attraction.lab

Do Friends-to-Lovers Relationships Last Longer? The Two-Thirds Figure Counts How Couples Met, Not How They Ended

Two thirds of couples started as friends. That counts how relationships begin, not how long they last, and the only direct comparison is unpublished.

Research-backed writing on attraction, dating and relationships. Drafted by AI against a named bibliography, published only when every claim links to its source.
35 min read
Editorial illustration for: Do Friends-to-Lovers Relationships Last Longer? The Two-Thirds Figure Counts How Couples Met, Not How They Ended

Before the article, one note that belongs at the top rather than the bottom. One section below reports a finding about jealous behaviour, and it is worth separating two things the subject tends to blur. A pattern of checking a partner's phone, requiring their location, needing to account for their time, or deciding who they are allowed to see is not evidence of how much someone cares and not a feature of any particular origin story. Those are controlling behaviours. Find A Helpline, in the sources at the foot of this page, 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.

With that said. The claim this page examines is everywhere: that relationships which started as friendships are stronger, deeper, and last longer than the ones that started with a date. The number quoted in support is almost always the same one — about two thirds of couples were friends first — and it is a real figure from a real paper.

It is also a figure about how relationships begin. Seven samples, 1,897 people, one question: were you friends with this person before you became romantically involved. Nobody in that research was asked how long the relationship lasted, whether it was still going, or how it compared with anything. A prevalence rate cannot answer a durability question, and the three researchers who produced it said as much in their own discussion, where they list comparing the success and the trajectory of the two pathways as work still to be done.

So this page is an audit of what has and has not been measured. It turns out that two studies have come closer to the question than the search results suggest, and that both are unpublished master's theses from the two Canadian labs that defined the field — one of them supervised by a co-author of the prevalence paper, the other by its first author. One of them is the only direct comparison of relationship quality between the two pathways that exists anywhere. Its effects are real, survive several covariates, and account for well under one percent of the variance. One of its five findings runs the wrong way. And the single number in that thesis that touches on lasting points away from the friends-first pathway, for reasons that make it useless as evidence in either direction.

It also turns out that the phrase “love lasts longer when partners are friends” traces to a 2017 blog post whose only citation is its own author's study, described then as in progress and still unpublished nine years later — and which measured self-esteem and life satisfaction, never asked how any couple met, and followed nobody over time.

What follows is fifteen sections: what the two-thirds figure counted, the gap its authors named, the one direct comparison and the size of what it found, the finding that went the wrong way, the length figure that proves nothing, whether the gap closes, why a longer friendship buys nothing, where the headline claim comes from, the strong research on friendship inside a relationship, the 2007 study in which love at first sight does not lose, the one difference that is solidly established, the better argument for the pathway, what it costs, and what can honestly be said today. Our editorial policy explains why every empirical claim here carries a linked source, and why the ones that cannot be sourced are named as such instead of smoothed over.

#1

The two-thirds figure, and the question it was asked

Almost every page answering this question opens with the same number, and the number is real. It is worth being precise about what was measured, because everything else on this page follows from the gap between that and what the number gets used for.

Danu Anthony Stinson, Jessica Cameron and Lisa Hoplock published The Friends-to-Lovers Pathway to Romance: Prevalent, Preferred, and Overlooked by Science in Social Psychological and Personality Science, volume 13, issue 2, pages 562–571, in 2022 (online July 2021). Their Study 3 pooled seven samples their labs had collected between 2002 and 2020 for other purposes, N = 1,897. Five were undergraduates in two regions of Canada; two were crowdsourced adults in Canada and the United States.

The percentage reporting that their relationship began as a friendship, sample by sample: 40.0, 65.2, 70.7, 72.8, 70.5, 62.2 and 70.6, for a weighted mean of 68.2%. A meta-analysis of proportions gave a median of 0.66, 95% CI [0.59, 0.72]. Heterogeneity was substantial — I² = 86% — which the authors attribute to the different wordings used across two decades of surveys.

It did not differ by gender (67% of 545 men against 68% of 808 women), by sample type (71% students against 68% crowdworkers), by ethnicity or by education. It did differ by age in the married sample — 84% under 29, 68% in the thirties, 65% in the forties, 75% at fifty-plus — and by the gender composition of the couple: 85% of the 84 people in same-gender or queer relationships, against 68% of women dating men.

Now the measurement itself. In six of the seven samples the question was a yes-or-no: were you friends with this partner before you became romantically involved. In the remaining one it was a six-option list and the researchers counted everyone who picked “friends.” That is a question about how the relationship started. Nobody was asked how long it lasted, whether it was still going, or how it compared with anything. The figure is a prevalence rate, and a prevalence rate cannot answer a question about durability no matter how large it is.

We went through these same numbers from the other direction in how many dates it takes to know, where the point is that for two thirds of people there were no dates during the part where the knowing happened.

#2

The authors' own paper says nobody has run the study

The reason no page answers the durability question with a citation is not that the answer is hidden in a journal. It is that the study has not been done, and the people best placed to say so said it in print.

Stinson and colleagues built their paper around that absence. In Study 1 they searched PsycInfo for empirical work on relationship initiation across eleven journals that regularly publish relationship science, and found 108 relevant publications. Of those, 85 concerned romantic relationship initiation; of those 85, 63 were about dating initiation and 7 about friends-first, with 8 covering both. Counting the overlap generously, 84% of the literature concerned dating initiation and 18% concerned friends-first. Interrater reliability was 95% agreement, κ = .87.

Study 2 did the same to two standard textbooks on intimate relationships. Of 43 relevant publications cited, 38 concerned romance, and of those, 30 were about dating initiation and 2 about friends-first. The authors note that both of those two were about friends-with-benefits relationships rather than the pathway itself.

Their discussion then lists what this means for anyone wanting to know whether one route produces better relationships. Researchers, they write, “may also need to examine whether these same variables moderate the success of each type of initiation, and whether such variables moderate the trajectory of relationships that form via dating or friends-first initiation.” They call for longitudinal studies and name their own retrospective design as a limitation, since reports about emotionally charged events can be reshaped by what happened afterwards.

That is the state of the evidence as described by the three people who established the prevalence figure, published in the paper the whole genre cites. The question in the title of this page was still open when they wrote it. The next three sections are about the closest anyone has come since, which is closer than the first page of search results suggests and much further from a verdict.

#3

The only direct comparison of the two pathways is an unpublished thesis

There is one study that took the two pathways and compared relationship quality between them. It is a master's thesis, it has never been published as a journal article, and as far as can be established not one page competing for this query mentions it.

Katherine Dyck's Examining differences in relationship quality by romantic relationship initiation pathway was submitted for an MA in psychology at the University of Manitoba and deposited on 28 July 2025. Her supervisor was Jessica Cameron — the second author of the prevalence paper — so this is the same research programme following its own question forward.

The design. 545 introductory-psychology students and 403 crowdsourced adults from Prolific in Canada and the United States consented, and the thesis reports excluding 88 “of the original 953 participants” to leave 865. Those three figures do not reconcile — 545 and 403 make 948, not 953, and 948 minus 88 is 860 rather than 865 — so somewhere a handful of participants are described inconsistently. Every analysis below is reported on 865, which is what the degrees of freedom in the thesis confirm. Mean age 28 (SD 11.93); most identified as women, white, straight and Canadian. Everyone had to be in a relationship at the time. Initiation pathway was classified with Stinson and colleagues' own item, which makes the two studies directly comparable: 538 friends-first against 327 dating-initiated. A power analysis set the target at 788 for a small effect, so the sample was adequate by its own plan.

The measures were established ones, which matters for comparing this to anything else: the three-item trust and commitment scales from Fletcher, Simpson and Thomas's Perceived Relationship Quality Components inventory — the trust one built on Rempel, Holmes and Zanna's — the four-item Couples Satisfaction Index of Funk and Rogge, a shortened uncertainty measure split into uncertainty about the self, the partner and the relationship, and a jealousy measure split into cognitive, emotional and behavioural. Reliability was high: α = .96 for commitment and .94 for satisfaction in this sample.

The result, by multivariate analysis of variance with Pillai's trace because several variables violated homogeneity of variance: Pillai's trace = .021, F(8,856) = 2.34, p = .017. Probing each quality separately, five differed by pathway. Friends-first initiators reported higher trust (6.24 against 6.09), higher satisfaction (5.07 against 4.90), higher commitment (6.51 against 6.37), lower self-uncertainty (1.66 against 1.79) — and, against prediction, higher behavioural jealousy (2.02 against 1.90). Partner uncertainty, relationship uncertainty, cognitive jealousy and emotional jealousy did not differ at all.

So the direction of the first four matches what the genre asserts. The next section is about the size.

#4

The difference is a fifth of a scale point and under one percent of the variance

The honest way to report a finding is in the units it was measured in, and this is where the friends-first advantage stops looking like the thing the headlines describe.

Every one of those differences is on a seven-point scale. Satisfaction: 5.07 against 4.90, a gap of 0.17 of a scale point, with standard deviations of 0.88 and 0.98 — roughly a fifth of the spread within either group. Trust: 0.15 apart. Commitment: 0.14. Self-uncertainty: 0.13.

The effect sizes are printed in the thesis and they are the part worth memorising. Partial eta squared was .005 for trust, .009 for satisfaction, .005 for commitment, .006 for self-uncertainty and .005 for behavioural jealousy. The multivariate effect was .021. In plain terms, which pathway a couple took accounts for somewhere between half a percent and one percent of the differences between people on these measures. The other ninety-nine percent is everything else.

It held up to the checks, and that matters. When personality, self-esteem and attachment were added as covariates the multivariate effect stayed significant. When recruitment sample was added it stayed significant, with four of the five individual effects surviving. When relationship length was added it stayed significant, again with four of five. Whatever this is, it is not an artefact of students differing from crowdworkers.

The author's own defence of the small effects is fair and deserves repeating rather than dismissing: a small effect is not meaningless, and something experienced daily over years can matter more than its variance share suggests. She also names the problem that cuts the other way. People with better relationships volunteer for surveys about their relationships, and in her data every quality measure averaged at the high end of its scale — so this is a comparison between contented friends-first couples and contented dating-first couples, which may not be the comparison a reader wants.

Two more limits, both stated in the thesis. Only one member of each couple reported, so nothing here establishes that both partners experience the advantage. And the design is correlational, which the author points out cannot be fixed, because assigning people to fall in love one way or the other is not an experiment anyone can run. What that leaves is a reliably detected, genuinely tiny difference — roughly the size of effect this site has had to describe before in whether the five love languages work, where the gap between a popular claim and a measured one is also mostly a gap in magnitude.

#5

One of the differences ran the wrong way, and the author says so

The finding the genre would never print is in the same table as the four it likes. Friends-first initiators reported more behavioural jealousy than dating-initiated ones: 2.02 against 1.90, F(1,863) = 3.90, p = .049.

Behavioural jealousy means the actions — checking, monitoring, keeping tabs on where a partner is and who they are with. It is the one component of jealousy in this literature that is consistently associated with worse outcomes, which is why it sits oddly next to higher trust and satisfaction in the same people.

The thesis treats it with the caution it deserves, and the caution is informative. The effect barely cleared significance at p = .049, and it disappeared when anxious attachment was added as a covariate (p = .155) and again when relationship length was added (p = .139). Four of the five original effects survived every covariate; this was the one that did not. Read conservatively, it is a maybe.

That it was anxious attachment that absorbed it is the most informative part. The checking behaviour tracked an individual pattern rather than the route the relationship took — which is the variable our attachment style test scores on the two dimensions the research actually measures, with the caveat on its own page that a cut-off in the middle of a continuous score makes two nearly identical people land in different categories. Whether that pattern is fixed is a separate question with a clearer answer than this one: can attachment styles change.

The author's proposed explanation is worth hearing because it is specific to the pathway rather than generic. Someone whose relationship began inside a friend group has direct evidence that a close friendship can turn into a romance — their own did. A new close friendship of their partner's is therefore a legible threat in a way it might not otherwise be. She connects it to the work of Jaimie Krems, Keelah Williams, Athena Aktipis and Douglas Kenrick, whose eleven studies of 2,918 people found that friendship jealousy is specifically triggered by third parties rather than by the prospect of simply losing a friend, is calibrated to cues of being replaced, and motivates friend-guarding behaviour. If that transfers into a romantic relationship, the thesis concludes, “then there is a dark side to the friends-first pathway” — and it needs replication before anyone builds on it.

One thing does not need replication, and it belongs here rather than in a footnote. A pattern of checking a partner's phone, requiring their location, needing to know where they are at all times, or vetting who they are allowed to see is not a quirk of a particular origin story and not a measure of how much someone cares. Those are controlling behaviours, and the services listed in the sources at the foot of this page exist for exactly that. What the research on jealousy does and does not show — including why the single most famous finding in the area turns out to depend on the answer format — is in the psychology of jealousy, and the smaller, more ambiguous end of the same spectrum is in what has actually been measured about micro-cheating.

#6

The one number that touches “lasting” points the other way, and it still proves nothing

Somewhere in the thesis there is a figure that bears directly on the question in the title. It is not in the abstract, no page quotes it, and its direction is the opposite of what the genre asserts.

Checking whether relationship length could explain her results, Dyck compared the two groups on how long their current relationship had lasted. Dating-initiated relationships averaged 7.07 years (SD 9.52). Friends-first relationships averaged 5.16 years (SD 6.61), a difference of nearly two years, F(1,861) = 12.00, p < .001, partial eta squared = .014 — a larger effect than any of the quality differences in the same study.

If this page were written the way the subject is usually written, that paragraph would be the headline, and it would be wrong. Here is why, and the reasoning is the whole skill the subject requires.

The number is the length of a relationship that is still going on, measured once. That is not durability; it is a snapshot of where people happen to be. And the sample it comes from is two populations stacked on top of each other. The undergraduates had a mean age of 20.57 and a mean relationship length of 2.20 years. The Prolific adults had a mean age of 36.90 and a mean relationship length of 10.24 years. The thesis also reports that Prolific participants were significantly more likely to be dating-initiated than the students were — consistent with the prevalence paper's finding that friends-first initiation is commonest among the under-thirties.

Put those three facts together and the two-year gap is mostly a statement about who ended up in which half of the survey. Older people have longer current relationships and were more often dating-initiated; younger people have shorter ones and were more often friends-first. Nothing about the resilience of either pathway follows.

That is why the figure appears here as a cautionary tale rather than an answer. It is exactly the inference the genre makes in the other direction — reading a cross-sectional difference as a verdict about the future — and it would be no better for pointing our way. A current-length difference measured once cannot settle a durability question, and this site has had to make the same distinction before about whether living together before marriage leads to divorce, where the answer flipped depending on which clock the researchers started.

#7

Do dating-first couples catch up? Somebody ran that analysis

A small advantage that shrinks over the years is a different claim from a small advantage that persists, and the thesis tested which one it is.

The analysis was a hierarchical regression with pathway dummy-coded (dating-initiated 0, friends-first 1) and relationship length mean-centred — the mean was 70.58 months, SD 94.63 — in the first step, and their interaction in the second. Two participants had not given a relationship length, so n = 863. Across all nine relationship-quality measures, the second step was not significant.

The author's reading: dating-initiated couples “do not seem to catch up” to friends-first couples, and the small advantage holds independent of how long people have been together. Her conclusion is that the choice of pathway may have consequences for the whole relationship rather than only its opening.

That reading is defensible and it is worth adding the two things it cannot mean. First, a null interaction is weaker evidence than a significant one; failing to detect a change over time is not the same as establishing there is none, even in a sample powered for small interactions. Second — and this is the structural limit — the study is cross-sectional. Comparing a couple together for two years with a different couple together for twenty is a comparison between people, not a measurement of change. The couples who broke up before the survey are not in the data, in either group, and they are precisely the couples a durability question is about.

The thesis says this plainly in its own limitations, and recommends following couples over months or years instead. So the state of play is: the small gap does not visibly close, measured in the only way it has been measured, which is not the way that would settle it.

The absent group — the people whose relationship has already ended — is where most of the information about durability lives, and almost no research on this topic has it. Our page on what longitudinal data says about thinking of ending a relationship is about the one design that does: a panel of people surveyed twice, so that what happened in between is visible.

#8

Being friends for longer buys nothing measurable

There is a folk version of this idea that goes one step further: not just that friendship first is better, but that the longer the friendship, the stronger the relationship. That one has been tested, and it is the cleanest null in the whole subject.

Dyck correlated two different clocks against all nine relationship-quality measures. One was how long the friendship lasted before it turned romantic, for the 395 friends-first participants whose answers were usable. The other was how long everyone had known their partner before becoming romantically involved, for 721 participants.

Across both sets, every single correlation, the largest coefficient was .08 — and the significant ones were a commitment correlation of .08 for pre-dating acquaintance and nothing else. Her conclusion was that these variables “were not related to my relationship qualities of interest in any meaningful way,” and she dropped them from further analysis. Having been friends for six months and having been friends for six years looked the same on every measure she took.

One methodological detail from that analysis is more interesting than the null itself. 144 participants had to be removed because their two answers contradicted each other — reporting, for instance, that they had been friends for two years but had known the person for five months. More than one in six people could not give a consistent account of their own timeline, which is a direct measurement of how much weight retrospective relationship chronologies can carry. Every prevalence figure on this page rests on reports of the same kind.

For context on the scale involved: in the prevalence paper's Study 4, the friendship phase lasted a mean of 21.90 months (SD 28.23), with a median of 12 and a range from 1 to 180. So the typical friends-first couple had about a year of knowing each other first, and the variation around that year appears to predict nothing about how the relationship turned out.

What a long acquaintance does change is harder to see than relationship quality and better evidenced, and it is in what actually builds attraction over time: the research there suggests the gain is in selection rather than in the relationship itself. And the number that exposes the retrospective-timeline problem most sharply is in how many dates it takes to know.

Further illustration for: Do Friends-to-Lovers Relationships Last Longer? The Two-Thirds Figure Counts How Couples Met, Not How They Ended
#9

Where “love lasts longer when partners are friends” actually comes from

The phrase in the headline of this section is the one the search results are built on, and it has a traceable source. It is a Psychology Today blog post by Suzanne Degges-White, published on 13 July 2017, and it is the first result for its own title.

Here is what it says about its evidence, in its own words: “In a new research study that we are just now completing with over 400 adult participants, we explored the role of relationships and friendships with happiness, self-esteem, and fidelity.” It presents its findings as preliminary. Its reference list credits “Degges-White, S. (in progress). Friends and lovers are key to a happy life.”

Nine years later that is still the citation. There is no published paper to check, no method section, no sample description beyond “over 400 adult participants,” no effect sizes.

And the content does not match the title either. What the post reports is that partnered people had the highest well-being of any group; that self-esteem was slightly higher among people who considered their partner a friend, and slightly higher again for “best friend”; that people who regard a partner as a friend were less likely to report infidelity; and that having at least one friend who cares about you strongly predicted life satisfaction. Those are cross-sectional measures of happiness and self-esteem. The post measures no relationship durations, follows nobody over time, and — crucially — never asks how the couple met. It is about friendship inside a relationship, not about the friends-first pathway at all.

The other high-ranking piece makes the same move more subtly. A December 2023 Psychology Today post by Wendy Patrick cites Stinson and colleagues accurately, with the DOI, and summarises their prevalence and preference findings correctly. Then its closing paragraph states that “easy, comfortable relationships often develop into healthy romantic, lifelong partnerships of love and respect” — with no citation, because there is none to give. A correctly cited prevalence finding becomes an uncited durability claim in the space of one page.

That substitution is the engine of the entire SERP. Once you can see it, the rest of the results read differently.

#10

Friendship inside a relationship is a real finding — about something else

The frustrating part is that the claim the genre is reaching for has good evidence behind it. It is simply a claim about a different thing, and keeping the two apart is the whole point of this page.

Laura VanderDrift, Juan Wilson and Christopher Agnew published On the benefits of valuing being friends for nonmarital romantic partners in the Journal of Social and Personal Relationships, volume 30, issue 1, pages 115–131, in 2013. Two two-wave studies — N = 190 and N = 184 — tested whether valuing the friendship aspect of a romance predicted anything. It did: concurrent love, sexual gratification and romantic commitment, increases in those qualities over time, and a negative association with dissolution. They also showed the benefit was specific to valuing friendship rather than to valuing any aspect of the relationship, the sexual aspect included.

Shawn Grover and John Helliwell published How's Life at Home? New Evidence on Marriage and the Set Point for Happiness in the Journal of Happiness Studies, volume 20, issue 2, pages 373–390, in 2019. Using the British Household Panel Survey they controlled for individuals' pre-marital well-being and still found married people more satisfied, which they read as a causal effect rather than selection. Exploring friendship as the mechanism, they found the well-being effects of marriage about twice as large for people whose spouse is also their best friend.

Both are stronger designs than anything in this subject: two waves and a national panel with pre-marriage baselines, against cross-sectional retrospective surveys. Both support something like “friendship is load-bearing in a romance.”

Neither asked how the couple met. A person can be their partner's best friend having met them on an app six months earlier, and a person can have been someone's friend for three years and not be their best friend now. The friends-first pathway is about initiation; this research is about a current property of a relationship. The two are not the same variable and nothing in the literature establishes that one produces the other.

If what you actually want to know is whether friendship is load-bearing in your own relationship, that is a question about the present and it is answerable. What the research on couples who last actually reports goes through that literature carefully, including why the famous percentages attached to it are smaller and more cautious in the published papers than in the retellings.

#11

Love at first sight does not lose, which is awkward for the whole premise

If the friends-first pathway were as decisive as the genre implies, its opposite should look bad in the data. The only study to set the two against each other found the opposite, and it was published in 2007.

Dick Barelds and Pieternel Barelds-Dijkstra published Love at first sight or friends first? Ties among partner personality trait similarity, relationship onset, relationship quality, and love in the Journal of Social and Personal Relationships, volume 24, issue 4, pages 479–496. The sample was 137 married or cohabiting couples, and both partners completed personality measures.

Two of their three predictions held. Couples who fell in love at first sight became romantically involved sooner after meeting. And they had less similar personalities, specifically on extraversion, emotional stability and autonomy — consistent with the idea that a long acquaintance lets people sort on more than first impressions.

The third prediction failed. The authors had hypothesised that love-at-first-sight couples would report relatively low relationship quality. They did not. What the paper found instead was that the link between personality similarity and relationship quality itself depended on how the relationship started and on which trait was being measured.

This is a single cross-sectional sample of couples still together, with the same selection problem as everything else here, and one failed prediction is not proof of equivalence. But it is the only head-to-head comparison of the two onsets on relationship quality that exists in a peer-reviewed journal, it is nearly twenty years old, and its answer to “is starting fast worse” is no.

The similarity half of that finding connects to a claim this site has been through in detail: whether opposites actually attract, where the measured effect of similarity on relationship satisfaction is far weaker than either folk theory predicts. And the related question of what intense early feeling does and does not predict is in the early signs of emophilia and in why the honeymoon phase ends.

#12

What the two pathways verifiably differ in is the clock

Set durability aside and ask what has been measured about the difference between the two routes. The answer is a timeline, and it comes from the second unpublished thesis in this story.

Erin Lowey's Friends-first romantic relationship initiation was submitted for an MSc at the University of Victoria and deposited on 23 December 2022. Her supervisor was Danu Stinson — first author of the prevalence paper. So the two pieces of research that come closest to this question are graduate theses from the two labs that defined the field, and neither has been published.

Her Study 2 recruited undergraduates who had begun a relationship within the previous two years, whether or not it was still going. Of 680 recruited, 114 were ineligible and 90 were excluded, leaving 476 participants (mean age 21.10). Each dated 28 relationship milestones and rated emotional intimacy, passion and romantic interest at each one.

The headline is the length of courtship, defined as the gap between first meeting and becoming an official couple. For dating initiators it was 174.17 days, about six months. For friends-first initiators it was 654.13 days, about 22 months — d = −.61, a large difference. Using a stricter definition, from first hanging out one-on-one to becoming official, the gap was still there: 94.41 days against 272.38, t(420.9) = −4.12, p < .001, d = −.39.

Three more differences, each measured:

The feelings arrive in a different order. Dating initiators had higher romantic interest at the moment of first meeting (d = .43). Friends-first initiators had higher passion at the first kiss, and higher emotional intimacy at every physically intimate milestone.

The social network comes in at the opposite end. Friends-first initiators introduced their partner to friends and family early, while intimacy and romantic interest were still developing. Dating initiators did it late, once high intimacy was already established. The thesis reads this as approval of the social circle being a precondition in one pathway and a milestone in the other.

There is a long gap neither group can account for. After first meeting, the next milestone was 78 days away for dating initiators and nearly 290 for friends-first ones. Something was happening in those months that a milestone checklist does not capture — and notably, dating initiators were not calling each other friends during it either.

On the question of this page, the thesis is careful in the way the genre is not. Its conclusion is that the pathways “differ in substantive ways that could have implications for longer-term relationship satisfaction, stability, and longevity.” Could. It measured none of them: participants' relationships had begun within two years, some had already ended, and the author treats that status only as a possible memory bias rather than as an outcome. She also offers her own hypothesis — that established intimacy may buffer a relationship better than initial passion — and labels it as one.

What this does supply is a reason the two groups are not comparable in the first place. One route takes three and a half times as long to reach the same official label. Any honest comparison of outcomes would have to decide when each clock starts, which is the problem that broke the cohabitation-and-divorce literature for thirty years.

#13

The better argument for the pathway is selection, not durability

There is a case for friends-first that does not require the durability claim at all, and it rests on the strongest study in this whole area.

Lucy Hunt, Paul Eastwick and Eli Finkel published Leveling the Playing Field in Psychological Science, volume 26, issue 7, pages 1046–1053, in 2015. They took 167 couples — 67 dating, 100 married — asked how long the partners had known each other before dating and whether they had been friends first, and had trained coders rate each person's physical attractiveness from video.

The overall attractiveness-matching correlation was .38, close to the meta-analytic average for the effect. The result is in the moderation. 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. Comparing the two pathways directly: couples who had not been platonic friends first matched at .67; couples who had been friends first, at .43.

The mechanism proposed is market-based. Competition for partners works through consensus about who is desirable, and consensus is strongest at first sight and decays as people get to know each other in more than one context. A long friendship is a period in which that consensus stops operating.

Two honest limits, both in the paper. The authors set aside 20 couples who had known each other longer than 19 months as extreme values; in a robustness check that kept them capped instead, the acquaintance-length interaction weakened to p = .084 on one measure and p = .153 on the other. And it is one mostly white, mixed-sex American sample.

Now the finding that makes this a selection argument rather than a quality one. In that same sample, how well a couple matched on attractiveness had no association with their relationship satisfaction, for either partner. So the friends-first pathway appears to change who ends up with whom, without that change showing up in how satisfied they are.

That is a smaller and much more defensible claim than “friends-first relationships last longer,” and it happens to be the one with the best evidence behind it. The full version of this research, including what mere exposure does and does not do, is in what actually builds attraction over time. The pathway seen from the side where it stalls instead of turning is in why people keep ending up in the friend zone.

#14

What the pathway costs, where there are numbers

Any page recommending this route owes its reader the other half, because the friends-first pathway has a downside the dating one does not: there is an existing relationship to lose. Here the numbers are thinner but they exist.

Start with what friends-first initiation usually is not. In the prevalence paper's Study 4, of 210 friends-first initiators, 12.0% said they had intentionally befriended their partner because they were attracted to them, 17.7% said their partner had done so, and 70.3% chose “we just became friends and then became attracted after getting to know each other.” The pathway is mostly something that happens, not a strategy that gets executed. Reading it as advice inverts how it actually works for seven people in ten.

For the version that does involve a sexual relationship first, there is a direct measurement, and it is sobering. Laura Machia, Morgan Proulx, Michael Ioerger and Justin Lehmiller published a two-wave longitudinal study of friends-with-benefits relationships in Personal Relationships in 2020, following 192 people from one time point to the next. The most common outcome was no relationship of any kind at Time 2 (31%). Of those who wanted the arrangement to become a friendship, 59% got that. Of those who wanted it to become a romantic relationship, 15% did. The prevalence paper notes in passing that 42% of its married friends-first initiators had been in a friends-with-benefits relationship with the person they went on to marry — so the route exists, and the 15% is the rate at which wanting it makes it happen.

And on attraction inside an ongoing friendship, April Bleske-Rechek and colleagues studied pairs of friends and then a wider sample of emerging and middle-aged adults, published in the Journal of Social and Personal Relationships in 2012. Men reported more attraction to their cross-sex friend than women did, regardless of either person's relationship status. In the second study, both men and women nominated attraction to a cross-sex friend as a cost more often than as a benefit. Among younger women and middle-aged participants, more attraction to a current cross-sex friend went with less satisfaction in their current romantic relationship.

None of that says don't. It says the risk is real, asymmetrically experienced, and absent from every page that presents this pathway as the superior option. The question of how to read a friend's behaviour at all runs into the wall this site has described before: nothing in the evidence separates romantic warmth from friendly warmth reliably, which is the subject of flirting or just friendly and, from the other side, coming across as flirty. What happens to a friendship afterwards, when there is a romance in its past rather than its future, is in what the research says about staying friends with an ex.

#15

What would answer the question, and what can honestly be said today

The study that settles this is not exotic. It would recruit people at the start of relationships formed both ways, follow them for years, and record dissolution as the outcome — the design Stinson and colleagues called for in 2022 and the one both theses name in their limitations. Nobody has run it. Until somebody does, the answer to the title of this page is that it is unknown, and that is a finding rather than an evasion.

What can be said. Friends-first initiation is how roughly two thirds of relationships begin, with a meta-analytic median of .66, and it is the most popular answer when university students are asked the best way to meet someone — 47.4% chose it, ahead of mutual friends and school at 18.1% each, with online dating at 0.3% and a blind date at 0.3%. It takes about three and a half times as long to reach the official-couple label. It appears to weaken attractiveness matching, which in the same sample had no bearing on satisfaction. And in the one direct comparison that exists, friends-first initiators report very slightly higher trust, satisfaction and commitment and slightly lower self-uncertainty, with effects under one percent of the variance, in an unpublished thesis.

What cannot be said. That these relationships last longer. That they are less likely to end. That a longer friendship first makes anything better — the correlations top out at .08. And that the prevalence figure speaks to any of this: it counts how couples met.

What is unknown and gets asserted anyway. Comparative breakup rates. Whether the small quality gap would survive a sample that includes unhappy couples. Whether it holds for both partners, since only one was ever asked. And anything about non-heterosexual couples specifically, even though friends-first initiation is most common among them — 85% in the prevalence data, from 84 people across four samples, which the authors describe as too small to be definitive.

If the reason for the question is a decision rather than curiosity, notice that none of this literature can inform it. What predicts how a relationship goes, in the research that has looked, is how the two people treat each other and how conflict runs — not the route by which they met. That is measurable in the present. Our relationship satisfaction test scores satisfaction and commitment separately, because the two come apart; the conflict style test is built on the demand-withdraw literature rather than on a workplace instrument; and both say on their own pages that they are educational tools and not diagnoses. The questions that actually distinguish compatible couples from the rest, and the ones that do not, are in what questions reveal about real compatibility, and the markers worth watching early are in signs a relationship has real potential. The rest of this section is under Dating.

Pulling it together

The question has a clear answer and it is not the one the subject advertises: nobody knows, because the study has never been run. What exists is a prevalence figure doing duty as a durability figure, two unpublished theses that measured adjacent things carefully, and a 2017 blog post whose evidence is a survey that never got published.

That is not nothing, though. The friends-first pathway is how most relationships actually begin, it is what people say they prefer by a wide margin, it takes about three and a half times as long to reach the official label, and it appears to loosen the sorting on physical attractiveness that operates between strangers — a sorting that, in the same study, had no bearing on how satisfied anyone was. In the single direct comparison, friends-first initiators do report slightly more trust, satisfaction and commitment. Slightly is the operative word: a fifth of a scale point, under one percent of the variance, in a sample of people who were all doing fairly well.

If there is a practical reading, it is that the pathway is a description of how people meet rather than a technique for making a relationship work. Nothing in this literature says a friendship first will protect a relationship, and nothing in it says starting fast will damage one. The research that does predict how relationships go measures what the two people do now — how they handle conflict, whether responsiveness is mutual, whether commitment and satisfaction are both present. Those are things that can be looked at this week, and they do not care how you met.

Frequently asked questions

Do friends-to-lovers relationships last longer?

No study has measured this, so the honest answer is that it is unknown. The figure cited in support — that about two thirds of couples began as friends — comes from Stinson, Cameron and Hoplock's meta-analysis of seven samples, N = 1,897, with a median proportion of .66, 95% CI [.59, .72]. That research asked one question: were you friends before becoming romantically involved. It did not measure relationship length, breakups or outcomes of any kind, and its authors explicitly list comparing the success and trajectory of the two pathways as future work. The closest anyone has come is an unpublished 2025 master's thesis finding slightly higher trust, satisfaction and commitment among friends-first initiators, with effect sizes of partial eta squared .005 to .009. No study has compared dissolution rates between the two pathways.

What does the two-thirds statistic about friends-first couples actually measure?

Prevalence — how often relationships start that way. Stinson, Cameron and Hoplock pooled seven samples collected in their labs between 2002 and 2020: five of Canadian undergraduates and two of crowdsourced adults in Canada and the United States, N = 1,897 in total. The percentage reporting friends-first initiation ran 40.0, 65.2, 70.7, 72.8, 70.5, 62.2 and 70.6, for a weighted mean of 68.2%, and a meta-analysis of proportions gave a median of .66. The rate did not differ by gender, sample type, ethnicity or education. It was higher among people under 29 (84% of the married sample) and higher among people in same-gender or queer relationships (85%, though from only 84 participants). All of it is retrospective self-report about how a relationship started.

Has anyone compared relationship quality between friends-first and dating-first couples?

Once, in a master's thesis that has not been published as a journal article. Katherine Dyck's 2025 MA thesis at the University of Manitoba, supervised by Jessica Cameron, surveyed 865 people — 538 friends-first and 327 dating-initiated — using the same classification item as the prevalence paper. The multivariate test was significant (Pillai's trace = .021, F(8,856) = 2.34, p = .017). Friends-first initiators reported higher trust (6.24 against 6.09 on a seven-point scale), satisfaction (5.07 against 4.90), commitment (6.51 against 6.37) and lower self-uncertainty (1.66 against 1.79). Partial eta squared ran from .005 to .009, so pathway accounted for well under one percent of the variance. Four of the five effects survived adding personality, attachment, recruitment sample and relationship length as covariates.

Does being friends for longer before dating make the relationship better?

Not measurably. The same 2025 thesis correlated two clocks — length of the friendship before it turned romantic, for 395 friends-first participants, and length of acquaintance before becoming involved, for 721 participants — against all nine of its relationship-quality measures. The largest correlation coefficient in either analysis was .08, and the author concluded the variables were not related to relationship quality in any meaningful way and dropped them. For scale: in the prevalence paper's Study 4, the friendship phase lasted a mean of 21.90 months with a median of 12 and a range from 1 to 180. One detail from that analysis is worth knowing: 144 participants had to be excluded because their two timeline answers contradicted each other, which is a direct measure of how reliable retrospective relationship chronologies are.

Why do so many pages say “love lasts longer when partners are friends”?

Because of a Psychology Today post of that title by Suzanne Degges-White, published on 13 July 2017, which is the first result for its own headline. The post describes “a new research study that we are just now completing with over 400 adult participants” and presents preliminary findings; its reference list credits “Degges-White, S. (in progress). Friends and lovers are key to a happy life.” Nine years later there is still no published paper to check. What the post reports is cross-sectional: partnered people had higher well-being, self-esteem was slightly higher among people who considered a partner a friend, and people who did were less likely to report infidelity. It measured no relationship durations, followed nobody over time, and never asked how any couple met — so it is about friendship inside a relationship, not about the friends-first pathway.

Is love at first sight worse for a relationship than being friends first?

The one study that compared them found no quality disadvantage. Dick Barelds and Pieternel Barelds-Dijkstra studied 137 married or cohabiting couples and published the results in the Journal of Social and Personal Relationships in 2007. Two of their predictions held: couples who fell in love at first sight got involved sooner after meeting, and their personalities were less similar on extraversion, emotional stability and autonomy. The third failed — they had hypothesised that love-at-first-sight couples would report relatively low relationship quality, and those couples did not report lower quality. Instead, the link between personality similarity and relationship quality depended on both the type of onset and the specific trait. It is one cross-sectional sample of couples still together, so it does not establish equivalence; it is also the only head-to-head comparison in a peer-reviewed journal.

What actually differs between the two pathways?

The timeline, and it has been measured precisely. Erin Lowey's 2022 MSc thesis at the University of Victoria, supervised by Danu Stinson, had 476 undergraduates date 28 relationship milestones. From first meeting to becoming an official couple took 174.17 days for dating initiators — about six months — and 654.13 days for friends-first initiators, about 22 months, a large difference at d = −.61. Using a stricter start point, from first hanging out one-on-one, it was 94.41 days against 272.38, t(420.9) = −4.12, p < .001, d = −.39. Dating initiators had higher romantic interest at first meeting (d = .43); friends-first initiators had higher passion at the first kiss and higher emotional intimacy at every physically intimate milestone, and introduced their partner to friends and family much earlier in the process.

How often does a friends-with-benefits relationship turn into a romantic one?

In the one longitudinal measurement, rarely for the people who wanted it to. Laura Machia, Morgan Proulx, Michael Ioerger and Justin Lehmiller followed 192 people in friends-with-benefits relationships across two time points, published in Personal Relationships in 2020. The most common Time 2 outcome was no relationship of any kind, at 31%. Of those who wanted the arrangement to become a friendship, 59% got that. Of those who wanted it to become a romantic relationship, 15% did. The route does exist: the prevalence paper reports that 42% of its married friends-first initiators had been in a friends-with-benefits relationship with the person they went on to marry. The 15% is the rate at which wanting that outcome produced it over the study's interval.

Is there a downside to the friends-first pathway?

There are two with evidence behind them. April Bleske-Rechek and colleagues, in the Journal of Social and Personal Relationships in 2012, found that men reported more attraction to a cross-sex friend than women did regardless of either person's relationship status, that both men and women nominated attraction to a cross-sex friend as a cost more often than a benefit, and that among younger women and middle-aged participants more attraction to a current friend went with less satisfaction in their existing romance. And the 2025 Manitoba thesis found friends-first initiators reporting more behavioural jealousy than dating initiators (2.02 against 1.90, p = .049) — a result that barely cleared significance and vanished once anxious attachment or relationship length was controlled, which the author flags as needing replication before anyone draws conclusions from it.

Sources

What this article draws on. For the broader bibliography see our sources page.

Keep reading