said in 2019 that it no longer used that score. Its description of the replacement said the order of potential matches is adjusted each time a profile is liked or passed, and that regular activity in the app raises a profile’s visibility. Observers noted at the time that the new method still appeared to depend heavily on swipe behavior, so profiles that collect likes quickly still appear near the top of many feeds at once.
Filters and Pool Size
User-set filters shrink the pool before any ranking happens. Age range and distance limits each remove candidates, and a tight radius in a small city leaves few active profiles in any given age band. Once the pool is that small, the same faces return within days because the app has little else to show. In a 2023 Pew Research Center report, 53% of adults under 30 said they had tried online dating, compared with 37% in the 30-to-49 group and 20% of people aged 50 through 64. An older user starts with fewer candidates before setting any filter.
Niche dating sites do part of the filtering before any algorithm runs. Options such as a sugar baby website, sites for rural singles or sites for vegans each start from a pool defined by one shared trait. In a smaller and more specific pool, the profiles that repeat are closer to what the user came for.
The Desirability Hierarchy
Elizabeth Bruch of the University of Michigan and Mark Newman studied heterosexual users of a free online service in four large American cities, using the messages they sent in January 2014. They ranked each person’s desirability by the number of messages received and by how desirable the senders were. Men and women alike contacted people about 25% more desirable than themselves on average, and the chance of a reply fell sharply as that gap widened. Users also wrote longer messages to more desirable people, although Seattle was the only city where longer messages showed signs of a better reply rate.
Women’s desirability scores in the study tended to decline from 18 to 60, while men’s tended to rise until about 50. A system trained on that behavior learns the same hierarchy. Profiles near the top receive the most messages and likes, and the system treats that attention as a sign of quality and shows those profiles to still more users.
Exposure Bias Toward Well-Liked Profiles
A 2024 study in the journal Manufacturing & Service Operations Management measured this effect directly. Using data from more than 240,000 users of a large platform in Asia over three months, the researchers found that the platform’s algorithms recommended attractive users more often, and that a user’s chance of being recommended rose sharply as that user’s average attractiveness score went up.
Their model also examined the platform’s incentives. Well-liked users generate more likes and messages, so the model suggested that unbiased recommendations would lower revenue and produce fewer matches, unless those users became so selective that others saw them as out of reach. The authors suggested that the bias may be low while a platform is growing and its reputation depends on a high match rate, and that it may increase as the platform matures.
Race and Preference Filters
In her 2024 book, media scholar Apryl Williams of the University of Michigan examined company patents and interviewed more than 100 app users, and she argued that the ranking systems of major platforms reproduce racial and sexual hierarchies. A 2018 paper by Jevan Hutson and colleagues at Cornell University examined bias and discrimination on what the authors called intimate platforms. Reviewing design features such as ethnicity filters, the authors recommended that platforms offer categories other than race and write matching systems that do not discriminate. Among the cases the paper cited was a 2016 news report that one app showed users only potential partners of their own race, including users who had said they had no preference.
When a user filters by ethnicity, or when that user’s swipe history shows a consistent racial pattern, the system has little reason to show anyone outside it. The repetition the user notices is partly a record of their own earlier choices.
The Composition of a Dating Feed
A dating feed is a ranked forecast. It combines the filters a user set, the profiles that user liked or passed, the behavior of users with similar histories, which is the core of collaborative filtering, and the attention each profile has collected from everyone else, and it orders the results by the predicted chance of a mutual like. About 3 in 10 American adults have used a dating site or app, according to Pew, and each of them sees a feed built mostly from their own past behavior and from the reactions of strangers with similar habits. Different results require different inputs, which in practice means wider filters and likes given to profiles outside the usual pattern.
The post Why Dating Apps Keep Showing You the Same Types of People appeared first on Daily Emerald.