How we met · United States · 1960–2020

They still say they met at a bar. Most of them didn’t.

Ask anyone how they met their partner and the answer used to be a room. A bar. A church. The office. A friend’s party. One by one, those rooms emptied. Here is what actually happened, in nine pictures.

01The bar

They still say they met at a bar.

Drag across the band to read any year

1960

The share whose story names a bar has barely moved since 1995. What changed is where they actually met.

Six keys fanned out on a ring.

02The rooms, one at a time

It didn’t stop at friends.

Online dating didn’t beat one way of meeting and stop. It passed every single one, in order, over twenty-five years. The small and local went first. The close and personal went last.

in 1990, 1 couple in 100 met online same building, different employers 1990 on vacation 1996 the military personal ads, matchmakers, speed dating 1997 neighbours a blind date 1998 church 2000 a voluntary organisation as customer and client out in public a private party 2001 college school family 2003 coworkers 2004 a bar or restaurant 2005 friends 2014 1990 1995 2000 2005 2010 2015 the year online dating passed it in 1990, 1 couple in 100 met online same building, different employers 1990 on vacation 1996 the military personal ads, matchmakers, speed dating 1997 neighbours a blind date 1998 church 2000 a voluntary organisation as customer and client out in public a private party 2001 college school family 2003 coworkers 2004 a bar or restaurant 2005 friends 2014 1990 1995 2000 2005 2010 2015 the year online dating passed it
Seventeen ways of meeting, each drawn at the year online dating first became more common than it. Ordered by that year; thickness is the share of couples the channel still accounted for at the moment it fell.

Each bar ends at the first year online dating became more common than that way of meeting. Five-year averages, weighted. We checked all 17 dates a second way — heterosexual couples only, unweighted — and 16 of them landed within two years. The one that didn’t, same building, different employers, never accounted for more than one couple in a thousand, so it is drawn as an outline. Only friends had been published before, at 2013. Source: How Couples Meet and Stay Together, Stanford University Libraries (HCMST 2009 waves 1-3; HCMST 2017/2020/2022)

Church fell in 2000. Your parents’ friends in 2003. Work in 2004. The bar in 2005. Meeting through friends held out until 2014 — and that is the only one of these dates anyone has published before.

An empty dining chair.

03Family

Your family used to introduce you.

Of every way of meeting that was ever big, being introduced by your own relatives fell the furthest and the fastest.

28% → 6%

More than one couple in four met through their own family in the early 1960s. Now it is about one in sixteen — a fall of 79%.

0%10%20%30%40%1955196519751985199520052015friendscoworkersfamilyschoolchurch
0%10%20%30%40%195519852015friendscoworkersfamilyschoolchurch
Each line is the share of couples who met that way. Family is the coral line. The grey lines are the other big ways of meeting, for scale — friends barely moved for fifty years, and family fell right past it.

Five-year groups, weighted, 1955–1959 to 2015–2019. Source: How Couples Meet and Stay Together, Stanford University Libraries (HCMST 2009 waves 1-3; HCMST 2017/2020/2022)

Grandparents met through relatives. Parents met through friends. The generation after that mostly met nobody through anybody.

A smartphone showing a single card on its screen.

04Who is actually on the apps

The youngest people are the least online.

Dating apps get filed as a young person’s thing. In the data it is the other way round, and it has been for twenty years.

16.6% vs 46.2%

People who met their partner under 25 are the least online group there is. People who met at 45 or over are the most.

school, college or friendsonlineunder 2553.6%16.6%132 couples25–3416.2%45.4%172 couples35–4419.1%36.8%75 couples45 and over10.3%46.2%128 couples
school, college or friendsonlineunder 2553.6% · 16.6%132 couples25–3416.2% · 45.4%172 couples35–4419.1% · 36.8%75 couples45 and over10.3% · 46.2%128 couples
Two numbers for each age group, both measured out of the same people. They are not two halves of anything — a story can mention a friend and an app at once — so they are never added together.

Couples who met 2015–2022, weighted. Age is age when they met. Source: How Couples Meet and Stay Together, Stanford University Libraries (HCMST 2009 waves 1-3; HCMST 2017/2020/2022)

The reason is in the left-hand bars. Under 25 you are still inside the institutions that introduce people to each other — school, college, a big group of friends. Past 35 those are gone: school and college are at zero. Online dating is what you use after school ends.

A stopwatch with one hand.

05Ten years early

Gay couples got there first.

Same-sex couples didn’t take to online dating a bit faster than everyone else. They got there roughly a decade earlier, and everyone else spent that decade catching up.

10 years

One in five same-sex couples were meeting online by 2002. Different-sex couples didn’t reach that until 2012.

5 in 100 couples1 year10 in 100 couples2 years20 in 100 couples10 years30 in 100 couples13 yearshow far ahead same-sex couples were at each level
5 in 100 couples1 year10 in 100 couples2 years20 in 100 couples10 years30 in 100 couples13 yearshow far ahead same-sex couples were at each level
Measured across rather than up: how many years it took everyone else to reach a level same-sex couples had already reached. The gap widens the higher up you look.

Seven-year average, weighted. Both studies deliberately over-sampled LGB respondents, which is why the numbers hold up; the level in any single year is noisy, so this is measured at crossings instead. Source: How Couples Meet and Stay Together, Stanford University Libraries (HCMST 2009 waves 1-3; HCMST 2017/2020/2022)

It isn’t a story about technology. It is a story about how thin the local options were. When the people you could actually meet in your own town are a few percent of the room, a search tool isn’t a convenience — it is the only thing that works. The same pattern shows up outside the cities, for the same reason.

A spiral-bound block of blank paper squares.

06The calendar

Love has a season, and December is the worst of it.

Couples do not meet at a steady rate through the year. They meet in June.

Three times as many couples start in June as in December — the month everybody associates with romance is the month it least often begins.

an even year would sit hereJanFebMarAprMay154JunJulAugSepOctNov51Dec
an even year would sit hereJanFebMarAprMay154JunJulAugSepOctNov51Dec
Each column is one month, adjusted so a short February isn’t punished for being short. The dashed line is what a year with no season would look like.

3,492 couples who remembered the month, weighted. December is the lowest month in every era measured. Source: How Couples Meet and Stay Together, Stanford University Libraries (HCMST 2009 waves 1-3; HCMST 2017/2020/2022)

The swing is flattening — it used to be four and a half times, now it is about two. The reason is the obvious one: the way of meeting that replaced all the others doesn’t have a summer.

A partly unfolded paper road map.

07Distance

You used to meet someone from your own town.

Every old way of meeting was really a machine for introducing you to someone who lived nearby. The one that replaced them is not.

39% → 18%

Share of couples who grew up in the same city or town. In the early 1950s it was nearly two in five. Now it is under one in five.

school59.3%out in public26.3%friends22.5%church22%family20%a bar12.5%online7.7%college7.1%couples who met from 2000 on
school59.3%out in public26.3%friends22.5%church22%family20%a bar12.5%online7.7%college7.1%couples who met from 2000 on
The same question, split by how the couple met. School introduces you to the person down the road. Online introduces you to someone who grew up somewhere else entirely.

Couples who met from 2000 on, weighted. This is a proxy: how far apart the two people actually lived is not recorded in any wave of the survey. Source: How Couples Meet and Stay Together, Stanford University Libraries (HCMST 2009 waves 1-3; HCMST 2017/2020/2022)

An hourglass with sand running through it.

08The one everybody repeats

“Online couples break up more” is wrong.

It is the most repeated claim about online dating, the raw numbers do say it, and it falls apart the moment you compare like with like.

2.4× — and then nothing

Couples who met online broke up at 26% over two years against 11% for everyone else. But the online couples had been together 3.5 years; everyone else, 16.

met onlinemet any other waytogether 0–2 years45.1%46.8%3–5 years9.3%24.5%6–10 years6.3%12.2%
met onlinemet any other waytogether 0–2 years45.1%46.8%3–5 years9.3%24.5%6–10 years6.3%12.2%
The same couples, now grouped by how long they had been together. New relationships end more often than old ones — that is all the first number was measuring. Compare couples of the same age and the gap closes, then reverses.

Couples followed up in 2010 and 2011, weighted. Group sizes 57–366. Source: How Couples Meet and Stay Together, Stanford University Libraries (HCMST 2009 waves 1-3; HCMST 2017/2020/2022)

An online relationship could not be twenty years old in 2009, because the websites weren’t twenty years old. The whole effect is that one fact.

A folded newspaper with its columns shown as flat blocks.

09Gone

Two whole institutions disappeared and nobody marked the date.

The blind date was a comedy genre. The matchmaker was a job. Both are now rounding errors.

0%2%4%6%8%196019701980199020002010the blind datepersonal ads and matchmakersthe same building, different employers
0%2%4%6%8%196019852015the blind datepersonal ads and matchmakersthe same building, different employers
Three ways of meeting that have effectively stopped happening. The blind date is the coral line.

Five-year groups, weighted. These are small numbers year to year, so read the direction rather than the exact height. Source: How Couples Meet and Stay Together, Stanford University Libraries (HCMST 2009 waves 1-3; HCMST 2017/2020/2022)

The blind date peaked at about 7% of couples in the late 1960s and is now under one in two hundred. Personal ads, video dating and matchmakers — a whole small industry — went from 2.2% to 0.3% within a few years of the web arriving.

And the first internet couples didn’t meet on a dating site at all. In the early 2000s more than a quarter of them met in a chat room, talking about something else entirely. That is now about one in fifty.

Every room that used to introduce people to each other still exists. Almost none of them still does the introducing.

Where these numbers come from

The survey

How Couples Meet and Stay Together, run by Michael Rosenfeld, Reuben Thomas and Sonia Hausen at Stanford. It asks a nationally representative sample of Americans how they met their partner, and it has been asking since 2009.

This page pools two of those surveys — the 2009 one and the 2017 one, plus its 2020 and 2022 follow-ups. That is 6,405 couples, who met between 1934 and 2022. Both sexes, and both same-sex and different-sex couples.

It is free to download for research. Every chart here is drawn from it, and nothing on this page is an estimate of ours.

How to read a year

Any one year holds too few couples to be worth much on its own, so every point on a line is a 7-year average centred on that year, and every bar is a five-year group. Nothing is drawn where the group holds fewer than 60 couples; the thinnest one that made it holds 227.

Answers are also weighted, so the sample matches the country rather than whoever happened to answer.

Why a bar story can also be an online story

Because the survey lets it. Its own instructions to coders say: “Sometimes people whose relationship starts online end up having a first date in a bar or restaurant… the meeting in person also qualifies for a ‘first meeting’ if respondents describe it as such.”

So a couple who matched on an app and met up at a bar is counted in both. That is not a flaw we found — it is written into the rules, and it is exactly what the first chart separates out.

What we are not claiming

  • These categories overlap. Two thirds of couples give more than one answer — a friend introduced them, at a party, at a bar. Nothing here is a pie chart or a stacked total, and no two numbers on this page should be added together.
  • Anything after 2017 is drawn faint. Only 228 couples in the survey met that recently. The direction is real; the exact height is not a number to quote.
  • People are remembering. Someone who met their partner in 1968 is telling you about 1968 from memory, so the further back you look the softer the year gets. That is why everything is a rolling average.
  • Some groups are small. The age panel rests on 75–172 couples per group and the break-up panel on 57–366. Read those as directions, not decimals.
  • No map. The survey records where someone lives no more finely than a census region, and it never records the partner's home town at all, so the distance section uses two stated proxies and says so.

What was already known

One chart from this data is famous: Figure 1 of Rosenfeld, Thomas and Hausen, “Disintermediating your friends”, PNAS 2019. It shows nine ways of meeting over time and it is where “online passed friends around 2013” comes from.

That paper states the bar/restaurant result in a single sentence of text and never draws it. Of the 17 dates in the second section, one — friends — had been published before. Our pipeline reproduces their published numbers on their own basis first, as a check, before departing from it.

Checking it yourself

Every number comes out of one pipeline: python pipeline/run.py for the full inventory, then export_hero.py, export_acts.py and export_sections.py for the series drawn here. verify_hero.py and verify_claims.py re-check every number printed on this page against those exports and fail if one drifts.

The first chart is drawn from the exact series — both edges of the band are the data, and every bit of texture lives inside the shape. No line is ever moved to make it look better.

Every number in the first chart (61 years)
The share of American couples meeting in each year whose story mentions a bar or restaurant, split by whether they met there or met online first. 7-year average, weighted. Rows after 2017 are in italics: they rest on the 2020 and 2022 follow-ups.
Year they met Mentions a bar Met at the bar Met online first Share that began online Couples in the average
1960 17.9% 17.9% 0.0% 0.0% 227
1961 15.9% 15.9% 0.0% 0.0% 240
1962 15.5% 15.5% 0.0% 0.0% 256
1963 13.4% 13.4% 0.0% 0.0% 275
1964 14.9% 14.9% 0.0% 0.0% 293
1965 15.6% 15.6% 0.0% 0.0% 315
1966 14.8% 14.8% 0.0% 0.0% 327
1967 15.4% 15.4% 0.0% 0.0% 370
1968 16.3% 16.3% 0.0% 0.0% 387
1969 16.9% 16.9% 0.0% 0.0% 406
1970 17.0% 17.0% 0.0% 0.0% 420
1971 17.7% 17.7% 0.0% 0.0% 443
1972 18.0% 18.0% 0.0% 0.0% 454
1973 18.5% 18.5% 0.0% 0.0% 458
1974 18.6% 18.6% 0.0% 0.0% 440
1975 18.4% 18.4% 0.0% 0.0% 471
1976 18.6% 18.6% 0.0% 0.0% 485
1977 19.0% 19.0% 0.0% 0.0% 519
1978 19.0% 19.0% 0.0% 0.0% 530
1979 19.5% 19.5% 0.0% 0.0% 541
1980 22.1% 22.1% 0.0% 0.0% 561
1981 24.2% 24.2% 0.0% 0.0% 589
1982 23.6% 23.6% 0.0% 0.0% 613
1983 22.9% 22.9% 0.0% 0.0% 642
1984 22.8% 22.8% 0.0% 0.0% 637
1985 23.1% 23.1% 0.0% 0.0% 649
1986 22.5% 22.5% 0.0% 0.0% 691
1987 20.4% 20.4% 0.0% 0.0% 731
1988 19.8% 19.8% 0.0% 0.0% 734
1989 19.0% 19.0% 0.0% 0.0% 729
1990 19.3% 19.3% 0.0% 0.0% 743
1991 18.6% 18.6% 0.0% 0.0% 800
1992 17.9% 17.9% 0.0% 0.0% 834
1993 17.5% 17.5% 0.0% 0.0% 852
1994 17.2% 17.2% 0.0% 0.0% 866
1995 16.5% 16.4% 0.1% 0.6% 940
1996 17.1% 16.8% 0.3% 2.0% 1,005
1997 17.3% 16.8% 0.4% 2.5% 1,078
1998 17.1% 16.5% 0.6% 3.6% 1,073
1999 16.6% 15.8% 0.8% 4.9% 1,093
2000 16.1% 14.9% 1.1% 7.0% 1,108
2001 16.2% 14.1% 2.1% 12.7% 1,143
2002 16.0% 13.3% 2.7% 16.7% 1,128
2003 16.4% 13.4% 3.0% 18.5% 1,136
2004 16.0% 12.4% 3.5% 22.1% 1,195
2005 16.3% 12.4% 3.9% 23.9% 1,267
2006 17.2% 12.7% 4.5% 26.3% 1,283
2007 18.1% 13.1% 4.9% 27.4% 1,215
2008 17.5% 12.8% 4.7% 27.1% 1,126
2009 17.8% 13.0% 4.8% 26.9% 1,070
2010 17.2% 12.3% 4.9% 28.2% 1,000
2011 18.5% 12.5% 6.0% 32.3% 833
2012 18.7% 11.8% 6.9% 37.0% 743
2013 19.4% 10.9% 8.5% 43.7% 717
2014 19.6% 10.5% 9.2% 46.7% 734
2015 20.7% 11.3% 9.5% 45.6% 684
2016 20.3% 10.8% 9.5% 47.0% 631
2017 20.4% 10.0% 10.4% 51.1% 555
2018 19.3% 9.0% 10.3% 53.5% 493
2019 19.8% 9.3% 10.5% 53.2% 373
2020 16.3% 7.8% 8.5% 52.3% 230

In short: the band peaked at 24% of couples in 1981, has held between 16% and 21% every year since 1995, and by 2019 53% of it was couples who had already met online. The two halves changed places in 2017.