Social media and communication
Relationships that start online
Meeting online has become the single most common route into a US couple, and no outcome study finds those relationships are worse or shorter.
Two findings carry most of the weight. First, meeting online overtook every other route into a heterosexual American couple around 2013 and by 2017 accounted for four in ten new couples. Second, the outcomes look ordinary: the one good breakup comparison finds no detectable difference between couples who met online and couples who did not, and the study claiming an advantage was paid for by a dating company. The commonest error in public discussion is arithmetic rather than direction — the share of newly formed couples who met online is roughly four times the share of all existing couples who did, and the two are routinely swapped for each other. For a workplace-oriented example of technology measurement in practice, see this page.
Online became the most common way to meet in the mid-2010s
The tracking data come from a long-running US survey that has asked how couples met the same way across decades.
39% of US heterosexual couples who met in 2017 met online, up from about 2% in 1995, while meeting through friends fell from 33% to 20%.
Direction: Increase. Strength of evidence: Strong.
Caveat The 39% is a share of recently formed couples rather than of all couples, and the survey covers the United States only.
The same survey put the figure for same-sex couples at about 65%, fitting the simplest explanation available: online search matters most where the offline pool of partners is thinnest (Rosenfeld, Thomas & Hausen, 2019). That subsample is small and its interval wide.
The flow figure and the stock figure are not interchangeable
Surveys of the whole adult population produce much smaller numbers, and they do not contradict the figures above. They count a different thing: every partnership currently in existence, most of which began before dating apps did.
Only 10% of all partnered US adults met their current partner online, rising to 20% of partnered adults under 30 and 24% of partnered LGB adults.
Direction: Increase. Strength of evidence: Strong.
Caveat A stock measure across all existing relationships, which is why it sits far below the flow measure for newly formed couples; the two are frequently conflated.
Use is also more occasional than headlines imply. Pew found in 2023 that 30% of US adults had ever used a dating site or app but only 9% had in the past year, with lifetime use at 53% among under-30s and 51% among LGB adults — a lifetime measure overstates current engagement roughly threefold.
The experience divides sharply by sex
Aggregate satisfaction is close to an even split, which is itself informative: a product half its users describe negatively remains the dominant route into new relationships. The split is not random.
53% of online daters describe the experience as positive and 46% as negative, but 56% of women under 50 who have used the platforms received unsolicited sexually explicit messages and 11% received threats of physical harm.
Direction: Cuts both ways. Strength of evidence: Strong.
Caveat Retrospective self-report over an unbounded window, so the harassment rates are lifetime-on-platform rather than annual.
Relationships that begin online do not end faster
This is the question most people want answered, and the answer is a null once the obvious differences between the two groups are adjusted for.
Couples who met online broke up at 7.0% per year against 7.3% for those who met offline, an adjusted odds ratio of 1.10 with a confidence interval from 0.77 to 1.59.
Direction: No detectable effect. Strength of evidence: Mixed.
Caveat The author states power was only modest for detecting small differences, and the data predate smartphone-era dating apps.
The best-known result pointing the other way should be read with its funding in view.
Marriages that began online showed a slightly lower breakup rate, about 6% against 7.6%, and marginally higher satisfaction, 5.64 against 5.48 on the study's scale.
Direction: Decrease. Strength of evidence: Weak.
Caveat Commissioned by eHarmony, with the lead author a paid eHarmony adviser; the differences are small, observational and vulnerable to selection into online dating.
Nothing measured beforehand predicts how a relationship goes
Matching services rest on a premise that has been tested directly and did not hold: a consortium reanalysed dozens of couples studies with machine learning to see what forecasts relationship quality.
Across 43 datasets, partners' pre-relationship individual differences added no predictive power beyond a person's own reported experiences, and change in relationship quality was largely unpredictable from any combination of self-report variables.
Direction: No detectable effect. Strength of evidence: Strong.
Caveat Limited to self-report predictors already in existing datasets, so behavioural or unmeasured variables might still predict better.
If traits measured before a relationship starts do not forecast its quality, an algorithm sorting people by those traits has nothing to work with. No compatibility algorithm has been independently validated against relationship outcomes.
Loneliness is not clearly rising, whatever the framing suggests
Online relationships are usually discussed against a backdrop of assumed epidemic isolation. The cross-national baseline is worth having, with a note on who paid for it.
24% of people aged 15 and over report feeling very or fairly lonely, and the loneliest age group is 19–29-year-olds at 27% rather than the over-65s at 17%.
Direction: No detectable effect. Strength of evidence: Strong.
Caveat Single-item self-report, China excluded, and Meta co-funded the study.
In England the trend is flat to slightly improving: 6.6% of adults reported feeling lonely often or always in 2024/25, down significantly from 7.1% the year before but still above the 4.9% to 5.6% range of the middle 2010s (DCMS, Community Life Survey 2024/25, 2025). A 2022 review in Nature Reviews Psychology called the cross-temporal evidence for rising loneliness mixed, with several cohort studies showing stability or decline. Instruments and response rates changed across decades, which is the main reason no clean trend can be established — the same problem that limits what can be said about social media and well-being and about how communication shifted between channels.
The short version
- Meeting online accounted for 39% of new US heterosexual couples in 2017 and became the most common route around 2013 (Rosenfeld, Thomas & Hausen, 2019).
- Only 10% of all partnered US adults met their partner online, because most existing relationships began before the platforms did (Pew Research Center, 2023).
- Couples who met online split up at 7.0% a year against 7.3% offline, a difference the study cannot distinguish from zero (Rosenfeld, 2017).
- The one study finding an advantage for online-formed marriages was commissioned by eHarmony, whose paid adviser led it (Cacioppo et al., 2013).
- Harassment is concentrated rather than general: 56% of women under 50 who use the platforms have received unsolicited explicit messages (Pew Research Center, 2023).