Why swiping doesn't work: the evidence

Not an opinion piece. A reading list with a conclusion.

In 2017, researchers Samantha Joel, Paul Eastwick, and Eli Finkel ran the experiment every dating app hopes you never read about. They took more than a hundred self-reported traits and preferences from people about to speed-date — the exact kind of data apps feed their algorithms — and pointed machine learning at one question: can we predict which two specific people will want each other? The answer, published in Psychological Science: essentially no. Around zero percent of pair-specific desire was predictable before two people met. You can predict who is generally pickier, who is generally desired — but not the spark between this person and that one. Nobody can. That's the entire premise of compatibility-score matching, gone.

It gets worse for the swipe itself. Strangers largely agree on who's attractive — that's what a swipe measures. But Eastwick and Hunt showed in 2014 that this consensus collapses as people actually get to know each other; with acquaintance, evaluations become unique to the evaluator. Swiping harvests exactly the signal that real life erases. And the interface trains your thumb toward no: Pronk and Denissen (2020) found acceptance rates drop roughly 27% from the first profile in a session to the last — they called it the "rejection mindset."

The lived experience matches the lab. Pew Research (2023): only about 1 in 10 partnered US adults met their current partner on an app, while among recent users, the overwhelming majority report regularly feeling disappointed, and most women rate the overall experience negative. Industry surveys put recent users reporting burnout near 4 in 5.

Here's the part the burnout headlines miss: meeting online doesn't doom anyone. Stanford's HCMST data (Rosenfeld et al.) finds couples who met online break up at about the same rate as anyone else. The marriages are fine. The process is the problem — a slot machine that pays out for a minority while teaching everyone else to reject and be rejected at scale.

What does work is almost embarrassingly unglamorous: repeated exposure to the same small group (the most replicated finding in attraction research), accumulated leisure hours (about 50 to a casual friend, 90 to a real one), novel shared activities, and gradually deepening conversation. None of it can be swiped. All of it can be scheduled.

So that's what PartyUp does: same 5–8 people, same table, every week, with a little adventure built in. We can't predict chemistry — the science says no one can. We can build the room it happens in.

Sources: Joel, Eastwick & Finkel 2017 (Psychological Science); Eastwick & Hunt 2014 (JPSP); Pronk & Denissen 2020 (SPPS); Pew Research Center 2023; Rosenfeld, Thomas & Hausen 2019 (PNAS); Hall 2018 (J. Social & Personal Relationships). Full bibliography: the research page.