Dating Apps: Expanding Horizons or Reinforcing Social Silos?

The DII Take

Dating apps sit at the centre of a social sorting mechanism with consequences far beyond the dating industry. If algorithms and filter options reinforce educational, economic, and racial homogamy, they contribute to social stratification. If they disrupt it by facilitating cross-class or cross-ethnic matching, they serve a democratising function. The evidence to date is mixed, but the design choices that platforms make—which filters to offer, how to weight preference signals, whether to show users outside their stated preferences—carry social implications that extend well beyond match rates. The platforms making these choices today are shaping the social fabric of the next generation, whether they recognise it or not.

What the Research Shows

Several research streams address assortative mating in the digital era. Educational homogamy has increased in recent decades. Research by Robert Mare and others has documented rising educational assortative mating in the United States and Europe: college-educated adults are increasingly likely to partner with other college-educated adults. The question is whether dating apps accelerate or moderate this trend. Platforms that include educational level as a filter (The League explicitly, and most platforms implicitly through profile display) facilitate educational sorting. Platforms that de-emphasise education in profiles may reduce it.

Ortega and Hergovich's modelling research (2017/2020) predicted that online dating would increase interracial marriage by connecting people across social network boundaries. Empirical data has broadly supported this prediction: interracial marriage rates in the United States have risen during the period of dating app expansion, though attributing this specifically to apps rather than broader cultural change is methodologically challenging.

The 'attractiveness matching' phenomenon—the tendency for people to pair with partners of similar physical attractiveness—may be intensified by dating apps. Research on swiping behaviour shows that users tend to 'swipe right' on profiles at or above their own perceived attractiveness level but are matched primarily with profiles at similar levels. The result is a market equilibrium where attractiveness matching is reinforced through the mechanics of mutual interest.

The tension between access expansion and preference reinforcement is the central social question of algorithmic dating.

The social consequences of algorithmic sorting deserve scrutiny. In an era of rising inequality and declining social mobility, the mechanisms through which people form partnerships matter for economic stratification. Couples who share educational and economic backgrounds pool resources in ways that concentrate wealth. Couples who form across class lines distribute resources more broadly. Dating apps are not neutral in this dynamic—they are sorting machines whose design choices influence the social structure of the next generation.

What Platforms Can Influence

Dating platforms cannot control the deep forces driving assortative mating—educational sorting, geographic segregation, and cultural homophily operate at scales far beyond any product design choice. But platforms can influence the margins, and the margins matter.

Default filter settings shape behaviour. If a platform defaults to showing only users within the same educational bracket, it reinforces educational homogamy. If it defaults to showing a diverse range, some users will match across class lines who would not have otherwise. The default matters because most users do not adjust default settings.

Recommendation diversity is a design choice. An algorithm that optimises purely for match probability will recommend similar profiles to similar users, reinforcing existing patterns. An algorithm that occasionally introduces diverse recommendations disrupts assortative patterns without significantly reducing match rates.

The profile information hierarchy matters. Platforms that display education, occupation, and neighbourhood prominently enable class-based screening before any interaction occurs. Platforms that foreground interests, values, and personality reduce the salience of socioeconomic markers, potentially enabling connections that would not survive the initial socioeconomic filter.

The Race and Ethnicity Dimension

Racial assortative mating is one of the most sensitive and consequential dimensions of the pattern. Research on dating app behaviour has consistently found that racial preferences are among the strongest and most persistent filters users apply. Studies by Curington, Lin, and Lundquist (2021) analysing millions of dating app interactions found that White users showed the strongest in-group preference, whilst users from minority groups showed more variable patterns—some preferring in-group matching, others showing out-group or no-group preferences.

The platform design implications are ethically complex. Explicit race-based filtering (allowing users to exclude entire racial groups from their recommendations) has been removed by some platforms but remains implicitly embedded in algorithmic recommendations that learn from user behaviour. If an algorithm observes that a user consistently swipes left on profiles of a particular race, it will learn to stop showing those profiles—effectively implementing a racial filter without explicit user action.

Some platforms have taken deliberate steps to increase cross-racial matching. Research by Ortega and Hergovich predicted that online dating would increase interracial unions by connecting people across segregated social networks, and the data on rising interracial marriage rates in the app era is consistent with this prediction. Platforms that include diversity in their recommendation algorithms—occasionally showing profiles outside a user's revealed preference pattern—may contribute to this trend whilst expanding users' horizons.

The Geographic Dimension

Geographic assortative mating—the tendency to partner with people who live nearby—has been dramatically disrupted by dating apps. Historically, the vast majority of partnerships formed within a few miles of both partners' homes. Dating apps expanded this radius substantially, connecting users across neighbourhoods, cities, and in some cases countries.

The consequences for social mixing are significant. Neighbourhood-level segregation by class, race, and education level means that geography-constrained matching reinforces socioeconomic homogamy. Apps that expand geographic radius introduce users to potential partners from different neighbourhoods, backgrounds, and socioeconomic contexts. The degree to which this geographic expansion translates into actual cross-class or cross-racial partnership formation, rather than merely increased browsing across boundaries, remains an active research question.

The Algorithm's Role in Social Sorting

Machine learning recommendation systems learn from user behaviour and optimise for engagement metrics (likes, messages, matches). This optimisation process has an inherent tendency toward assortative matching because people tend to engage most with profiles similar to their own. The algorithm observes these patterns and reinforces them, creating a feedback loop that narrows the diversity of recommendations over time.

Research on algorithmic bias in recommendation systems, primarily from the computer science literature, has documented this pattern across multiple domains: music recommendation systems narrow listener taste over time, news recommendation systems create filter bubbles, and social media algorithms reinforce existing social networks. Dating app algorithms face the same dynamic, with the additional complication that the social consequences of romantic assortment (who pairs with whom for the purpose of creating the next generation) are arguably more significant than the consequences of music or news consumption.

Dating apps are not neutral utilities. They are sorting mechanisms that influence who pairs with whom, and therefore who raises children together, who pools economic resources, and how social stratification evolves across generations.

Some platforms have experimented with diversity-enhancing algorithms. Hinge's 2025 algorithm updates, which nudge users to be more open-minded about certain filter preferences like distance, represent a modest step toward disrupting algorithmic assortment. More aggressive approaches—deliberately surfacing profiles that differ from a user's historical preference patterns, or de-emphasising socioeconomic indicators in recommendation ranking—face a tension between social benefit and user satisfaction. Users may prefer recommendations that match their existing patterns, even if broader recommendations would produce better long-term outcomes.

Implications for Social Mobility

The societal implications of dating app assortative mating extend beyond individual relationship outcomes to questions of social structure and inequality. Research by economists including Greenwood, Guner, Kocharkov, and Santos has estimated that increasing educational assortative mating accounts for a measurable share of the rise in income inequality in the United States. When college graduates consistently marry other college graduates, and non-graduates consistently marry other non-graduates, household income distributions widen. If dating apps intensify this pattern, they contribute to a broader social sorting mechanism with generational consequences.

The counterargument is that dating apps also facilitate matches across geographic, racial, and social boundaries that would not have formed through traditional channels. The Ortega-Hergovich research on increasing interracial marriage supports this perspective. The net effect of dating apps on social mobility likely depends on which force dominates: the homogamy-reinforcing tendency of algorithmic recommendation, or the boundary-crossing capacity of access to diverse partner pools. Both forces operate simultaneously, and the balance between them is a function of platform design choices that operators have the power to influence.

The Age of Transparency

Growing public awareness of algorithmic sorting in dating creates both a risk and an opportunity for operators. Media coverage of dating app algorithms has highlighted the ways in which platforms reinforce existing social patterns, leading to calls for greater algorithmic transparency and user control.

Platforms that proactively disclose how their algorithms work—what factors influence recommendations, how user behaviour shapes future suggestions, and what diversity mechanisms (if any) are built into the system—build trust whilst positioning themselves ahead of potential regulation. The EU's Digital Services Act already requires certain transparency obligations for large platforms, and dating-specific algorithmic transparency requirements may emerge in future regulatory cycles.

The research shows that most users' stated preferences are less rigid than their swiping behaviour suggests. Platforms that gently expand the range of profiles users see, whilst maintaining high match probability, serve both individual users and broader social goals.

The research on assortative mating ultimately raises a question that extends beyond product design into social policy: should dating platforms be neutral sorting mechanisms that reflect users' existing preferences, or should they be active agents of social mixing that challenge those preferences? The answer likely lies in a middle ground: platforms should respect user autonomy whilst designing defaults and recommendations that expand rather than constrain the range of connections users consider. The research shows that most users' stated preferences are less rigid than their swiping behaviour suggests. Platforms that gently expand the range of profiles users see, whilst maintaining high match probability, serve both individual users and broader social goals.

The assortative mating research raises questions that extend beyond commercial strategy into social responsibility. Dating platforms are not neutral utilities. They are sorting mechanisms that influence who pairs with whom, and therefore who raises children together, who pools economic resources, and how social stratification evolves across generations. The design choices that operators make about filters, recommendations, and defaults carry consequences that outlast any individual subscription. The research does not prescribe specific design choices but it demands that operators make those choices with awareness of their broader implications for the societies their platforms serve.

More articles by DII Editorial