Dating Apps' Age Filter Dilemma: Stated Preferences vs. Real Attraction

Dating apps may be optimising for the wrong signals. Women prefer younger partners in blind dates despite consistently telling researchers they want older men, according to new research that exposes a fundamental gap between stated preferences and revealed attraction. The finding carries direct commercial implications for platforms that have spent two decades building recommendation engines on self-reported filter data.

A University of California, Davis study of 4,500 speed-dating encounters found that women, like men, showed marginally higher attraction to younger partners. Women showed a 3% higher likelihood of being attracted to partners younger than themselves compared to older ones. Men showed a 4% preference for younger partners. The statistical difference between the sexes was negligible.

Yet when surveyed about their preferences before the dates, women stated they preferred older partners. The gap between stated preference and revealed attraction is not trivial for an industry that has architected its core product around what users say they want. The finding, published in Proceedings of the National Academy of Sciences, comes from speed-dating sessions where participants rated their attraction to each partner after brief conversations.

The DII Take

If stated preferences systematically misrepresent actual attraction, particularly for women, who are likelier to convert matches into conversations, then platforms are optimising for the wrong outcome.

This matters because dating apps are architected around what users say they want, not what they respond to in practice. The commercial implication is direct: lower match quality means lower engagement, which means weaker retention and worse unit economics. Apps that can surface revealed preference through behavioural data rather than filter selections have a structural advantage.

Why the gap exists

The researchers posit that social norms around male provisioning and status drive women's stated preferences for older men, even when immediate attraction pulls in a different direction. Women may internalise the expectation that suitable partners should be older, more established, or more financially secure, criteria correlated with age, and report preferences accordingly. When those filters are removed in a blind date setting, where age is visible but biographical data is minimal, attraction patterns align more closely with men's.

The finding echoes earlier research showing that when women are primed to think about status or long-term compatibility, their stated age preferences skew older. Remove those primes, and preferences flatten. Dating app operators have long observed a version of this dynamic.

Match Group (MTCH) executives have noted in earnings calls that user behaviour often diverges from stated filters, with members engaging with profiles outside their declared age ranges at meaningful rates. Bumble (BMBL) reported in its 2023 annual user survey that 62% of women said they preferred partners at least two years older, yet swipe data from the platform showed women engaging with younger men's profiles more often than those stated preferences would predict.

Hinge, which positions itself as the relationship-focused alternative to swipe apps, removed age filters entirely from its free tier in 2021: a decision the company framed as encouraging users to be 'open-minded' but which also served to bypass the stated-preference problem. According to Hinge's product disclosures, conversation rates increased 15% after the change, suggesting that reducing user control over age filtering improved match quality rather than degrading it.

What this means for matching algorithms

Dating platforms face a basic architectural question: do they match people based on what users say they want, or what they appear to want based on behaviour? Most apps use a hybrid approach, weighting stated preferences but adjusting for engagement signals: swipes, likes, message rates, conversation length. The tension is commercial as well as technical.

Restricting user control over filters (as Hinge did) risks alienating paying subscribers who expect granular search functionality as part of premium tiers. Yet surfacing profiles that contradict stated filters risks appearing broken or unresponsive to user input. Apps that lean heavily into behavioural signals: Tinder's Elo-style desirability scoring, The League's engagement-based ranking: already deprioritise stated preferences in favour of revealed ones.

The UC Davis study suggests that behavioural approaches may be closer to predictive accuracy than filter-first matching, at least for initial attraction.

The challenge is converting initial attraction into relationship formation. The researchers acknowledge their data captures only first-impression chemistry, not compatibility or longer-term interest. Real-world couple formation data consistently shows men as older in heterosexual pairings: by a median of 2.3 years in the UK, according to Office for National Statistics figures, and 2.1 years in the US, per Census Bureau data.

That gap between blind-date attraction and ultimate pairing suggests other factors override initial chemistry. The researchers flag male initiation patterns as one explanation: if men disproportionately approach younger women, and women are likelier to respond to inbound interest than to initiate, then age distributions in formed couples will skew male-older regardless of women's attraction patterns.

Economic factors matter too. Women's stated preference for older partners correlates with preferences for financial stability and career establishment: traits that, in aggregate, correlate with age. A 25-year-old woman may feel equal attraction to a 23-year-old and a 28-year-old in a blind date, but consider only the latter a viable long-term partner if she's filtering for income or career stage.

The reliability question

The study is large by academic standards, 4,500 dates across multiple speed-dating events, but it's a single dataset from one US university town. Replication across geographies, age cohorts, and socioeconomic groups would strengthen confidence in the findings. Speed dating also selects for people willing to participate in structured romantic evaluations, a group that may not represent the broader singles market.

More fundamentally, the study measures initial attraction in a controlled setting with minimal information exchange. How that translates to app-based matching, where users see photos, bios, and often occupation and education data before deciding to engage, is not clear. Age may function differently as a signal when it's isolated versus when it's bundled with other status markers.

Dating apps operate in the bundled-signal environment, not the blind-date one. That makes direct application of these findings to algorithmic design non-obvious. But the directional insight, that stated preferences are noisier signals than behavioural ones, reinforces a shift already underway in product development. Apps are moving toward less user control and more algorithmic curation, trading autonomy for match quality.

Regulatory pressure is pushing the other direction. The EU Digital Services Act (DSA) requires platforms to provide users with at least one non-algorithmic ranking option, and the UK Online Safety Act (OSA) mandates transparency in how algorithmic recommendations work. Compliance teams should track whether revealed-preference matching raises explainability challenges under those frameworks, particularly if apps surface profiles that directly contradict user-set filters.

The next product generation will likely split the difference: more behavioural weighting in recommendations, but clearer explanations of why profiles are surfaced. Expect to see 'we think you'll like this' carousels alongside traditional filter-based search, with the former quietly driving more engagement than the latter.