Key Takeaways

  • Algorithms trained on self-reported dating preferences predict exactly zero percent of mutual attraction between two people.
  • Speed-dating experiments run by Paul Eastwick show that people who meet someone with their three ideal traits like them, but they like a person with three completely different positive traits just as much.
  • In a study led by Samantha Joel, machine learning models given full profiles of personal values, preferences, and traits still failed to forecast compatibility.
  • Predicting attraction requires expanding past three conscious deal-breakers to at least 35 simultaneous traits processed beneath conscious awareness.

Checklists Predict Zero Percent of Real Attraction

Ask someone what they want in a partner, and they hand you a neat list. They want someone funny, ambitious, and kind. They believe those specific traits explain who they fall for. The data shows this belief is fiction.

Psychologist Paul Eastwick tested this assumption across speed-dating experiments. He asked participants to list their top partner preferences, then tracked who they liked in person. The results broke the standard model of romantic selection. As Eastwick explained: “So we do studies where we introduce folks to potential partners. And it's true, if I introduce you to somebody who has those three qualities, you're probably going to like her more. But the tricky part is this: if instead, she had three different qualities, she was especially attractive, and sociable, and a really supportive, kind person, you would have liked her just as much, even though those weren't the three things that you listed.” (play from 21:01)

In other words, people love positive traits in general, but they have almost no introspective accuracy about which specific traits drive their attraction. As Eastwick put it, “the things that we say we're looking for don't match up all that well with the traits that actually appeal to us once we meet folks face-to-face.” (play from 21:36)

The breakdown became undeniable in a study led by Samantha Joel. Researchers gathered exhaustive self-reports covering personalities, political values, and personal standards. They fed the data into machine learning algorithms to predict which pairs would feel mutual chemistry. The predictive power was non-existent. Eastwick summarized the outcome bluntly: “And what we found was that when it comes to compatibility, we could predict exactly 0% of what makes two people compatible. It is not possible to do with troves of self-reported information about who you are.” (play from 36:19)

The Brain Weighs Thirty-Five Traits at Once

The failure of preference lists does not mean attraction is random. It means human evaluation is mathematically too dense for conscious checklists. When you look at an upfront profile, you focus on three or four salient attributes. But the moment you sit across from a living person, your brain processes dozens of nonverbal and verbal signals at the same time.

Eastwick pointed out the mathematical gap: “part of the issue is that it's computationally very complex to think about all the attributes that somebody might have. If we look at more than somebody's three deal-breakers, and we expand it out to 35 traits, we can start to make a little headway on why you like one person, but not somebody else.” (play from 23:10)

Static profiles invite you to act like an auditor grading a scorecard. In reality, human connection is dynamic. You cannot pre-filter your way into chemistry because the weights your brain assigns to each trait change the second interaction begins. The spreadsheet never survives contact with the person.

What to Do With This

Take your current hiring rubric or co-founder wish list and cut the requirements down to non-negotiable ethics. Stop disqualifying candidates on paper because they lack one specific background trait you thought you needed. Schedule a live, 60-minute working session with the next applicant who looks unusual on paper, and let real-time chemistry override your static checklist.