What Builds Attraction Over Time: Why Who Seems Desirable Changes With Acquaintance

First, what this page used to say

This article was published under the title nine quiet power moves that build real attraction, and the promise was that a reader could deploy them to produce attraction in someone else. It was rewritten in September 2026 because that frame treats another person as a mechanism with inputs, and the change is logged on our corrections page rather than made quietly.

That rewrite fixed the framing and left the page thin: about 2,200 words, the shortest article on this site, resting on five references, none of which supported a specific sentence in it. One of those references was reported badly enough to count as an error. The article said positive emotions make it easier to include another person in the self, then described the study behind it as having hypothesised exactly that, and never said what it found. It did find it, and the numbers are below; a hypothesis presented as a conclusion is a defect whether or not the conclusion turns out to be right.

This is the second pass. Every empirical claim below has a study behind it in the sources list, the numbers are quoted from the papers, and where the evidence is weak or a famous result has failed to replicate, the article says so instead of rounding it up. That includes the central claim of this page: the best-known experiment supporting it produced null main effects the last time somebody preregistered a version of it.

Two different questions are wearing the same words

If you are reading this because a relationship has gone quiet and you are trying to work out whether that means something is wrong, a conversation with a therapist or counsellor is a reasonable step and not an overreaction. Nothing here is a diagnosis, and none of it is a substitute for one.

With that said: what builds attraction over time is asked by two different people who want two different things.

The first is asking whether attraction to a particular person can grow from little or nothing as you get to know them — whether the absence of a spark on date one settles the question. The second is already in a relationship and wants to know what keeps attraction from draining out of it.

These have separate research literatures and separate answers, and the genre routinely quotes one at the other. The first question has a clean, surprising, well-measured answer that has nothing to do with anything you can perform. The second has a decent answer with a caveat attached, and a much larger body of evidence about what fails.

One boundary before starting, because it constrains everything that follows. The largest study ever run on relationship quality found that whether a given relationship will get better or worse from here is essentially not predictable from anything either person can report about it. Nothing on this page is a forecast about your relationship. It is a description of what moves the average, and averages are not people.

Agreement about who is desirable decays with acquaintance

Start with the finding the first question turns on, because once you have it the rest of this literature stops looking like a list of tips.

When a group of people rate how romantically desirable they find each other, those ratings split into two kinds of variance. Target variance is consensus: everyone agrees this person is attractive. Relationship variance is uniqueness: this particular rater finds this particular person appealing, and nobody else does to the same degree.

Eastwick and Hunt measured that split three ways — asking people what mate value means to them, running a social relations model on classmates rated at the start and the end of a semester, and running a one-with-many design on long-term acquaintances. Across the studies, there was more relationship variance than target variance: romantic evaluations were more likely to be unique to a particular perceiver than shared. The dominance of uniqueness was strongest for judgments about whether someone would make a good relationship partner, and weaker for the classic trait ratings — attractiveness, resources — where consensus does better. And consensus decreased as participants got to know one another, with long-term acquaintances showing, in the authors' words, enormous amounts of relationship variance.

The follow-up paper puts the same result in percentage points. Consensus exceeded uniqueness by two points when raters had known each other for two weeks. Uniqueness exceeded consensus by ten points at fourteen weeks, and by thirty-seven points among well-acquainted others.

Why it moves: longer acquaintance means more contexts and more chances to disagree about what a behaviour means. One person is impressed by the performance onstage; another finds the same performance pretentious. Early on, the raters have seen the same thin slice and produce the same ranking. Later they have seen different things.

That is what growing on someone actually is, stated as a variance decomposition rather than as encouragement. It is also why reading a first impression is a much weaker instrument than it feels, a point worked through in what body language actually predicts.

What that decay does to real couples

If consensus is what powers competition — everyone chasing the same people, and sorting into pairs of similar desirability — then a decay in consensus should show up as a decay in how well couples are matched. Hunt, Eastwick and Finkel tested exactly that.

They took 167 couples (67 dating, 100 married) from a longitudinal study, asked how long the partners had known each other before they started dating, and had trained coders rate each person's physical attractiveness from video. The coders were reliable, and the overall matching correlation was .38 when each partner was rated alone with the other half of the screen covered — close to the meta-analytic average of about .39 for this effect.

The result is in the interaction. The longer the couple had known each other before dating, the weaker the matching: b = −0.16, t(163) = −2.53, p = .012 for the separate ratings, and b = −0.21, t(163) = −3.15, p = .002 for the joint ones. Couples who started dating within a month of meeting had a predicted matching correlation of .53 to .72. Past roughly nine months of prior acquaintance, matching on attractiveness was no longer distinguishable from zero.

The friends-first comparison points the same way. Couples who had not been platonic friends before dating showed a matching correlation of .67; couples who had been friends first, .43. That is the measured version of a pattern discussed from the other side in why people keep ending up in the friend zone — a period of friendship is not a failed romance so much as a different selection process with different results.

Two honest limits, because they matter. First, the authors treated prior acquaintance longer than 19 months as an extreme value and set those 20 couples aside; when a robustness check kept them capped instead, the joint-rating interaction dropped to p = .084 and the separate-rating one to p = .153. The finding is real but it is sensitive to that decision, and the paper says so. Second, this is one mostly White, mixed-sex American sample.

And one result worth keeping: in this sample, how well a couple matched on attractiveness had no association with their relationship satisfaction, for either partner.

The largest attempt to say what makes a relationship good

For the second question — what keeps attraction alive once a relationship exists — the best single starting point is the biggest study of its kind.

Joel, Eastwick and 84 co-authors pooled 43 dyadic longitudinal datasets from 29 laboratories, covering 11,196 couples, and ran machine-learning models (random forests) to ask which of the hundreds of variables relationship science has proposed actually predict how satisfied people are.

The top relationship-specific predictors were perceived-partner commitment, appreciation, sexual satisfaction, perceived-partner satisfaction and conflict. The top individual-difference predictors — life satisfaction, negative affect, depression, attachment avoidance, attachment anxiety — also performed respectably on their own.

Then the finding that reorganises the field. Relationship-specific variables explained up to 45% of the variance in current satisfaction. Individual differences explained up to 21%. But adding individual differences and partner reports to a model that already had the person's own relationship-specific variables improved nothing. Whatever a personality trait or a partner's own experience does to a relationship, it appears to do it through how the person experiences the relationship they are in.

It is also the reason no questionnaire filled in before two people meet can forecast this, which is the structural problem behind whether a dating app's matching does anything: every variable that carried weight here is a judgement about a relationship that already exists.

That is an argument against the entire premise of self-improvement-as-attraction-strategy, and it is made with 11,196 couples rather than an anecdote. The traits are not where the variance lives. Which of your attachment patterns you bring is worth knowing — our attachment style test is built on the standard two-dimensional measure — but it is not the thing that decides how a relationship feels.

And the limit, stated by the authors: predicting change was a failure. No model accounted for more than 5% of the variance in whether satisfaction rose or fell over the study period, typically one to two years, from any combination of self-report variables. Whether things get better from here is not visible in what people can tell you today. If you want a snapshot of where a relationship currently sits rather than a forecast, that is what our relationship satisfaction test is for, and it says the same thing on its own page.

Self-expansion: the model that predicts the direction

The theory that organises most of the rest is Aron and Aron's self-expansion model. Its core claim is that people are motivated to expand the self — to increase their resources, perspectives and identities — and that close relationships are one of the main ways this happens: another person's resources and viewpoints start being experienced as partly one's own. The literature review of this work traces the same principle through perspective taking, social comparison, humour and even how physical pain is shared between partners.

What makes it the right frame for attraction over time is that it predicts a direction, and the direction is unflattering to the advice industry.

Early attraction can run on novelty and appearance, because at that stage there is nothing else to run on. Sustained attraction, on this account, runs on whether being with someone keeps making your world bigger. That is not a claim that you should perform interestingness. It is a claim that expansion requires material, and that a person with a life — work that matters, friendships that are not joint, some difficulty of their own — has material to include someone in.

It also explains the most common complaint in long relationships without needing anything to have gone wrong. If both people's worlds have gradually contracted to the shared one, there is nothing left to include, and the loss of the feeling gets attributed to the relationship rather than to the absence of new material. The authors of the experiments in the next section say precisely this: once two people know each other fairly well, the opportunities for rapid expansion from getting to know each other inevitably decrease.

One measured detail about how the inclusion happens, since this page previously asserted it without reporting it. Waugh and Fredrickson followed 147 first-year college students prospectively, with daily emotion reports for 28 days, and tracked self-other overlap with their new roommate. Students reporting more positive emotions during their first weeks reported greater overlap, and the association survived controls for extraversion and negative emotions (β = 0.25). Prospectively, the ratio of positive to negative daily emotion predicted overlap a month later controlling for overlap at the first measurement (β = 0.17, p = .05), and overlap in turn mediated the path to a more complex understanding of the other person. The caveat is that these are randomly assigned roommates, not romantic partners; the everyday reading is not be more cheerful but that the emotional weather of the time two people spend together shapes how much of each other they take on.

The habit side of this is covered in which maturity behaviours actually have evidence behind them, and the same mechanism observed from the far side — what happens when separation stops supplying novelty — is in what distance actually does to attraction.

The seven-minute experiments, in detail

The evidence usually cited for all of this is a single 2000 paper with five studies in it, and it is worth reading properly rather than through the version that reaches magazines.

The two surveys. A questionnaire printed in a Santa Cruz newspaper produced 112 usable responses; a door-to-door survey in the same town produced 80. In both, how exciting people rated the things they did with their partner correlated with relationship satisfaction (r = .51 and β = .45), and in both the association survived a control for socially desirable answering. In both, it was mediated by relationship boredom: with boredom in the model, the exciting-activities path fell to .10 and .22, while boredom held a strong path of its own.

The three experiments. Couples came into a laboratory, filled in a pretest, then did a task. In the novel and arousing condition they were strapped to each other at the wrist and ankle with Velcro and had to cross nine metres of gym mats on hands and knees, over a rolled-up mat barrier, carrying a cylindrical pillow between them without using hands or teeth. In the mundane condition, one partner crawled to the middle of the mat and back, then the other did the same. Seven minutes.

Study 3 (28 couples): the novel-arousing group showed greater improvement in experienced relationship quality, F(1, 24) = 6.07, p < .05, partial r = .45. Study 4 (63 married couples) replicated it and added a no-activity control, which is the study that matters. Novel-arousing beat mundane, and it also beat doing nothing, partial r = .30 — so the effect was an increase in the novel condition, not a decrease in the boring one. Study 5 (35 married couples) found the same pattern in blind ratings of videotaped discussions before and after: less hostility, more acceptance and support.

Seven minutes, in a laboratory, measured minutes later. That is the whole evidence base for the claim that doing something new together does something, and it is a good deal smaller than the way it is usually reported. It is also the only real seven in this genre, which is worth knowing when you meet the numbered dating rules with no research behind them.

The authors list three limitations themselves: the laboratory paradigm has limited ecological generalisability, the samples are entirely American, and the work demonstrates an effect without establishing the mechanism. They name misattribution of arousal, shared positive affect and simple cooperation as alternatives they cannot rule out.

Exciting, not merely pleasant: the one field test

There is one randomised field experiment in this literature, and its result is more specific than the headline.

Reissman, Aron and Bergen randomly assigned 53 married couples to spend an hour and a half a week for ten weeks doing activities they themselves defined as exciting, or activities they defined as pleasant, or to a no-special-activity control. Satisfaction was measured before and after, adjusted for socially desirable responding.

Two contrasts, and both are informative.

The exciting group ended up significantly more satisfied than the pleasant group, with a moderate effect size. But the contrast comparing both activity groups against the control was not significant, and its effect size was small. In plain terms: doing more things together did not itself raise satisfaction over ten weeks. What separated the groups was what kind of thing.

This is the result that kills the most common version of the advice. Spend more quality time together is not what was tested and not what was found. The operative variable is whether the activity adds something — a capability, a perspective, an experience neither person had — and a restaurant you have not been to before is a change of scene rather than an addition. Learning something at which one or both of you are visibly bad is the kind of thing the model predicts will do the work, and the discomfort is not a side effect of it.

It is also the reason the date-night literature is weaker than it is sold as. The best survey evidence for regular couple time is cross-sectional, and when its authors added communication, sexual satisfaction and commitment to the model the association with couple time disappeared entirely — a result unpacked in the 3-6-9 rule and the other invented numbers.

Where the effect does not replicate cleanly

Everything above was run before preregistration was normal. The honest question is what happens when somebody preregisters it, and the answer is mixed enough to belong in the article rather than in a footnote.

Balzarini, Sharma and Muise ran two preregistered laboratory experiments — 93 couples and 142 couples — using shared novel activities in virtual reality. Study 1 compared a novel VR experience against watching video; Study 2 compared novel against mundane VR.

The manipulations worked on some outcomes. VR increased the sense of the partner being present and reduced boredom compared with video. Novel VR produced less boredom and greater closeness than mundane VR.

And then: no main effects on self-expansion or relationship quality in Study 1, and none on self-expansion or relationship satisfaction in Study 2. Effects on self-expansion appeared only when immersion was controlled for, with downstream effects on relationship quality from there.

How much this counts against the original work is a real question rather than a rhetorical one. The authors' own reading is that virtual interaction may have less potential than in-person interaction to produce self-expansion, which is plausible: crossing a gym mat strapped to another human is not the same class of event as sharing a headset. But two preregistered studies with 235 couples between them moved boredom and closeness while leaving satisfaction alone, and that is a narrower result than the one in circulation.

What survives is the pattern rather than the promise. Shared novelty reliably moves boredom. Whether it moves the thing you actually care about is supported by older, smaller, unpreregistered work and by one field experiment, and is not yet nailed down. This site's editorial policy commits to saying that where it is true, and it is true here.

Boredom is the early warning, and it shows up nine years later

The boredom half of this literature is better evidenced than the novelty half, partly because it does not require an experiment.

Tsapelas, Aron and Orbuch used a long-running panel of married couples in the Detroit area — same-race couples in first marriages, with the wife under 35 at the start — and looked at 123 couples interviewed at year seven and again nine years later, at year 16.

The boredom measure was one question: during the past month, how often did you feel that your marriage was in a rut, or getting into a rut, that you do the same thing all the time and rarely get to do exciting things together as a couple.

Greater boredom at year seven predicted significantly less satisfaction at year 16, and the association ran through closeness: boredom appears to undermine closeness, which in turn reduces satisfaction.

Two things to hold about this. It is longitudinal, which is much better than cross-sectional, but it is not an experiment — nobody assigned anyone to be bored, and a marriage that is quietly going wrong for other reasons will also feel like a rut. And it is one item, in one region, in one cohort of first marriages.

What it adds is a time horizon. The thing people notice as the spark went is measurable years before it becomes a complaint, and it is measurable with a question about whether anything new happens rather than a question about the relationship. The behavioural side of the same finding — what couples who stay together actually do — is in the habits of couples who stay in love.

Nearly half of what a partner does is not noticed

Appreciation came second on the list of the strongest predictors in the 11,196-couple project. The study that shows what that looks like day to day is smaller and much more specific.

Algoe, Gable and Maisel gave 67 cohabiting couples nightly diaries for 14 days — 1,768 completed reports — asking, among other things, whether their partner had done something thoughtful for them that day, whether they had done something thoughtful for their partner, and how grateful and how indebted they felt.

The first number is the one to sit with. Of the days on which a partner reported doing something thoughtful, the other person agreed 51.2% of the time. Nearly half of the thoughtful things people reported doing went undetected by the person they were done for.

The second is what gratitude then did. Feeling grateful from interactions on one day predicted increases in relationship connection and satisfaction the next day, for both men and women, controlling for the previous day's level — and it predicted the same increases in the partner, the person who had done the thing. Feeling indebted did not; indebtedness predicted no improvement in relationship quality for either sex.

The design allows the reverse test, and the authors ran it: yesterday's relationship satisfaction did not predict today's gratitude. The arrow points the way it is usually claimed to, which in this literature is not something to take for granted. The effects also survived controlling for admiration, so this is not simply positive feeling generally.

The practical residue is unglamorous and it is not a technique. About half of what a partner does for you is currently invisible to you, on average, and noticing it out loud has a measurable next-day effect on both of you. Saying what you noticed rather than what you concluded is also the single most transferable finding in how to get someone to open up.

What happens when something goes right

There is a second, separate line of work on the same theme, and it is about good news rather than bad.

Gable, Reis, Impett and Asher ran four studies on what they called capitalization: telling someone about a positive event. Two showed that communicating a personal positive event was associated with increased daily positive affect and well-being above and beyond the impact of the event itself and other daily events. Two more showed that close relationships in which one's partner typically responds enthusiastically were associated with higher relationship well-being — intimacy, daily marital satisfaction.

The part that matters for this article is the moderator. The benefits were enhanced when others were perceived to respond actively and constructively, and not when they responded passively or destructively. The good news does less work than the response to it.

This is worth stating plainly because the ordinary intuition about supportive partners is entirely about bad days. Being present for someone's difficulty is not in dispute. What this work adds is that the response to someone's good day carries its own weight, and that the failure mode is not hostility but flatness: hearing it, registering it, moving on.

It connects back to the self-expansion account without much strain. Someone else's good news is new material entering the relationship, and an enthusiastic response is what includes it. A passive one leaves it outside.

These are American self-report and diary studies like most of this literature, and the associations are correlational in the two relationship studies. The claim is not that enthusiasm manufactures satisfaction; it is that the two travel together, and that the direction has held up in the day-to-day designs that can test it.

The partner who makes you more like who you want to be

The strongest version of attraction grows where a person adds to who you are is not a metaphor, and it has a name.

Drigotas, Rusbult, Wieselquist and Whitton described the Michelangelo phenomenon: close partners sculpt each other's selves, and the relevant variable is partner affirmation — the degree to which a partner perceives you, and behaves toward you, in ways congruent with your own ideal self. Across four studies with different designs, perceived partner affirmation was strongly associated with movement toward the ideal self.

The design that makes it credible is the friend-as-observer one. Rather than relying on a single person's report, pairs of friends each rated the same thing — one as the target ratings their own partner's affirmation and their own progress, the other as an observer rating both from outside. The associations held across friends: when a friend described someone's partner as highly affirming, that person reported more movement toward their ideal self.

Later work adds a behavioural test. Married couples were videotaped discussing a goal relevant to each person's ideal self; trained coders scored how affirming the partner was — helping clarify plans, offering assistance, praising the pursuit. Four months later, targets whose partners had been more affirming were more likely to have achieved the goals they had discussed.

The distinction the researchers insist on is the one this genre would blur. Affirmation is not enhancement — seeing someone more positively than is warranted, which is a separate and separately beneficial phenomenon. Affirmation is specific to the ideal that person actually holds. Flattery does not qualify, and neither does encouraging someone toward a version of themselves you would prefer. The model's negative case is explicit: a partner can also disaffirm, by communicating indifference or pessimism, by undermining the pursuit, or by affirming qualities antithetical to the person's ideal self.

Which is the mechanism this page's hook points at, and the reason it is not advice about being impressive. The markers of this in a new relationship, such as they are, are collected in the signs a relationship has real potential.

Does it have to fade?

The assumption underneath the whole question is that attraction declines and the best you can do is slow it. That assumption is half right, and the half that is wrong is the interesting one.

Acevedo and Aron examined a large dataset of community couples and found that romantic love and obsession loaded as independent factors — that intensity, engagement and sexual interest are separable from the preoccupied, anxious, checking component that dominates early attraction. A subsample reported very high romantic love without obsession, and that held after controlling for socially desirable responding.

They then meta-analysed 25 studies. Romantic love without obsession was strongly associated with relationship satisfaction in both long-term and short-term relationships. Obsession behaved differently depending on the stage: positively correlated with satisfaction in short-term relationships, negatively correlated in long-term ones.

So the thing that reliably fades is not the thing people mourn. What fades is the obsessive component, and in a long relationship its persistence is associated with things being worse rather than better. The component that does not have to fade is the one people are usually describing when they say the spark went.

This matters for reading your own experience. Intensity that consists of not knowing where you stand, checking, and being unable to think about anything else is a different construct from intensity that consists of wanting to be around someone. The first is what manufactured uncertainty produces, and it is the dynamic examined in love or anxious attachment and in the psychology of jealousy. Mistaking it for the good kind is the most expensive error in this whole subject.

The usual caveats: this is a review plus a meta-analysis of self-report studies, and cross-sectional comparison of long- and short-term couples cannot separate what time does from who stays. The companion question — how far and how fast satisfaction itself declines, and why the trajectory studies say most couples do not fall off a cliff — is taken apart in why the honeymoon phase ends.

What does not build attraction over time

A shorter list, and it is better evidenced than most of the positive claims.

Playing hard to get. The classic paper is worth reading for its structure alone: Walster, Walster, Piliavin and Schmidt ran five experiments trying to show that people value hard-to-get partners more, and all five failed. The sixth, which succeeded, tested something different — a woman who was easy for the participant to get but hard for everyone else. Participants preferred that over uniformly hard to get, uniformly easy to get, and no information. The finding that survives is selective availability toward one person, which is close to the opposite of the strategy the paper's title gets cited for. Dai, Dong and Jia later separated the two responses: playing hard to get increased wanting and decreased liking, and only when the person was already psychologically committed to pursuing. Otherwise it decreased both.

Being someone's stated type. The meta-analysis of ideal partner preferences, spanning 97 studies, found physical attractiveness predicting romantic evaluations at about r = .40 and earning prospects at about r = .10 — for both sexes, with a sex difference of r = .03 that was uniformly non-significant. Stated preferences do much less work than their prominence suggests, which is the starting point of why smart people keep dating the wrong people.

Mirroring and body-language technique. The seminar version rests on misreadings, of which the 7-38-55 figure is the most famous, and what survives careful analysis is far narrower. The detail is in what body language actually predicts and, for phones specifically, in what phone behaviour on dates actually shows.

Timing rules. Waiting three days, texting on a schedule, the numbered cadences: folklore delivered confidently. The 72-hour rule traces where that one came from, and the texting habits that actually push people away covers what the evidence supports instead.

Being matched on looks. In the 167-couple study above, how well partners matched on attractiveness had no association with either partner's satisfaction.

The common thread is that all of these treat attraction as something produced in another person by managing the signal you send. Every well-evidenced finding on this page is about what the relationship does to both people.

When the world getting smaller is not your choice

This section exists because the central idea of this article has a failure mode that looks identical from the outside to something much worse, and the difference is not a matter of degree.

The self-expansion account says that relationships stall when both people's worlds contract to the shared one. Read carelessly, that sits uncomfortably close to a description of what a controlling partner produces deliberately.

The difference is who did it. Two people who drifted into having no separate friendships have a problem they can both see and both work on. One person being cut off from friends and family, a little at a time, is not that. UK statutory guidance on controlling or coercive behaviour names isolating a person from sources of support, monitoring their time and movements, controlling access to money, and depriving them of independence, as recognised markers of that pattern.

The practical reason for naming it here is that the self-expansion frame invites the wrong response. If the problem is read as we have got boring, the remedy looks like doing something new together — which puts more of the person's time inside the relationship, with the other person, which is the direction the control was already pushing. Nothing on this page applies to that situation, and no shared activity fixes it.

Some markers that the line has been crossed, stated behaviourally: your seeing someone reliably produces a cost afterwards; you find yourself accounting for hours; access to your phone or your location is required and is not mutual; interests of your own are treated as evidence of disloyalty; the withdrawal or the anger arrives as a sanction. The version of this specific to phones is in what phone behaviour on dates actually shows, and the boundary between jealousy as a feeling and control as a behaviour is worked through in the psychology of jealousy.

If any of that is recognisable, the useful next step is a person rather than an article. Find A Helpline, in the sources below, lists free and confidential services by country. In the United States the National Domestic Violence Hotline is 1-800-799-7233, or text START to 88788, 24 hours a day.

What follows from all of this

Four things, and none of them are moves.

Have something to bring. The self-expansion model requires material, and the 11,196-couple project found that no combination of personal traits predicted satisfaction beyond a person's own experience of the relationship. Both point the same way: a life with things in it is not a way of becoming more attractive, it is the raw material the process runs on. Its absence is what leaves a relationship with nothing to include.

Do things that add something, not things that are merely different. The only randomised field test in this literature found that exciting activities beat pleasant ones, while activities in general did not beat doing nothing. The operative variable is addition — a capability, a perspective, a context neither of you had. Being visibly bad at it for a while is part of the design.

Say what you noticed. Nearly half of what a partner does goes undetected, and noticing it out loud predicted next-day improvements for both people in the only design that can test the direction. The same applies to good news: the response does more work than the event.

Watch for the rut rather than for conflict. One question about whether anything new happens predicted satisfaction nine years out. That is a more sensitive early indicator than whether you are arguing, and it points at something you can act on.

And one thing not to do, which is the reason this page was rewritten twice. None of this works as behaviour performed to generate a result in another person. The measured effects are all effects on both people at once, and the strategies aimed at producing attraction in someone else — engineered scarcity above all — are the part of this literature with the clearest record of failure.

What this evidence cannot tell you

The limits go here rather than in a footnote, because on this subject they are most of the honest answer.

Change is not predictable. In the 11,196-couple project, no model accounted for more than 5% of the variance in whether satisfaction rose or fell over one to two years, using every self-report variable 29 laboratories had collected. That is the best-resourced attempt anyone has made, and it says that the trajectory of a particular relationship is largely not visible from the inside today.

The central experiment is old and thinly replicated. The seven-minute studies are three laboratory experiments with 28, 63 and 35 couples, run in the 1990s at one American university, before preregistration. Their authors name the ecological limits, the American samples and the unestablished mechanism themselves. The one preregistered attempt at the same idea in a new medium found the boredom effect and not the satisfaction effect.

The samples are narrow. American undergraduates and community couples, overwhelmingly White in several of the key studies, almost entirely mixed-sex. The 43-dataset project is mostly North America, Western Europe and Israel, and its predictors are 99% self-report rating scales rather than observed behaviour.

One key result is sensitive to an analysis decision. The acquaintance-and-matching finding weakens to non-significance on one of its two measures if the 20 outlying couples are capped rather than excluded.

And there is no measured method for making a specific person become attracted to you. Not in this literature, and not in the parts of it that get sold as one. What the evidence describes is a process that either has material to run on or does not, observed on average, across people who are not you.

Every empirical claim on this page carries a linked source, which is the promise made in our editorial policy; claims this site has had to retract are listed on the corrections page. The rest of this section is under Mindset.