If you've been dating seriously for a while, you've probably had both kinds of first date. The one where the conversation runs long, you're aware of your own pulse, and you walk home already imagining a second date. And the one that was pleasant, competent, and flat.

The natural next question is: how much should I trust this? Is chemistry the thing you should follow, or the thing that keeps misleading you?

That question can't be answered as asked, because "chemistry" is doing two jobs at once. It is a good signal about one thing and weak evidence about another. This piece is about separating them, and about what to watch and ask in the first few dates, while the information you need is still cheap to gather.

What chemistry actually measures

In speed-dating research, where strangers meet for three or four minutes and rate each other immediately afterward, the most consistent predictor of momentary romantic interest is physical attractiveness 4. Reciprocal liking matters too: finding out that someone likes you raises your interest in them.

So the first thing to be clear about is that chemistry is fast and it is about approach. It answers "do I want more of this?" It does not answer "is this good for me?"

Why your first-date read of their character is partly your attraction talking

Here's the problem with using that feeling as a character assessment. Once a global positive evaluation forms, it spreads. Psychologists call this the halo effect, and there's a specific version of it in the first-date literature: the more attracted you are to someone, the more extraverted, agreeable, and conscientious you perceive them to be. In one Japanese speed-dating study, the researchers described physical attractiveness as the first attribute people register, with the rest of the impression forming around it 4.

The cleanest demonstration of this is about the feeling of "we have so much in common." In a speed-dating study that measured actual similarity and perceived similarity in the same way, actual similarity did not predict attraction at all. Perceived similarity did, and the general, holistic version of it — the sense that you and this person have things in common — was an extremely strong predictor of liking 3. So strong that it's best read not as independent evidence but as the same judgment in different words.

That matters practically. "We just clicked," "he really gets me," "the conversation flowed" are not three separate findings about compatibility. They are usually one finding — high attraction in this interaction — restated three ways. The trait-level conclusions you form on date one are contaminated by how much you liked the date.

What the data can and cannot predict

Now the honest accounting. Samantha Joel, Paul Eastwick, and Eli Finkel had speed-daters fill out more than 100 self-report measures before the events, covering traits and preferences that mate-selection researchers had flagged as relevant, then used machine learning to predict how much each person desired each specific partner.

The models did fine at two things: predicting how much a person generally desires others (4–18% of that variance) and how desirable a person generally is to others (7–27%). They could not predict, at all, how much one specific person would desire one specific other person — the compatibility part 1. When the researchers used measures collected after the interaction instead, prediction jumped to 16–29% of that variance.

A second, larger study shows what happens next. Across three speed-dating studies with follow-up surveys (559 participants, over 6,600 dates), both how consensually desirable a date was and how much the participant uniquely liked that date predicted who actually messaged whom and who still felt romantic interest weeks later 2. The general "I'm into everyone" tendency mostly did not.

Put together, these findings say something specific. The part of attraction that is unique to the pair of you is real — it lingers and it drives behavior. But it is not predictable from any list, profile, or pre-meeting measure. It lives in the interaction. Relatedly, the traits you say you want in a partner do not predict which real people you end up drawn to 4.

What follows is not "chemistry is meaningless." It's that chemistry is the only place the dyad-specific data lives, and also that you cannot have the data in advance.

The useful reframe: a signal, not a finding

Treat chemistry as a signal. A signal means: something is happening here, keep collecting information. It is a reason to book a second and third date, not a conclusion about who this person is or whether the two of you fit.

Two consequences fall out of the evidence:

A true negative is informative. If there is no pull on either side after a genuine encounter, that is real information, because initial desire is a practical prerequisite — two people have to want more time together before anything else can happen 1.

A false negative is likely, and so is an inflated positive. The same research showing that compatibility can't be predicted beforehand also means your read after one meeting is a thin sample. That doesn't mean a flat first date reliably turns into something; there's no good evidence for that either. It means one meeting is a low-information encounter, and the fix is more interaction rather than a verdict.

One pattern is worth flagging, since it connects to the last piece. If your strongest chemistry shows up mostly in situations with uncertainty — intermittent contact, a person who is hard to read, a relationship that keeps you slightly anxious — the intensity may be your attachment system's alarm rather than compatibility. Anxiety-driven arousal is real arousal, and it feels like chemistry because your body doesn't label the source.

The one behavior worth watching

If attraction contaminates your read of personality, look for something you can observe as behavior instead, inside the interaction.

The most useful thing here comes from Harry Reis and Phillip Shaver's model of intimacy, tested in daily-diary studies with married couples. Intimacy develops through a cycle: one person discloses something personally relevant, the other responds. What predicts the first person's sense of intimacy is not the disclosure alone but perceived responsiveness — the sense that the other person understood, validated, and cared. Disclosing emotions and feelings predicted intimacy better than disclosing facts 6.

So the thing to watch is what happens after you say something that costs you a little. A real thing: that a friend hurt you, that you're scared about work, that something about your family is complicated.

Watch whether the response contains understanding (a follow-up question that shows they tracked what you actually said), validation (they don't rush to tell you that you're overreacting), and care. Or whether it pivots to their own similar story, offers a quick fix, or moves back to logistics. Also watch what they do with it later — whether the thing you mentioned comes back in a way that shows it was retained.

This one is comparatively trustworthy because you're not rating a trait you can't see. You're watching one observable moment and what follows it.

The one asymmetry worth knowing: what kind of pull is it?

Diane Felmlee's research on "fatal attractions" studied the qualities that first drew people to a partner and the qualities they later disliked most. In about 29% of the ended relationships, the complaint was the negative face of the original attraction: funny became "doesn't take life seriously," easygoing became "never on time," strong became controlling, considerate became "says sorry too much." The risk was higher when the attracting quality was extreme (about three times more likely to become the complaint) and when it was unlike the person's own qualities. Physical characteristics were much less likely to turn fatal 5.

The practical use is a question to ask yourself on date two or three: if this trait were turned up 20%, what would I be complaining about in a year? And a second one: is the pull toward a trait — how interesting, how confident, how spontaneous they are — or toward how they treat me? Felmlee's data suggests the first category is where the reversals live.

Caveat: this is retrospective self-report about relationships that already ended. It describes a pattern, not a prediction about your specific date.

What to ask, and what asking is for

The point of asking is not to extract facts. Facts predict poorly, as the Joel study showed. The point is to create the conditions in which responsiveness becomes observable, and to find out whether the conversation can hold weight.

That means asking how, not whether:

  • Not "what are you looking for?" but "what does a good week look like for you?"
  • "What do you do when you're having a bad month?"
  • "How did your last relationship end, and what was your part in it?" Note that the second half is where the information is — anyone can tell a story in which they were reasonable.
  • "How do you two handle it when you disagree?" — about a friend, a sibling, a colleague. Conflict style with other people is easier to describe than conflict style in the abstract.

And then ask one follow-up question about whatever they just said. Following up is how you find out whether they're curious about you, and it's also how you give them the chance to be responsive to you.

Honest boundary: none of these specific questions is a validated predictor of relationship success. They are instruments for observing a process, and the process — what happens between the two of you — is where the predictive variance lives.

What changes

The first few dates are for collecting interaction data, not for reaching conclusions. Since the dyad-specific information only exists once you're in the room, more encounters with more people will tell you more than deeper analysis of profiles, and a checklist cannot validate or invalidate what you feel.

One limit worth stating plainly: most of this evidence comes from short speed-dating interactions, largely with college and community samples in Western countries. It measures desire, not durability. It cannot tell you who to marry.

Chemistry tells you to keep walking. It doesn't tell you what's in the room.