A Market for Strangers

On 19 April 1965 the Harvard Crimson reported that more than 7,800 students at New England colleges, among them 900 Harvard men and 32 women from Radcliffe, were about to receive their matches. Each had filled in a 75-part questionnaire and mailed it with $3 to a company called Compatibility Research. The questions ran from “Which means most to you: prestige, security, power, or money?” to IQ, hair colour and sexual experience. Two undergraduates, Jeff Tarr and Vaughan Morrill, fed the answers into an IBM 1401 rented by the hour in Roxbury, and the machine paired people who had answered alike.¹ They called it Operation Match. The founders later said they had been discussing the irrationality of two social evils, the blind date and the mixer.²

Sixty years later, about two in five American heterosexual couples, and about two in three same-sex couples, meet online.³ One company, Match Group, owns Tinder, Hinge, Match.com, OkCupid and Plenty of Fish, and took in $3.49 billion in 2025.⁴ Tinder says it has made more than 100 billion matches.⁵ Americans reported losing about $1.48 billion to romance scams in 2025.⁶ And in 2025, for the first time, Tinder’s revenue fell.⁴

This essay follows how the questionnaire in Roxbury became the swipe on a phone, through the decisions that shaped it: to match people who are alike, to design for women, to sell science, to give matching away, to require that both people say yes, to charge for relief from limits the free product imposed, to promise to be deleted. Its deeper subject is what happens when the search for a partner, which for most of history was carried out by families, friends and neighbours, is handed to a market of strangers run by companies that are paid for the searching. Two questions run underneath. Can anyone, human or machine, know in advance who will love whom? And what does choosing from a menu do to the people who choose and the people chosen?

Before the machine

The commercial search for a spouse is older than the newspaper. In July 1695 John Houghton, a London apothecary who sold space to advertisers, printed a notice from a man who wanted “some Good Young Gentlewoman, that has a Fortune of 3000 l. or thereabouts”. Houghton added that such advertisements might prove useful. The date is the conventional one for the first printed personal advertisement in Britain; others are claimed, and the exact wording survives only in later quotation.⁷ What matters is that the first known ad already named a price.

For the next two centuries most matches were still made by families and communities, but the lonely-hearts column grew alongside them. Commercial bureaus followed. In April 1939 two women aged 24, Heather Jenner and a partner who called herself Mary Oliver, opened the Marriage Bureau at 124 New Bond Street. Their first clients were largely men home on leave from the empire, soldiers, planters and colonial servants, who had little time and few local contacts.⁸ It was a thin market: people who wanted to marry but did not meet enough others to choose from. Every later technology in this history has been sold as a cure for it.

Courtship itself had already moved out of the house. The historian Beth Bailey describes how, in early twentieth-century America, “calling”, in which a young man visited a young woman at her family’s home, gave way to “dating” in restaurants, theatres and dance halls, paid for by the man.⁹ Moira Weigel later defined dating as the form courtship takes when it happens in the free market.¹⁰ Each step took the search further from the family’s supervision and closer to commerce. Software continued a move that was already under way.

The computer as matchmaker

Operation Match was not the first computer dating service. Mar Hicks has shown that Joan Ball, who grew up in London’s working-class East End and already ran a marriage bureau, made her first computer run to pair clients in 1964 under the name St James Computer Dating Service. It merged the following year with another bureau run by women and became Com-Pat.² Its clients were older than Harvard’s, many divorced or widowed.

What these services assumed is more interesting than which came first. Com-Pat’s director said the computer could “eliminate the embarrassment of introducing people with the wrong background and nationality and politics and religion”, and users could list what they “would not tolerate” in a partner. Hicks’s reading is that early computer matchmaking “usually operated on a principle of shunning, and even fearing, difference”, and that the goal remained heterosexual marriage.² The machine was used to keep like with like, faster.

Hicks makes a second point that the industry would take decades to absorb. The questionnaires, she writes, were “of little importance in comparison with the user base upon which they operated.”² A service that recruited a homogeneous pool had solved most of the matching problem before the computer ran. Operation Match grew because of the pool, not the program: about 70,000 students sent questionnaires in its second year, and by 1968, when the business was sold, its founders counted more than a million respondents.¹¹ Tarr later called it a fad whose interest “died out”.¹² In New York, Project TACT, short for Technical Automated Compatibility Testing, aimed at wealthy singles on the East Side, sent women five blue cards and men five pink ones, signed up about 5,000 people and faded within a few years.¹³ In Britain, John Patterson’s Dateline, founded in 1967 after he saw a Harvard matching service, bought Com-Pat in 1974 and ran until his death in 1997.²

Hicks contrasts two business models visible by the early 1970s: Com-Pat’s fewer, marriage-minded pairings and Dateline’s volume. The same choice returns in every era of this history.

Video, chat and the Minitel

In Los Angeles on 14 February 1976 a videographer named Jeffrey Ullman opened Great Expectations, where members recorded video interviews and browsed each other’s tapes. A meeting happened only if both people agreed.¹⁴ Ullman later claimed thousands of marriages; the figures are his own and differ between accounts. The rule of mutual consent is the one to remember.

The first mass market for meeting strangers through a screen was not about compatibility at all. In France the state telephone company’s Minitel terminals, from 1984, carried a kiosk of paid services, and the messageries roses, anonymous chat lines, many of them sexual, became among the busiest. A French account puts their late-1980s revenue for the operator at several hundred million francs a year.¹⁵ On CompuServe’s CB Simulator, two users married in a ceremony held online on Valentine’s Day 1983.¹⁶ People wanted to talk to strangers before anyone knew how to match them.

Designing for women

Gary Kremen founded Electric Classifieds in San Francisco in 1993 with Peng Tsin Ong, who built the database, and bought the domain Match.com for $2,500. The site went live in April 1995. Early users were mostly in San Francisco, and men outnumbered women four to one.¹⁷ Kremen’s reported principle was blunt: “You have to design the whole system for women, not men.”¹⁸ Fran Maier, who joined in late 1994 and became chief executive in 1996, wrote the first profile questionnaire, removed a question about body weight and argued against charging per message in favour of membership.¹⁹

This was the first deliberate decision about a dating platform as a two-sided market, the kind of business discussed on this site’s page Markets With Two Sides. The men already online would pay, but they would pay only if women were there, and women would come only if the site felt safe and respectful. Match.com accepted slower growth among the paying majority in order to win the scarcer side.

The company was sold to Cendant over Kremen’s objections, for $7 million or $8 million depending on the account, and in May 1999 Cendant sold it to Ticketmaster Online-CitySearch, part of what became IAC, in a stock deal announced at about $50 million and valued at about $43 million when it closed in June. Match then had 1.8 million registered users.²⁰ The subscription model it settled on, free to browse and paid to message, would define the industry for a decade.

Stigma was the main obstacle. In a Pew survey in 2005, 29 per cent of American internet users agreed that people who used online dating were “desperate”, and 11 per cent had visited a dating site.²¹ By 2013 the “desperate” figure had fallen to 21 per cent and 59 per cent called online dating a good way to meet people.²²

Twenty-nine dimensions

eHarmony, launched in 2000 by the clinical psychologist Neil Clark Warren with his son-in-law Greg Forgatch, sold the opposite of browsing. Users answered a questionnaire of several hundred items (sources give 400, 436 and 480) and received matches chosen on what the company called 29 dimensions of compatibility, derived from Warren’s work on married couples.²³ The pitch was that science could do what friends and luck could not.

It matched only men with women. In 2005 Eric McKinley filed a complaint with New Jersey’s Division on Civil Rights, which found probable cause in 2007. In November 2008 eHarmony settled without admitting a violation, agreeing to launch a same-sex service, Compatible Partners, by 31 March 2009, with separate pools that could not be matched across. A California class action settled in January 2010 with payments to users and a requirement to link the two sites.²⁴ The exclusion followed from the founders’ assumptions about whose relationships their research described; the law treated it as discrimination.

The scientific claim fared no better. In 2012 Finkel and four other relationship scientists reviewed more than 400 studies for Psychological Science in the Public Interest. Online dating, they found, gave people access to far more potential partners than traditional methods, and online communication could build intimacy between strangers. But browsing many profiles could lead people to “commoditize potential partners”, and “there is little evidence that these algorithms can predict whether people are good matches.” Finkel put it more sharply: “To date, there is no compelling evidence that any online dating matching algorithm actually works.”²⁵

A year later a paper in PNAS reported, from a survey of 19,131 Americans married between 2005 and 2012, that more than a third of marriages had begun online and that those couples had slightly fewer break-ups (5.96 per cent against 7.67) and slightly higher satisfaction. eHarmony had commissioned the survey; the lead author, J. T. Cacioppo, was an adviser to the company, and a co-author had directed eHarmony Labs. Independent statisticians checked the analysis, an agreement guaranteed publication whatever the results, and the authors called the effects small.²⁶ The paper said nothing about whether eHarmony’s algorithm worked. Pew’s figure for the same period, that 6 per cent of internet users in committed relationships had met online, measured a different population with a different method, and the two are hard to reconcile.²²

Free, and almost free

The challenge to subscriptions came from people who gave matching away. OkCupid was launched in 2004 by four Harvard graduates who had built SparkNotes, among them Sam Yagan and Christian Rudder. It was free, funded by advertising, and built on questions users wrote and answered for each other, which produced a match percentage.²⁷ PlentyOfFish was begun in Vancouver in 2003 by Markus Frind, then 24, who ran it from home on Google’s advertising and was its only employee until about 2008, when his revenue was reported at about $10 million a year.²⁸

Match’s answer was to own both models. In February 2011 it bought OkCupid for $50 million in cash plus performance payments; that year it also raised its stake in the European service Meetic to about 80 per cent through a tender costing $272 million.²⁹ IAC’s annual report for 2011 described a portfolio of “both subscription-based and advertising-supported offerings” serving about 2.7 million subscribers, and said its revenue came “principally from subscription fees”.²⁹ PlentyOfFish followed in 2015 for $575 million in cash.³⁰ The free challengers became brands inside the incumbent.

A subscription dating service has an awkward incentive, and commentators named it early. Every couple that forms leaves, so a successful match loses the service two customers.³¹ Match Group rejects the argument, saying its business depends on people having good outcomes, which bring new users by word of mouth.³² No study has measured how large the effect is. But the incentive shapes how every decision in the rest of this history is read.

We experiment on human beings

OkCupid’s founders published what their data showed. Rudder’s blog, OkTrends, and his book Dataclysm (2014) reported that on OkCupid black women received lower ratings and fewer messages from men of other races, and Asian men fared worst among men.³³ Critics questioned the sampling, but the pattern recurred across years.

In July 2014 Rudder disclosed experiments under the title “We Experiment On Human Beings!”. In one, pairs who were actually a 30 per cent match were told they were a 90 per cent match. They sent more first messages and were more likely to keep talking. When actual 90 per cent matches were shown as 30 per cent, they did worse than they otherwise would have, though true 90s still did better than true 30s whatever the label. Users were told the correct figures afterwards. Rudder’s summary was that “the mere myth of compatibility works just as well as the truth.”³⁴ Reports differ on the exact conversation rates, about 10 per cent against 17 per cent, and a critic noted that success was defined as an exchange of four messages, which measures trust in the number more than attraction.³⁵ The experiment showed something Hicks had found in the 1960s: the score was doing part of the work it claimed only to measure.

The grid

The phone changed what a dating service could know. Grindr opened on Apple’s App Store on 25 March 2009, built for Joel Simkhai in Los Angeles after the iPhone gained GPS. It showed nearby men in a grid ordered by distance, nearest at the top left.³⁶ Gay men had used websites such as Gaydar to meet for years, and the population any one man was looking for was smaller and harder to find in public; Grindr made the question “who is near me now?” the whole interface. Badoo, launched in 2006 by Andrey Andreev, grew a freemium social-discovery business out of his Russian dating site Mamba.³⁷ Location, not compatibility, became the first filter.

Matchbox

Tinder came from inside IAC. In early 2012 Sean Rad, who had been hired to run a loyalty app at IAC’s Hatch Labs incubator, and an engineer, Joe Munoz, built a prototype called Matchbox at an internal hackathon. Its central rule was a “double opt-in”: nobody could message anybody until both had said yes.³⁸ The first version had buttons. Jonathan Badeen, who joined that March, built the swipe in the autumn of 2012; he later said he had “a nagging desire to gamify” the stack of profile cards.³⁹

The double opt-in was the decision that mattered. On open messaging sites women received far more messages than they wanted, and many were unpleasant; the rule made every conversation one both people had asked for. It was Great Expectations’ rule of mutual consent, run at the speed of a thumb. The swipe made it fast and a little like a game, and Tinder was free. By 2014 it reported more than a billion swipes and 12 million matches a day.³⁸

The early company was turbulent. Whitney Wolfe, who said she had named the app and was designated a co-founder in 2012, sued in 2014 alleging harassment and discrimination; the case settled that September without admission of wrongdoing.⁴⁰ Years later Rad and other early employees sued IAC and Match over a 2017 valuation of Tinder; Match agreed in December 2021 to pay $441 million, the day before closing arguments.⁴¹

Paying for relief

Tinder stayed free for about two and a half years. On 2 March 2015 it launched Tinder Plus, which offered unlimited likes, the ability to undo a swipe and to swipe in other cities. Free users now got a limited number of right swipes and had to wait 12 hours when they ran out; Tinder said the limit would encourage people to be honest about whom they liked. Plus cost $9.99 a month for users under 30 in the United States and $19.99 for those 30 and over.⁴²

Each later tier sold something the free product withheld. Super Like (October 2015) let a user signal special interest. Tinder Gold (August 2017) added “Likes You”, a list of the people who had already swiped right, and Match told investors it had raised conversion across Tinder.⁴³ Platinum, announced in 2020, sold priority.⁴⁴ An interpretation suggests itself: the double opt-in was built to hide who liked whom until both said yes, and Gold sold exactly that hidden information back to the users willing to pay.

Pricing by age was challenged in court. In January 2018 the California Court of Appeal revived a class action, Candelore v. Tinder, holding that age-based pricing as alleged violated the state’s Unruh Civil Rights Act. Tinder denied wrongdoing and agreed to a $60.5 million settlement covering about 268,000 Californians who paid at 29 or older; it received preliminary approval in early 2026.⁴⁵

Tinder also ranked its users. In January 2016 Rad confirmed to Fast Company an internal “Elo score”, named after the chess rating, which he called a measure of “desirability”: being liked by someone who was much liked counted for more. A data analyst described every swipe as “casting a vote”.⁴⁶ In March 2019 Tinder wrote that “Elo is old news” and that its technology no longer relied on it, without saying in detail what replaced it.⁴⁷ Researchers found the hierarchy without help from the company. In a study of messages on a free dating site in four American cities, Bruch and Newman measured desirability by who received messages from desirable people, found a hierarchy “remarkably consistent across cities” that correlated with age, race and education, and found that men and women alike wrote to people about 25 per cent more desirable than themselves.⁴⁸ A study of Tinder using test profiles found that men liked a much larger share of the profiles they saw, matched far less often and became less selective in response, while women became more selective; male test profiles matched on about 0.6 per cent of likes and female ones on about 10 per cent.⁴⁹

Women first

Wolfe, now Wolfe Herd, launched Bumble in December 2014 with Andreev’s Badoo, which took about 79 per cent for an initial investment. Its rule was the most distinctive in the industry: in a match between a man and a woman, only the woman could send the first message, and the match expired if she did not within 24 hours.⁵⁰ The rule answered the same complaint as Tinder’s double opt-in, the flood of messages women received, by giving women control over whether a conversation began at all.

Match tried to buy Bumble in 2018, reportedly for $450 million, then sued for patent and trademark infringement; Bumble countersued, and both suits were dropped. In 2019 Blackstone bought Andreev’s stake at a valuation of $3 billion, and in February 2021 Bumble went public at $43 a share, raising $2.2 billion.⁵¹

Designed to be deleted

Hinge began as an app that showed friends of friends from Facebook. On 27 September 2016 it announced it would retire swiping. Its founder, Justin McLeod, said: “Most of the dating apps out there are games designed to keep you single, and swiping is the most iconic feature of these games.” The relaunch in October offered fuller profiles, no swiping and a fee of $7 a month, part of which it soon rolled back.⁵² Its slogan became “designed to be deleted”.⁵³

In July 2018 Hinge added Most Compatible, a daily suggestion based on the Gale–Shapley algorithm, the method for finding stable pairings that two mathematicians published in 1962 and that later won a Nobel prize for Lloyd Shapley and Alvin Roth. Hinge adapted it to a single pool rather than two sides and fed it with each user’s recent activity, and said users in early tests were eight times more likely to exchange numbers with that match.⁵⁴ The figure is the company’s own.

Match took 51 per cent of Hinge in June 2018 and the rest early in 2019.⁵⁵ The app that called swiping a game designed to keep people single now belonged to the company that owned Tinder. Hinge’s revenue rose 44 per cent in 2022, to about $396 million in 2023 and $550 million in 2024.⁵⁶

One company, many doors

Match Group went public on 19 November 2015 at $12 a share, the bottom of its range, with 2014 revenue of $888 million.⁵⁷ IAC completed its separation from Match in July 2020.⁵⁸ Revenue reached $2.4 billion in 2020 and $3.2 billion in 2022. Tinder produced about 56 per cent of it, and its paying users averaged 10.9 million in 2022; in 2023 Tinder’s payers fell while revenue per payer rose, after what the company called pricing optimisations.⁵⁹

The fight over app-store fees, discussed in this site’s essay on app stores, hit dating apps directly because they sell subscriptions inside apps. Match sued Google in May 2022 over Play billing and settled before trial in 2023 on lower rates.⁶⁰ Facebook Dating arrived in the United States in September 2019 with a feature that revealed a crush on a friend only if it was mutual.⁶¹ In China, Momo bought Tantan, often called China’s Tinder, in 2018 for $600.9 million in cash plus shares; in India, Info Edge, owner of the matrimonial site Jeevansathi, bought most of the dating app Aisle in 2022.⁶² India’s matrimonial sites, Shaadi.com (begun as Sagaai.com in 1997) and BharatMatrimony (1997), followed an older model in which the goal was marriage and families took part.⁶³

The pandemic tested the format. On 29 March 2020, with much of the world at home, Tinder recorded more than 3 billion swipes in a day for the first time.⁶⁴

By then the change in how people met was measurable. Rosenfeld, Thomas and Hausen, using nationally representative surveys, found that the share of American heterosexual couples who met online rose from zero before 1995 to about 22 per cent in 2009 and about 39 per cent in 2017. Around 2013 meeting online overtook meeting through friends, which had been the leading way since the end of the Second World War. Meeting through family, church, the neighbourhood and school had been declining since 1940. Their test showed that online dating displaced friends instead of working through them, and that how couples met made no significant difference to whether they broke up.³

Who is on the other side

A market of strangers needs ways to know whom one is meeting. The industry’s record here is mixed.

Fraud came first. In September 2019 the Federal Trade Commission sued Match Group, alleging among other things that Match.com had sent non-paying users notices that someone was interested in them, prompted by accounts it had flagged as likely fraudulent, in order to sell subscriptions. In 2022 a court dismissed the counts about fake-profile communications under Section 230 of the Communications Decency Act, which shields online services from liability for what their users post. The rest settled in August 2025: Match paid $14 million in redress and agreed to clear disclosure of its six-month guarantee, simple cancellation and no penalties for users who disputed charges, without admitting liability. Between 2013 and 2016 nearly 2.5 million subscriptions had carried the guarantee, and 32,438 free six-month packages had been given out under it.⁶⁵ Ashley Madison, a site for people seeking affairs, was breached in 2015, exposing about 36 million accounts, and settled with the FTC and states in 2016; the FTC found that staff had created fake female “engager” profiles.⁶⁶

Romance scams spread far beyond dating apps. Americans reported losing $1.3 billion in 2022 and about $1.48 billion in 2025; the FTC says social media, not dating apps, is the most common starting point, and that reported losses understate the total.⁶ The United Nations Office on Drugs and Crime estimated in 2025 that scam centres in Southeast Asia, staffed partly by trafficked workers, make just under $40 billion a year.⁶⁷

Physical safety was a harder problem. In December 2019 Columbia Journalism Investigations, ProPublica and BuzzFeed News reported that Match.com had screened paying subscribers against sex-offender registries since a 2011 lawsuit, but that Tinder, OkCupid and Plenty of Fish, the free apps, did not. A Match spokesperson said: “there are definitely registered sex offenders on our free products.” Match later called the reporting inaccurate and said free apps did not collect enough information for meaningful checks.⁶⁸ In February 2025 the Dating App Reporting Project, an 18-month investigation by The Markup with the Pulitzer Center, published with The Guardian and The 19th, reported that Match kept on its apps users it knew had been accused of sexual violence. Its central case was a Denver cardiologist, Stephen Matthews, first reported to Hinge in 2020, who remained on Match apps until 2023 and was sentenced in October 2024 to 158 years to life for crimes against 11 women. Match pointed to its AI tools, ID verification and a portal for police. In December 2025 six women sued Match Group in Denver, alleging among other things that a user could unmatch a victim before she reported him, after which the report option disappeared.⁶⁹ The case was pending at the time of writing.

The law had mostly protected the platforms. In Herrick v. Grindr a man whose former partner created fake profiles that sent hundreds of strangers to his home and work sued Grindr for defective design; in 2019 the Second Circuit held the claims barred by Section 230.⁷⁰ The Denver suit frames the harm as a product defect in an attempt to get around that ruling.

Privacy was the third front. In 2018 researchers found that Grindr had sent users’ HIV status to two analytics companies; Grindr said it would stop. Norway’s data protection authority fined it NOK 65 million in 2021 on the reasoning that using Grindr reveals a person’s sexual orientation, a special category of data, and the fine was upheld through the Court of Appeal in October 2025. A British group claim over the same period settled in 2026 for £26 million without admission of liability.⁷¹ In 2021 a senior official of the American Catholic bishops’ conference resigned before a publication reported that commercially bought app-location data showed his device using Grindr.⁷²

The platforms’ answers came in layers: photo verification and a panic button on Tinder in 2020; then, in October 2025, Face Check, a mandatory video selfie that checks a new user is live, matches their photos and detects duplicate accounts, starting in California and seven countries.⁷³

The bill for attention

Researchers have asked what all this swiping does to the people who do it. Their answer is more cautious than the popular one. Studies fairly consistently link dating-app use to worse body image; links to depression and anxiety appear in about half of studies. A review of 45 studies reported in 2025 found negative associations with body image in 19 of 22 and with mental health in 14 of 29. Almost all are snapshots in time, so they cannot say whether apps make people unhappy or unhappy people use apps more.⁷⁴ A widely quoted figure that 78 per cent of users feel burned out comes from a commercial poll with a broad definition.⁷⁵

Lawsuits tested the idea that the apps were built to be compulsive. In February 2024 six users sued Match, alleging that Tinder, Hinge and The League were designed to be addictive; Match called the claims “preposterous”, and in December 2024 a federal judge sent the case to arbitration, where it remains out of public view.⁷⁶ Regulators moved on narrower ground. Britain’s Online Safety Act required age checks on dating apps from July 2025; Ofcom found a year later that more than one in ten 15- to 17-year-olds still visited the biggest dating apps.⁷⁷ Australia’s industry code for dating apps became enforceable in April 2025.⁷⁸ On the day this essay was published, 11 November 2026, the European Commission presented a Digital Fairness Act that would ban listed manipulative designs, tighten rules on subscriptions and free trials, and let users switch off addictive features, with fines of up to 6 per cent of turnover; it still needs the approval of the Parliament and the member states.⁷⁹

Fatigue

Tinder’s direct revenue peaked at $1.94 billion in 2024, up 1 per cent, while its paying users fell 7 per cent. In 2025 its revenue fell about 4 per cent, to about $1.9 billion, its first annual decline, and its payers ended the year at 8.8 million, down 8 per cent. Match Group’s total was flat, held up by Hinge.⁴ In Britain, Ofcom found that the share of adults visiting any dating site held steady at about 11 per cent, but Tinder’s reach fell from 5.2 per cent in May 2023 to 3.1 per cent in May 2025, level with Hinge, and among women Tinder fell to third behind Hinge and Bumble. Men outnumbered women on Tinder by about three to one.⁸⁰

That ratio is the mechanism of decline the industry had always feared, described on this site as network effects running in reverse: when women leave, the app is worse for the men who stay, and they leave too. Pew’s last large survey, in July 2022, found that 54 per cent of women who had used dating apps felt overwhelmed by the number of messages they received, while 64 per cent of men felt insecure about how few they got, and that 56 per cent of women users under 50 had been sent sexually explicit messages or images they did not ask for.⁸¹ Thirty per cent of American adults had used a dating site or app, and 10 per cent of partnered adults, 20 per cent of those under 30, had met their partner that way.⁸¹

Match changed its leadership. Spencer Rascoff became chief executive in February 2025 and in May cut 13 per cent of staff, including one manager in five.⁸² In December 2025 McLeod stepped down at Hinge to start Overtone, an AI and voice dating service in which Match took a large stake.⁸³ Tinder’s first product keynote, in March 2026, introduced Chemistry, an AI layer that replaces endless liking with “a daily curated recommendation” drawn from questions and, if a user agrees, a scan of their phone’s photos; new “modes” for music and astrology; group dates; video speed dating; and AI filters for harassing messages.⁵ In November 2026 Match reported that Tinder’s daily users had grown year on year in September “after more than three years” of decline, and that Hinge’s revenue was growing at about 19 per cent a year, on course for $1 billion in 2027; total revenue was falling slightly.⁸⁴

Bumble’s fall was steeper. Wolfe Herd returned as chief executive in March 2025, cut 30 per cent of staff and pursued what she called “a deliberate reset of our member base”, choosing “quality over quantity”. Revenue fell about 10 per cent in 2025 and by 15 to 16 per cent a quarter in 2026, while profits rose. On 13 October 2026 Bumble agreed to be taken private by Ithaca Holdings at $6.05 a share, about $1.3 billion, against its offering price of $43 in 2021.⁸⁵

Elsewhere the same signs appeared. In China, paying users of Tantan fell from 0.9 million to 0.6 million in a year.⁸⁶ The Tokyo metropolitan government launched its own AI-assisted matchmaking app in September 2024 and reported 265 marriages by June 2026, with no comparison group to show what difference the app made.⁸⁷

The agents at the door

Generative AI reached dating from both sides. A Match Group and Kinsey Institute survey reported that 26 per cent of American singles, and about half of those in Generation Z, had used AI for help with dating, from choosing photos to writing messages.⁸⁸ Surveys of romantic conversation with AI companions range from about 19 to 28 per cent of American adults, depending on how the question is asked.⁸⁹ Startups offered AI matchmakers that charge per introduction or, in one case, ask users to pay $50,000 if the service finds their spouse.⁹⁰

In November 2026 Rascoff said that Tinder and Hinge would not allow AI agents to review profiles or message matches on a user’s behalf, and urged other companies to think twice; if agents did the talking, he said, there was little chance the people would connect authentically and end up meeting.⁹¹ The company’s own AI, which recommends and coaches, continues. Of AI coaching that happens outside the apps, he told Bloomberg: “there’s not much we can do about that.”⁸⁸ Hinge’s new chief executive, Jackie Jantos, said: “it’s really important for humans to be humans.”⁸⁸

What online dating is today

Online dating in 2026 is the way most new couples in rich countries meet, and an industry that many of its users say they are tired of. One company, Match Group, owns most of the large Western brands and takes about $3.5 billion a year, more and more of it from Hinge. Bumble is leaving the stock market. Grindr, the largest app for gay men, rejected a take-private offer from its largest shareholders in 2025.⁹² In India the matrimonial sites keep families involved; in China and Japan, governments and local giants run their own services.

The product’s form is the result of the decisions above. A profile is a set of photographs and a few prompts, because Match.com and OkCupid showed that people judge on little and the phone made photographs the cheapest signal. Messages wait for mutual consent, because open messaging drove women away. The free tier rations, and the paid tiers sell relief from the rationing and information about who is interested. Ranking is hidden, because people object to being ranked. Matching science is claimed more cautiously than eHarmony claimed it, because the evidence never arrived. Safety is a mix of verification, filters and a reporting system the courts are now examining. And AI is arriving as a curator, which the companies welcome, and as a stand-in for the person, which Match has decided to refuse.

How online dating changed courtship

From introduced by friends to introduced by a company

For most of the twentieth century a partner was most often met through friends, who vouched for each other and knew both sides. Online dating displaced them, in Rosenfeld’s word “disintermediating” them, and put a company in their place. The company knows far more people than any friend, and far less about each of them, and it is paid differently.

From a circle to a menu

A person’s choices used to be bounded by the people they happened to meet: the thin market of the Marriage Bureau’s soldiers home on leave. The apps thickened the market enormously and presented it as a sequence of options, one after another, to be judged in seconds. The experience moved from encountering someone to choosing among many.

From open contact to mutual consent

Personal ads, Match.com and OkCupid let anyone write to anyone. Tinder made a conversation require two yeses and Bumble gave the first word to women. This was the industry’s most important change for the people, usually women, who had borne the cost of open contact, and it changed what a first message means: an invitation already half accepted.

From the person to the profile

What a stranger first sees is a few photographs and a few sentences, built for the purpose. Profiles have always been performances, but the swipe made the first judgement almost entirely visual and made the performance continuous, tested and, increasingly, written with help from AI.

From being chosen to being ranked

Families and friends sorted people too, but quietly. The apps measured desirability, first openly, as Tinder’s Elo, then less visibly, and researchers found the hierarchy in the data whether the companies scored it or not. Many users now experience dating as a market in which they have a price, and adjust their aims accordingly.

What the history suggests

Several of this site’s patterns recur. Who pays shapes what gets built: subscriptions made retention the measure of success in a product whose success is losing users; free tiers built the volume and paid tiers sold relief from the free tier’s limits; screening against sex-offender registries followed payment. Incumbents absorb the challenger’s best idea: Match bought OkCupid and PlentyOfFish, Tinder was built in its parent’s own incubator, and in 2018 it began buying Hinge, the app that had defined itself against swiping. Taking choices away can be the product: the double opt-in, Bumble’s first-message rule and Hinge’s single daily match each restricted what users could do, and each was the feature people came for. Network effects run in both directions: Tinder’s three-to-one ratio of men to women is the mechanism of its decline. Defences against abuse concentrate the market: verification at scale, Face Check and AI moderation are costly, and favour the largest operator.

What sets online dating apart is the good being traded. Every other product in this series moves information, money, goods or attention. This one introduces people to each other with a view to love, the relationship people most want to have chosen freely and least want to feel was bought. Com-Pat’s director thought a questionnaire could spare people the embarrassment of difference; Rudder found that a number on a screen could start a conversation by itself; Rascoff decided that a first conversation had to be typed by the person having it. Each was a judgement about how much of choosing a partner can be organised by someone else, and what is lost when it is. The philosophy starts from that question.

Understanding the philosophy

The thinkers below are connected to online dating by the author’s interpretation. Most of them wrote before anyone swiped, and several wrote about markets, gambling or love in general rather than about dating. The aim is to use their ideas to see the decisions in this history more clearly, not to claim that they foresaw them.

1. Too many to choose from

The idea in plain terms

More options are supposed to make a chooser better off: if the best choice from ten is good, the best from a thousand should be better. The idea of choice overload says that past some point this fails. A large set makes choosing harder, makes people less satisfied with what they choose, and can make them put off choosing at all. In dating, the worry is sharper still, because the options are people and the act of passing over them may change the person doing it.

The tradition

The idea’s modern history begins with jam. In 2000 Sheena Iyengar and M. R. Lepper reported a field study in an upmarket grocery store: a tasting booth showed either 6 jams or 24. The larger display drew more people to stop, but about 30 per cent of those who stopped at the small display bought jam against about 3 per cent at the large one.⁹³ Barry Schwartz built the finding into The Paradox of Choice (2004), which argued that abundance breeds regret and paralysis, especially in “maximisers” who must find the best.

Then the evidence wobbled. Scheibehenne, Greifeneder and Todd pooled 63 conditions from 50 experiments, more than 5,000 participants, and found a mean effect “virtually zero”, with large variation between studies.⁹⁴ Attempts to replicate the jam study failed. In 2015 Chernev, Böckenholt and Goodman reanalysed the field and found that overload is real under particular conditions: when the set is complex, the task difficult, preferences uncertain and the chooser keen to save effort.⁹⁵ Choosing a partner meets at least the third condition more fully than almost anything else people choose.

Dating studies point the same way. Lenton and Francesconi studied 84 British speed-dating events and found that when there were more people to choose from, choosers paid more attention to things quick to judge, such as height and weight, and less to things that take time, such as education and occupation.⁹⁶ D’Angelo and Toma had 152 students choose from 6 or 24 profiles; a week later, those who had chosen from 24 were less satisfied with their choice and more likely to want to change it.⁹⁷ Pronk and Denissen ran three experiments in which people went through a series of dating profiles; the further they went, the more likely they were to reject, as satisfaction with the photographs and their own sense of success fell. They called it “a rejection mind-set”.⁹⁸

Where it shows up

The whole arc of this history is a movement from scarcity to abundance. The Marriage Bureau’s clients needed more people; Tinder’s billion daily swipes supplied them. Finkel’s review in 2012 warned that browsing many profiles could lead users to “commoditize” partners.²⁵ Tinder’s own answer, the swipe limit of 2015, was presented as encouraging honesty, and its 2026 Chemistry feature replaces endless liking with one recommendation a day. Hinge’s single daily Most Compatible match is a product built on the same bet that fewer options can be better. The page Bounded Rationality and Satisficing discusses the older idea underneath: Herbert Simon’s point that real choosers stop at good enough, because they cannot search everything.

The strongest objection

The meta-analysis is a serious problem for anyone who wants to blame apps for indecision. If choice overload averages out to nothing across fifty experiments, the dating studies may be finding a weak, conditional effect and the popular story may be mostly anecdote. And the alternative to abundance was not a happy medium. For a gay man in a small town, a divorced woman of fifty or anyone whose circle was small, scarcity meant loneliness, and the large set was the point. Rosenfeld found no evidence that couples who met online broke up more often.³

What it means today

The defensible claim is narrower than the slogan. Large sets do not stop people choosing. They seem to change how people choose, toward quick visual cues, and how they feel afterwards, less sure and more inclined to keep looking, and long sessions build a habit of saying no. For people who build these products, the evidence points to limiting the number shown at once, and to designing the end of a search as carefully as its beginning: a moment to stop and commit, where now there is always a next card.

2. The stable marriage problem

The idea in plain terms

Matching people to people is different from selling goods, because both sides must choose each other. A matching is “stable” if no two people would both rather be with each other than with the partners they have, since such a pair would simply leave. Economists have studied what kind of procedure produces stable matches, and what markets need to work when prices cannot do the matching.

The tradition

In 1962 David Gale and Lloyd Shapley published “College Admissions and the Stability of Marriage” in the American Mathematical Monthly. They proved that for equal numbers of men and women with ranked preferences, a stable set of marriages always exists, and gave a procedure, deferred acceptance, for finding one: one side proposes, the other holds its best offer so far and rejects the rest, and the process repeats until nobody is rejected.⁹⁹ The version in which one side proposes is the best stable outcome for that side, which is why designers care who proposes. Shapley and Alvin Roth shared the Nobel memorial prize in economics in 2012 “for the theory of stable allocations and the practice of market design”.¹⁰⁰

The economist Gary Becker’s “A Theory of Marriage” (1973) asked what pattern of marriages a market would produce, and found that where partners’ traits complement each other, like will pair with like.¹⁰¹ Roth’s Who Gets What and Why (2015) set out what a working matching market needs: it must be thick, with enough participants; it must relieve the congestion thickness creates, giving people time to consider offers; and it must be safe, so that people can reveal what they really want without being punished for it.¹⁰² Hitsch, Hortaçsu and Ariely estimated preferences from users of an American dating site, computed the Gale–Shapley stable matches, and found them close to the matches the site actually produced.¹⁰³ Lee and Niederle gave users of a Korean dating site a few virtual “roses” to attach to proposals as a costly signal of special interest.¹⁰⁴

Where it shows up

The history reads well through Roth’s three conditions. The Marriage Bureau’s clients faced a thin market; the apps made it thick. Thickness produced congestion: women flooded with messages, men ignored. Tinder’s double opt-in and Bumble’s first-message rule were congestion controls. Super Likes and Hinge’s limited “roses” are costly signals in Lee and Niederle’s sense. Hinge’s Most Compatible borrowed Gale and Shapley directly. Bruch and Newman’s finding that people write to partners about 25 per cent more desirable than themselves, and are mostly ignored, is congestion in data.⁴⁸ The page Designing Rules People Will Not Game covers the wider field.

The strongest objection

The mathematics assumes what dating lacks. Gale and Shapley need each person to hold a complete, fixed ranking of everyone on the other side. A dater has no such list, and the research in the next section suggests one cannot be written before meeting. Stability is also a thin virtue for love: it means nobody wants to defect, not that anybody is happy. And an app is not a clearing house. Medical residency matching works because every hospital and doctor submits once and accepts the result. Dating apps run continuously, nobody is bound by a match, and the company’s revenue depends on people coming back.

What it means today

Market design is still the most useful lens on how dating apps fail, because most of their problems, from unread messages to fatigue, are failures of congestion and safety. Prediction is a smaller part of it. The practical lessons are Roth’s: keep the market thick, ration attention so that signals mean something, and make it safe to say what one wants. What the theory cannot supply is the preference list. Anyone selling a dating algorithm as a solution to the stable marriage problem is selling the easy half.

3. The chemistry no one can predict

The idea in plain terms

Every matching service since Com-Pat has implied that, given enough information about two people, someone can tell in advance whether they will be drawn to each other and whether they will last. Relationship scientists have tested the claim. Their answer, so far, is that the parts of attraction that are predictable are mostly about individuals, and the part unique to a pair is almost invisible before they meet.

The tradition

The 2012 review by Finkel and colleagues concluded that sites mostly measure similarity in personality and attitudes, which relationship science does not link strongly to lasting happiness, and that there was “little evidence” any algorithm could predict good matches.²⁵ Joel, Eastwick and Finkel then tested it directly. In two speed-dating studies, each person answered more than 100 questions and then went on four-minute dates. Machine-learning models predicted a fair part of how much a person desired others in general and how much they were desired in general. They predicted none of the desire unique to a particular pair. Joel said the team could not anticipate how much individuals would uniquely desire each other “with any meaningful level of accuracy”.¹⁰⁵

A larger study in 2020 pooled 43 datasets covering more than 11,000 couples. What predicted relationship quality was the couple’s own perceptions of the relationship: how committed each believed the other was, appreciation, sexual satisfaction, perceived partner satisfaction and conflict. Traits of the individuals added little once those were known, and changes in quality over time were largely unpredictable.¹⁰⁶ The things that predict a good relationship exist only once there is a relationship.

Where it shows up

eHarmony’s 29 dimensions were a bet against this result, and the PNAS study it commissioned was careful to say nothing about its algorithm. Hicks’s finding that the 1960s services succeeded through their pools, not their programs, is the same result found in an archive. OkCupid’s experiment adds a twist: telling a 30 per cent match that they were a 90 per cent match made them talk more. A score can produce some of the outcome it claims to forecast, which this site discusses as Predictions That Make Themselves True. Tinder, by contrast, made almost no claim to predict anything. Finkel’s review had said that online dating works insofar as it gets people to meet quickly and find out for themselves, which is close to a description of the swipe.

The strongest objection

Joel’s studies used students, short dates and self-reports. Apps now hold behavioural data on hundreds of millions of people, what they look at, whom they answer and how long they talk, and that may predict what questionnaires cannot. Collaborative filtering does not need to understand chemistry to notice that people who liked the same profiles tend to like each other’s. Hitsch and his colleagues found that stated and revealed preferences on a dating site could reproduce real marriage patterns. And a weak predictor applied across a huge pool can still raise the rate of good first dates.

What it means today

The claim that survives is modest: software can find people who are generally attractive to someone, filter out obvious mismatches, and increase the number of meetings. What it cannot yet do is see the spark between two particular people, and the evidence suggests that the most important predictors of a lasting relationship are built by the couple after they meet. For anyone designing or marketing a matching product, honesty about this is both an ethical and a practical matter. Promise introductions, not compatibility, and move people to meet quickly, because meeting is the only test that works.

4. A self made for strangers

The idea in plain terms

A dating profile is a person presented to people who do not know them, in a form designed to be judged quickly. It is a performance, shaped by what the presenter expects to happen next. The format shapes the audience too, who see the person as one card among many that look alike.

The tradition

Erving Goffman’s The Presentation of Self in Everyday Life (first published in 1956 and expanded in 1959) described social life as a performance, with front regions where the performance is given and back regions where it is prepared.¹⁰⁷ Ellison, Heino and Gibbs interviewed users of a large dating site in 2006 and found that the expectation of meeting face to face governed what they wrote: people managed impressions they would have to live up to.¹⁰⁸ Toma, Hancock and Ellison then measured 80 online daters’ height, weight and age against their profiles. Small lies were common, men’s mostly about height and women’s about weight; photographs were the least accurate element. The lies were small and strategic, because the daters knew they would be met.¹⁰⁹ Ellison, Hancock and Toma later described a profile as a “promise” about the person who would turn up.¹¹⁰

Martha Nussbaum’s essay “Objectification” (1995) listed seven ways of treating a person as a thing, among them instrumentality, denial of subjectivity and fungibility, treating someone as interchangeable with others of the same type. She also argued that some objectification is harmless in context.¹¹¹

Where it shows up

Maier’s removal of the weight question from Match.com’s first questionnaire in the 1990s was an early decision about what self-presentation should include. The swipe made the first judgement almost entirely visual and turned a profile into a card in a stack, close to Nussbaum’s fungibility. Lenton and Francesconi’s finding that larger sets shift attention to height and weight shows the format pushing judgement toward what a photograph shows. Face Check and verification badges are attempts to keep the promise enforceable. AI-written messages and AI-improved photographs stretch it: Tinder’s own Photo Enhance and users’ outside coaching raise the question of whose performance a profile is.

The strongest objection

People have always presented themselves selectively to potential partners. Clothes, manners and a friend’s introduction are performances too, and the 1695 advertiser’s fortune of £3,000 was a profile reduced to a single line. Toma’s findings are, in a way, reassuring: daters lie less than people assume, because they expect to meet. And a profile that someone can make carefully, at leisure, may present a shy person better than a crowded bar would.

What it means today

The promise holds only while the meeting that tests it is likely. Ellison’s daters stayed close to the truth because they expected to be seen. A format that delays or replaces meeting, with endless chat, AI-polished photographs or an agent that writes the messages, weakens the check that kept self-presentation honest. Designers can strengthen the promise by verifying the things that are cheap to verify, by moving people toward meeting, and by being cautious about tools that improve a profile beyond what its owner can live up to.

5. Exchange value

The idea in plain terms

When finding a partner works like a market, the language and habits of the market enter romance. People assess what they can “get”, given what they have to offer; they invest, compare, trade up. The idea, held in different forms by a psychoanalyst, a sociologist and a political philosopher, is that this changes love itself, or at least what people take love to be.

The tradition

The psychoanalyst Erich Fromm argued in The Art of Loving (1956) that in a market culture two people fall in love when each has found the best object available given the limits of their own “exchange values”, and that modern people trade “personality packages” and hope for a fair bargain. Against this he set love as an art to be practised, made of care, responsibility, respect and knowledge, and he distinguished the experience of falling in love from the lasting state of standing in it.¹¹²

The sociologist Eva Illouz gave the idea empirical form. In Cold Intimacies (2007) she described “emotional capitalism”, in which emotional and economic relations come to shape each other, and used internet dating sites as evidence. Why Love Hurts (2012) argued that romantic suffering is shaped by social forces, especially the rise of “marriage markets”, more than by individual psychology, and The End of Love (2019) described how relationships now fade and dissolve amid endless choice.¹¹³ Michael Sandel, in What Money Can’t Buy (2012), offered a test: markets can be objected to on grounds of fairness, when the poor are disadvantaged, and of corruption, when putting a price on a good degrades it by valuing it in a lower way than it deserves.¹¹⁴

Where it shows up

The first advertisement in 1695 named a fortune. Bruch and Newman’s desirability hierarchy, Tinder’s Elo score and the experience of many users that they have a price are Fromm’s exchange values made measurable. Sandel’s two tests fit the paid tiers. Tinder’s age-based prices raise the fairness question, and a court agreed they might be unlawful. Selling “Likes You”, knowledge of who wants you, raises the corruption question: the hope that someone is interested, which mutual consent kept private, becomes a product.

The strongest objection

Marriage was a market long before software, and a harsher one. For most of history matches were arranged around land, dowries, class and family interest, and individual feeling counted for little. Becker’s economics of marriage describes sorting that happened without any app. Measured against arranged marriage, online dating gives individuals more say than they have ever had, and its “market” language may describe the freedom to choose more than it corrupts love. Rosenfeld’s data show couples who met online doing as well as others.

What it means today

The objection is right about history and does not answer the present concern, which is about experience. A user who spends years being ranked and ranking others may come to see themselves as an asset with a price, and that self-image can travel from the app into the relationship. Products can lean into that or against it. Showing scores, selling information about who wants whom and pricing by personal characteristics lean into it. Rationing attention, hiding rank and moving people toward meeting lean against it. The choice is the company’s, and the test Sandel offers is a fair one to apply.

6. Falling without risk

The idea in plain terms

To love someone is to expose oneself to a person one cannot control, who may change, disappoint or leave. This exposure may be part of what love is. If so, a product that promises love without risk is promising something else.

The tradition

The philosopher Alain Badiou opened In Praise of Love (2009), revised from a public conversation at the Avignon festival, with the advertisements of the French dating site Meetic, which in translation promised “Be in love without falling in love!” and “Get perfect love without suffering!” He compared the promise to military talk of wars with “zero dead”: an attempt to secure in advance against the encounter with difference that love is.¹¹⁵ The philosopher Byung-Chul Han argued in The Agony of Eros (2012) that eros requires a real Other, whom one cannot reduce to oneself, and that a culture of comparison and self-optimisation levels everyone into “the same”.¹¹⁶

An older tradition answers the risk with commitment. Søren Kierkegaard’s Works of Love (1847) argued that love made a duty, and not left to inclination, is secured against change, and warned that loving someone because they are more distinguished than others can be a form of self-love.¹¹⁷ Harry Frankfurt, in The Reasons of Love (2004), held that love is not necessarily a response to the beloved’s value: what we love acquires value for us because we love it.¹¹⁸ On this view, the reasons for loving a particular person come after the loving, not before.

Where it shows up

eHarmony’s promise of scientific compatibility was Meetic’s promise in a lab coat: love chosen to minimise risk. The ever-present next card makes every relationship revocable before it begins; Illouz’s “unloving” and Pronk’s “rejection mind-set” describe what that may do to commitment. Frankfurt’s point turns the logic of search around. If love creates the beloved’s value, then a search that waits to find enough value before committing has the order wrong, and the person who keeps swiping in search of the right reasons may never find them in a profile.

The strongest objection

The romantic praise of risk is easy for people who are not at risk. Women meeting strangers face real dangers, as the Matthews case showed, and verification, mutual consent and screening are reductions of risk that nobody should be asked to give up for the sake of eros. Badiou’s targets were advertising slogans, not the people who used the site. And commitment chosen with care is not the enemy of love: Kierkegaard’s own point was that duty makes love steady, which a careful choice may support.

What it means today

The distinction worth keeping is between two kinds of risk. Physical danger, fraud and deceit are risks a platform should reduce as far as it can. The risk of being changed by another person, of choosing someone without proof and staying, is the one a product cannot remove without removing the thing it sells. Products that sell certainty about the second kind sell something they cannot deliver. Hinge’s promise to be deleted is, read generously, a promise to hand people over to that second risk and step out of the way.

7. The machine that pays out sometimes

The idea in plain terms

A reward that comes unpredictably keeps people going longer than one that comes every time. Dating apps deliver matches and messages on an unpredictable schedule. Put that schedule inside a business paid by continued use and two things may follow: habits users would not choose on reflection, and a company whose incentive pulls against its stated purpose.

The tradition

C. B. Ferster and B. F. Skinner’s Schedules of Reinforcement (1957) reported years of experiments with pigeons and rats. Animals rewarded after a variable number of responses responded at a high, steady rate, and the behaviour was especially hard to extinguish.¹¹⁹ Natasha Dow Schüll’s Addiction by Design (2012), based on about fifteen years of fieldwork in Las Vegas, described how gambling machines are built to maximise “time on device” and the trance-like “machine zone” players enter, and located addiction in the interaction between person and machine rather than in the person alone.¹²⁰ Schüll wrote about gambling, not dating; applying her account to apps is an analogy. Orosz and colleagues developed a six-item Problematic Tinder Use Scale in 2016, which measures compulsive use but is not a clinical diagnosis.¹²¹ Nir Eyal’s Hooked is the product designer’s version of the same mechanism, written as advice.

Harry Frankfurt’s account of second-order desires, discussed on this site’s page Wanting What We Want to Want, adds the moral frame: a person may want to keep swiping and want not to want it.

Where it shows up

The swipe was built, in Badeen’s words, out of a desire to “gamify” a stack of cards. A match arriving after an unpredictable number of right swipes fits the variable schedule closely. The paid tiers sell more pulls (unlimited likes), better odds (Boost, Super Like) and a peek at the outcome (Likes You). The subscription paradox, that a successful couple leaves, gives the business a reason to value engagement over endings, though Match denies it acts on that reason. The 2024 lawsuit alleging addictive design went to arbitration and was never tested in public.⁷⁶ Hinge’s slogan, and the EU’s proposal of a right to switch off addictive features, are two answers from opposite directions.

The strongest objection

The analogy is loose. Slot machines take money on every pull; most dating-app users pay nothing, and many stop when they find someone, which Rosenfeld’s figures show they often do. Variable rewards are part of all social life, including meeting people at parties, and nobody calls a party addictive. The evidence on harm is mostly cross-sectional and mixed. And the market itself offers a test: if apps were successful at trapping people, Tinder’s users would not be leaving.

What it means today

The decline of Tinder is an argument against the strongest version of the charge and for a weaker one. People can tire of a variable-reward loop, and when they do, the business suffers. That suggests the incentive to maximise swiping is real but self-defeating over time, which is perhaps the most useful thing a product leader can learn from this history. A design whose engagement metric rises while its users’ outcomes fall is the problem this site discusses as When a Measure Becomes a Target. The better metric for a dating service is the one Hinge advertised: people leaving because they found someone, counted and rewarded inside the company.

8. The politics of wanting

The idea in plain terms

Whom people find attractive feels private and beyond argument. Yet desire falls into patterns by race, age, body and disability, and dating platforms record those patterns, sort by them and sometimes sell filters for them. Once a platform adds up millions of private preferences and sorts by them, preference becomes a public matter, even though nobody has a right to be desired.

The tradition

Christian Rudder’s Dataclysm (2014) published OkCupid’s data on ratings and messages by race, showing persistent penalties for black women and Asian men.³³ Bruch and Newman found a desirability hierarchy that correlated with age, race and education.⁴⁸ In 2018 Hutson, Taft, Barocas and Levy examined the design of 25 leading “intimate platforms” in “Debiasing Desire” and argued that features such as race filters, sorting and recommendations can deepen or soften bias, and that platforms can change patterns while leaving users free to choose.¹²² The philosopher Amia Srinivasan’s essay “Does anyone have the right to sex?” (2018), later expanded in The Right to Sex (2021), held that nobody is owed desire, while arguing that desire is shaped by politics and that people may have reason to try to change their own.¹²³

Where it shows up

Com-Pat’s founders designed against difference; their system let users list what they “would not tolerate”. OkCupid’s data made the resulting pattern visible. Grindr announced in June 2020 that it would remove its ethnicity filter, after criticism of a statement of solidarity with Black Lives Matter, and some users argued that the filter had helped people of colour find each other.¹²⁴ Tinder’s age pricing, held potentially unlawful in California, was discrimination by the company itself. Recommendation systems that learn from swipes can reproduce the patterns without any filter at all, a problem this site’s essay on recommendation feeds discusses in general.

The strongest objection

Desire is the last place anyone should want a company to intervene. A platform that quietly adjusts whom users see in order to change their tastes is manipulating them, however good its intentions, and Srinivasan herself rejects any right to be desired. Users who want partners of their own religion, language or background, as matrimonial sites in India have always allowed, may have good reasons. And the removal of Grindr’s filter suggests that crude interventions can harm the people they mean to help.

What it means today

Coercion is ruled out. What a platform should do by default is still open. Every design choice already shapes whom users see, so neutrality is not an option. A platform can decide not to sell race filters, not to rank users by a single desirability score that encodes bias, and to show people a slightly wider range than their past swipes predict, while letting them choose. Those are decisions of the kind described in The Architecture of Choice, and a dating company makes them whether it admits it or not.

Bringing the threads together

Two of these ideas are about what can be known. The research on prediction says that the bond between two particular people cannot yet be seen in advance, and the economics of matching says that markets for people work through thickness, congestion and safety more than through clever matching. Together they suggest that the honest product is an introduction service, not an oracle.

Three are about what the format does to the person using it. Choice overload, the variable-reward loop and the profile as performance describe a self that is shaped by long hours of judging and being judged: quicker to reject, more attentive to surfaces, harder to satisfy. None of these effects is large in any single study, and all are disputed. They add up to a reason to design for stopping.

The rest are about value. Fromm, Illouz and Sandel ask whether market habits corrupt love; Badiou, Kierkegaard and Frankfurt answer that love begins where calculation ends; Srinivasan and Hutson ask whose desires a market amplifies. They do not agree. They share a suspicion that a product optimised for searching may teach people to keep searching.

A market for strangers

Operation Match’s founders thought the blind date and the mixer were irrational. Sixty years on, the industry they started has removed most of the irrationality they had in mind. Nobody now has to wait for a friend’s introduction or a dance. What it has not removed is the part of meeting someone that no questionnaire or swipe can settle in advance, and the history suggests some working principles for anyone who builds, runs or regulates a service that introduces people.

Sell introductions, and say so. The evidence since Finkel’s review has not changed: software can widen the pool and screen out mismatches, and cannot yet foresee a spark. Claims of compatibility beyond that are the riskiest thing a dating company can say, and the most tempting.

Ration attention on purpose. The double opt-in, Bumble’s rule and Hinge’s daily match were all limits, and each was the feature users chose. A thick market without congestion control serves the most desired users and exhausts everyone else.

Make safety independent of payment. The screening that covered paying subscribers and not free users is the clearest case in this history of a decision that looked like cost control at the time and like a moral failure afterwards. Verification, reporting and bans should work the same for every user.

Count the endings. A service whose success is people leaving needs to measure that and reward the teams that produce it, or it will drift toward measuring swipes. Hinge grew while Tinder shrank, which suggests users notice the difference.

Decide where the person must be present. Match’s ban on AI agents draws a line between help and substitution. Wherever a company draws it, the question is worth asking early: which parts of meeting someone are worth doing oneself, slowly and at some risk?

Tokyo’s government app had produced 265 marriages by the middle of 2026, and Tinder now offers to read a user’s camera roll and choose one person a day. Nobody has yet measured how many of those marriages the software made, and how many it merely witnessed. Rosenfeld’s surveys say only where couples met. What they did next was never in the data.

Selected bibliography

• Gale, David, and Lloyd S. Shapley. “College Admissions and the Stability of Marriage.” American Mathematical Monthly 69, no. 1 (1962): 9–15.
• Fromm, Erich. The Art of Loving. 1956.
• Goffman, Erving. The Presentation of Self in Everyday Life. Edinburgh: University of Edinburgh Social Sciences Research Centre, 1956; expanded edition, New York: Doubleday Anchor, 1959.
• Ferster, C. B., and B. F. Skinner. Schedules of Reinforcement. 1957.
• Becker, Gary S. “A Theory of Marriage: Part I.” Journal of Political Economy 81, no. 4 (1973): 813–846.
• Bailey, Beth. From Front Porch to Back Seat. Baltimore: Johns Hopkins University Press, 1988.
• Nussbaum, Martha C. “Objectification.” Philosophy & Public Affairs 24, no. 4 (1995): 249–291.
• Iyengar, S. S., and M. R. Lepper. “When Choice Is Demotivating: Can One Desire Too Much of a Good Thing?” Journal of Personality and Social Psychology 79, no. 6 (2000): 995–1006.
• Frankfurt, Harry G. The Reasons of Love. Princeton: Princeton University Press, 2004.
• Schwartz, Barry. The Paradox of Choice: Why More Is Less. 2004.
• Ellison, N., R. Heino, and J. Gibbs. “Managing Impressions Online: Self-Presentation Processes in the Online Dating Environment.” Journal of Computer-Mediated Communication 11, no. 2 (2006): 415–441.
• Illouz, Eva. Cold Intimacies: The Making of Emotional Capitalism. Cambridge: Polity, 2007.
• Toma, C. L., J. T. Hancock, and N. B. Ellison. “Separating Fact From Fiction: An Examination of Deceptive Self-Presentation in Online Dating Profiles.” Personality and Social Psychology Bulletin 34, no. 8 (2008): 1023–1036.
• Badiou, Alain, with Nicolas Truong. Éloge de l’amour. 2009. Translated by Peter Bush as In Praise of Love. London: Serpent’s Tail, 2012.
• Hitsch, G. J., A. Hortaçsu, and D. Ariely. “Matching and Sorting in Online Dating.” American Economic Review 100, no. 1 (2010): 130–163.
• Lenton, A. P., and M. Francesconi. “How Humans Cognitively Manage an Abundance of Mate Options.” Psychological Science 21 (2010): 528–533.
• Scheibehenne, B., R. Greifeneder, and P. M. Todd. “Can There Ever Be Too Many Options? A Meta-Analytic Review of Choice Overload.” Journal of Consumer Research 37, no. 3 (2010): 409–425.
• Finkel, E. J., P. W. Eastwick, B. R. Karney, H. T. Reis, and S. Sprecher. “Online Dating: A Critical Analysis From the Perspective of Psychological Science.” Psychological Science in the Public Interest 13, no. 1 (2012): 3–66.
• Ellison, N. B., J. T. Hancock, and C. L. Toma. “Profile as Promise: A Framework for Conceptualizing Veracity in Online Dating Self-Presentations.” New Media & Society 14, no. 1 (2012): 45–62.
• Illouz, Eva. Why Love Hurts: A Sociological Explanation. Cambridge: Polity, 2012.
• Sandel, Michael J. What Money Can’t Buy: The Moral Limits of Markets. New York: Farrar, Straus and Giroux, 2012.
• Schüll, Natasha Dow. Addiction by Design: Machine Gambling in Las Vegas. Princeton: Princeton University Press, 2012.
• Han, Byung-Chul. Agonie des Eros. Berlin: Matthes & Seitz, 2012. Translated by Erik Butler as The Agony of Eros. Cambridge, MA: MIT Press, 2017.
• Cacioppo, J. T., S. Cacioppo, G. C. Gonzaga, E. L. Ogburn, and T. J. VanderWeele. “Marital Satisfaction and Break-ups Differ Across On-line and Off-line Meeting Venues.” PNAS 110, no. 25 (2013): 10135–10140.
• Rudder, Christian. Dataclysm: Who We Are (When We Think No One’s Looking). New York: Crown, 2014.
• Chernev, A., U. Böckenholt, and J. Goodman. “Choice Overload: A Conceptual Review and Meta-Analysis.” Journal of Consumer Psychology 25, no. 2 (2015): 333–358.
• Roth, Alvin E. Who Gets What, and Why: The New Economics of Matchmaking and Market Design. Boston: Houghton Mifflin Harcourt, 2015.
• Hicks, Mar. “Computer Love: Replicating Social Order Through Early Computer Dating Systems.” Ada: A Journal of Gender, New Media, and Technology, no. 10 (2016).
• Orosz, G., I. Tóth-Király, B. Bőthe, and D. Melher. “Too Many Swipes for Today: The Development of the Problematic Tinder Use Scale (PTUS).” Journal of Behavioral Addictions 5, no. 3 (2016): 518–523.
• Weigel, Moira. Labor of Love: The Invention of Dating. 2016.
• Joel, S., P. W. Eastwick, and E. J. Finkel. “Is Romantic Desire Predictable? Machine Learning Applied to Initial Romantic Attraction.” Psychological Science (2017).
• D’Angelo, J. D., and C. L. Toma. “There Are Plenty of Fish in the Sea: The Effects of Choice Overload and Reversibility on Online Daters’ Satisfaction with Selected Partners.” Media Psychology (2017).
• Bruch, E. E., and M. E. J. Newman. “Aspirational Pursuit of Mates in Online Dating Markets.” Science Advances 4, no. 8 (2018): eaap9815.
• Hutson, J. A., J. G. Taft, S. Barocas, and K. Levy. “Debiasing Desire: Addressing Bias and Discrimination on Intimate Platforms.” Proceedings of the ACM on Human-Computer Interaction 2, CSCW (2018): article 73.
• Srinivasan, Amia. “Does Anyone Have the Right to Sex?” London Review of Books 40, no. 6 (2018).
• Rosenfeld, M. J., R. J. Thomas, and S. Hausen. “Disintermediating Your Friends: How Online Dating in the United States Displaces Other Ways of Meeting.” PNAS 116, no. 36 (2019): 17753–17758.
• Illouz, Eva. The End of Love: A Sociology of Negative Relations. New York: Oxford University Press, 2019.
• Pronk, T. M., and J. J. A. Denissen. “A Rejection Mind-Set: Choice Overload in Online Dating.” Social Psychological and Personality Science 11, no. 3 (2020): 388–396.
• Joel, S., P. W. Eastwick, C. J. Allison, et al. “Machine Learning Uncovers the Most Robust Self-Report Predictors of Relationship Quality Across 43 Longitudinal Couples Studies.” PNAS 117, no. 32 (2020): 19061–19071.
• Srinivasan, Amia. The Right to Sex. London: Bloomsbury, 2021.

Research limitations

Much of the early history rests on secondary sources. The wording of the 1695 advertisement comes from later quotation, and other dates (1692, 1685) are claimed for the first personal ad; a popular story of a Manchester advertiser committed to an asylum in 1727 could not be verified and is left out. Accounts of Operation Match differ on the computer used (an IBM 1401 or a machine at Avco), on whether Vaughan Morrill or David Crump was the co-founder, and on its scale, with figures of 7,800, 70,000, 100,000 and more than a million measuring different things at different times. Joan Ball’s priority rests on Hicks’s research. Revenue for the Minitel messageries roses comes from one French secondary source; Great Expectations’ marriage figures are its founder’s claims. Match.com’s sale to Cendant is given as $7 million in 1998 or $8 million in 1997, and Kremen’s remark about designing for women is reported from a summary of a Financial Times profile. The quotations from Beth Bailey and Moira Weigel are taken from reviews. eHarmony’s launch date and questionnaire length vary between sources, and the totals of the California settlement conflict. OkCupid’s 2014 conversation rates differ between reports. Tinder’s daily swipe and match figures, Hinge’s claim about Most Compatible and Match’s descriptions of its features are company claims. Tinder’s Elo score is known from secondary reports of the Fast Company article. The date Hinge adopted “designed to be deleted” rests on a single source. The Pew and Cacioppo figures for relationships begun online are not directly comparable. Figures for Match Group’s third quarter of 2026 come from trade press, and Rascoff’s remarks on AI agents from secondary reports of a New York Times interview that could not be read. Research on dating apps and mental health is mostly cross-sectional, and the 2025 review and 2026 meta-analysis were seen only in news summaries; the 78 per cent burnout figure is a commercial poll. Surveys of romantic use of AI vary widely with wording. At the time of writing, the Denver lawsuit against Match Group, the final approval of the Candelore settlement, the arbitration of the addictive-design claims, the closing of Bumble’s sale and the passage of the EU Digital Fairness Act were pending. Fromm’s phrases are checked only against excerpts, Bauman’s views were not used because his wording could not be verified, and the publisher of Ferster and Skinner’s book is given differently by different sources. The connections to Iyengar, Lepper, Schwartz, Scheibehenne, Greifeneder, Todd, Chernev, Böckenholt, Goodman, Lenton, Francesconi, D’Angelo, Toma, Pronk, Denissen, Simon, Gale, Shapley, Roth, Becker, Hitsch, Hortaçsu, Ariely, Lee, Niederle, Bruch, Newman, Finkel, Eastwick, Joel, Goffman, Ellison, Heino, Gibbs, Hancock, Nussbaum, Fromm, Illouz, Sandel, Badiou, Han, Kierkegaard, Frankfurt, Ferster, Skinner, Schüll, Orosz, Eyal, Rudder, Hutson, Taft, Barocas, Levy and Srinivasan are the author’s interpretation; their works are paraphrased apart from short, attributed phrases. The decisions in the decisions box are judged by the author, separating what could be known at the time (process) from what happened (call).

Footnotes

1. Harvard Crimson, “Operation Match Draws 32 Cliffies,” 19 April 1965. ↩

2. Mar Hicks, “Computer Love: Replicating Social Order Through Early Computer Dating Systems,” Ada no. 10 (2016). ↩ ↩² ↩³ ↩⁴ ↩⁵

3. M. J. Rosenfeld, R. J. Thomas and S. Hausen, “Disintermediating your friends,” PNAS 116 (2019): 17753–17758. ↩ ↩² ↩³

4. Match Group, fourth quarter and full year 2025 results, 3 February 2026; Q4 2025 earnings call transcript; Match Group, 8-K exhibit 99.1 for Q4 2024. ↩ ↩² ↩³

5. Tinder, “Tinder Debuts Inaugural Product Keynote, Tinder Sparks 2026,” 12 March 2026. ↩ ↩²

6. Bitdefender summary of FTC 2025 data; FTC, “Romance scammers’ favorite lies exposed,” February 2023. ↩ ↩²

7. Literary Review on Francesca Beauman, Shapely Ankle Preferr’d (2011); Futility Closet. ↩

8. Pan Macmillan on Penrose Halson, Marriages Are Made in Bond Street; The Oldie. ↩

9. American Heritage, “Front Porch, Back Seat”, on Beth Bailey’s book (1988). ↩

10. Ploughshares review of Moira Weigel, Labor of Love (2016). ↩

11. Harvard SEAS, “Alumni profile: Jeff Tarr”; Hicks, “Computer Love”. ↩

12. Harvard Crimson, “Operation Match,” 16 February 2018. ↩

13. BBC Culture (mirror), 6 February 2025; Hicks, “Computer Love”. ↩

14. Atlas Obscura; UPI, 12 April 1981. ↩

15. Le Tribunal du Net. ↩

16. Internet Society, “Love and Relationships”. ↩

17. “Match.com L.P.,” International Directory of Company Histories. ↩

18. The Week, “Financial Times Match.com profile: 5 takeaways”. ↩

19. The Story Exchange; Albuquerque Journal. ↩

20. Chief Marketer; Ticketmaster Online-CitySearch, Form 10-Q, 1999; Wikipedia, Gary Kremen; “Match.com L.P.,” International Directory of Company Histories. ↩

21. Mary Madden and Amanda Lenhart, Online Dating, Pew Internet & American Life Project, 5 March 2006. ↩

22. Pew Research Center, “Online Dating & Relationships,” 21 October 2013. ↩ ↩²

23. Psychotherapy Networker, “Match Made on Earth”; Business Wire, January 2013; DatingNews. ↩

24. The Register, 20 November 2008; NBC News; Courthouse News; Claims Journal, 29 January 2010. ↩

25. E. J. Finkel et al., “Online Dating: A Critical Analysis From the Perspective of Psychological Science,” Psychological Science in the Public Interest 13, no. 1 (2012): 3–66; ScienceDaily, 6 February 2012. ↩ ↩² ↩³

26. J. T. Cacioppo et al., “Marital satisfaction and break-ups differ across on-line and off-line meeting venues,” PNAS 110 (2013); Nature News. ↩

27. Wikipedia, OkCupid; Mixpanel interview with Christian Rudder. ↩

28. Global News, 15 July 2015; Mathew Ingram. ↩

29. IAC, Form 10-K for 2011; TechCrunch, 2 February 2011. ↩ ↩²

30. TechCrunch, 14 July 2015. ↩

31. Amplitude, “The dating app paradox”. ↩

32. NPR Planet Money, 13 February 2024. ↩

33. TIME and NPR on Christian Rudder, Dataclysm (2014); OkCupid, “Race and Attraction, 2009–2014”; mathbabe critique. ↩ ↩²

34. Christian Rudder, “We Experiment On Human Beings!”, OkCupid blog, 28 July 2014; CNN Money, 28 July 2014; MediaPost. ↩

35. Asher Snyder, “OkCupid’s data just doesn’t match”. ↩

36. PinkNews, 22 February 2011; Frieze, 23 December 2019; Wikipedia, Grindr. ↩

37. Wikipedia, Badoo; The Next Web, 17 July 2011. ↩

38. Wikipedia, Tinder; CNBC, 1 December 2021. ↩ ↩²

39. CNBC, 6 January 2017; VentureBeat. ↩

40. TechCrunch, 8 September 2014. ↩

41. CNBC, 1 December 2021; TechCrunch. ↩

42. CNN Money, 3 March 2015; Fortune, 3 March 2015. ↩

43. TechCrunch, 1 October 2015; TechCrunch, 29 August 2017; Motley Fool. ↩

44. TechCrunch, 5 August 2020. ↩

45. Law.com/The Recorder, 30 January 2018; Top Class Actions, 4 February 2026; settlement agreement. ↩

46. Digital Trends and Refinery29, reporting Austin Carr’s Fast Company article of January 2016. ↩

47. Tinder, “Powering Tinder: The Method Behind Our Matching,” March 2019, quoted by Engadget, 18 March 2019. ↩

48. E. E. Bruch and M. E. J. Newman, “Aspirational pursuit of mates in online dating markets,” Science Advances 4, no. 8 (2018). ↩ ↩² ↩³

49. Tyson et al., “A First Look at User Activity on Tinder,” ASONAM 2016; MIT Technology Review, 15 July 2016. ↩

50. Business Insider; Wikipedia, Bumble. ↩

51. Taipei Times/Reuters, 13 February 2021; Investing.com/Reuters. ↩

52. TechCrunch, 27 September 2016; Refinery29; Digital Trends. ↩

53. Design Week. ↩

54. TechCrunch, 11 July 2018; Engadget, 12 July 2018. ↩

55. CNN Money, 20 June 2018; Global Dating Insights, 7 February 2019. ↩

56. Match Group, 8-K exhibit 99.1 for Q4 2022; Form 10-K for 2023; 8-K exhibit 99.1 for Q4 2024. ↩

57. CNN Money, 16 October 2015; 24/7 Wall St. ↩

58. IAC and Match Group complete full separation, 1 July 2020. ↩

59. Match Group, 8-K exhibit 99.1, 2 February 2021; 8-K exhibit 99.1, 31 January 2023; Form 10-K for 2023. ↩

60. TechCrunch, 9 May 2022; Engadget. ↩

61. Engadget, 5 September 2019. ↩

62. Momo, acquisition of Tantan, 23 February 2018; Business Today, 8 March 2022. ↩

63. Wharton Knowledge, “Matrimonial portals”; The News Minute. ↩

64. Yahoo, Tinder’s 2020 year in review. ↩

65. FTC, “Match Group Agrees to Pay $14 Million,” 12 August 2025; VitalLaw on the 2022 ruling; ConsumerAffairs; Fortune, 13 August 2025. ↩

66. FTC, “Operators of AshleyMadison.com settle FTC, state charges,” December 2016. ↩

67. UNODC, Inflection Point, April 2025. ↩

68. ProPublica, December 2019; The Next Web. ↩

69. The Markup, 16 December 2025; Pulitzer Center methodology; Colorado Public Radio, 16 December 2025. ↩

70. Eric Goldman, Technology & Marketing Law Blog; Techdirt, 29 March 2019. ↩

71. Axios, 2 April 2018; Digital Policy Alert on the Borgarting Court of Appeal ruling; Attitude, 7 September 2026. ↩

72. Axios, 21 July 2021. ↩

73. Good Morning America; Match Group, Face Check release, 22 October 2025; 9to5Mac, 23 October 2025. ↩

74. Holtzhausen et al., BMC Psychology 8 (2020); The Conversation, via Interaksyon, 11 February 2025; Bonilla-Zorita, Griffiths and Kuss (2021). ↩

75. Global Dating Insights on the Forbes Health/OnePoll survey. ↩

76. Match Group, Form 10-K for 2024; Insurance Journal, 15 February 2024. ↩ ↩²

77. Ofcom, “Age checks helping make online experiences safer for UK children”. ↩

78. Minister for Communications, “Online dating platforms now subject to enforcement”. ↩

79. Agence Europe; Quartz. ↩

80. Ofcom, Online Nation 2025, 10 December 2025. ↩

81. Pew Research Center, “From Looking for Love to Swiping the Field,” 2 February 2023; key findings. ↩ ↩²

82. Bloomberg, 8 May 2025. ↩

83. CNBC, 9 December 2025; TechCrunch. ↩

84. Match Group, third quarter 2026 results; Global Dating Insights, 3 November 2026; Match Group, 8-K exhibit 99.1 for Q2 2026. ↩

85. Quartz; TechCrunch, 5 May 2026; Bumble, full year 2025 results; Bumble 10-Q for Q3 2026; Bumble 8-K exhibit 99.1, 13 October 2026. ↩

86. Hello Group, fourth quarter and fiscal year 2025 results. ↩

87. The Star, 28 July 2026; Global Dating Insights. ↩

88. The Next Web, on “chatfishing”. ↩ ↩² ↩³

89. Wheatley Institute, via PR Newswire, February 2025; Newsweek. ↩

90. TechCrunch, 25 June 2025; AlleyWatch, December 2025. ↩

91. Mashable; Dexerto, 11 November 2026. ↩

92. Grindr special committee statement, Business Wire, November 2025. ↩

93. S. S. Iyengar and M. R. Lepper, “When Choice Is Demotivating,” Journal of Personality and Social Psychology 79, no. 6 (2000): 995–1006, as summarised in Psychology Today and CS Monitor. ↩

94. B. Scheibehenne, R. Greifeneder and P. M. Todd, “Can There Ever Be Too Many Options?” Journal of Consumer Research 37, no. 3 (2010): 409–425. ↩

95. A. Chernev, U. Böckenholt and J. Goodman, “Choice overload: A conceptual review and meta-analysis,” Journal of Consumer Psychology 25, no. 2 (2015): 333–358. ↩

96. A. P. Lenton and M. Francesconi, “How Humans Cognitively Manage an Abundance of Mate Options,” Psychological Science 21 (2010): 528–533. ↩

97. University of Wisconsin–Madison, 13 June 2017, on D’Angelo and Toma, Media Psychology (2017). ↩

98. T. M. Pronk and J. J. A. Denissen, “A Rejection Mind-Set: Choice Overload in Online Dating,” Social Psychological and Personality Science 11, no. 3 (2020): 388–396. ↩

99. David Gale and Lloyd S. Shapley, “College Admissions and the Stability of Marriage,” American Mathematical Monthly 69, no. 1 (1962): 9–15. ↩

100. NobelPrize.org, press release, 2012. ↩

101. Gary S. Becker, “A Theory of Marriage: Part I,” Journal of Political Economy 81, no. 4 (1973): 813–846. ↩

102. Cato Institute review and Minneapolis Fed interview on Alvin E. Roth, Who Gets What, and Why (2015). ↩

103. G. J. Hitsch, A. Hortaçsu and D. Ariely, “Matching and Sorting in Online Dating,” American Economic Review 100, no. 1 (2010): 130–163. ↩

104. S. Lee and M. Niederle, working paper on signalling with virtual roses. ↩

105. Association for Psychological Science, “Romantic matches are hard to predict”, on S. Joel, P. W. Eastwick and E. J. Finkel, Psychological Science (2017). ↩

106. S. Joel et al., “Machine learning uncovers the most robust self-report predictors of relationship quality,” PNAS 117, no. 32 (2020): 19061–19071. ↩

107. Encyclopedia.com, “Goffman, Erving”. ↩

108. N. Ellison, R. Heino and J. Gibbs, “Managing Impressions Online,” Journal of Computer-Mediated Communication 11, no. 2 (2006): 415–441. ↩

109. C. L. Toma, J. T. Hancock and N. B. Ellison, “Separating Fact From Fiction,” Personality and Social Psychology Bulletin 34, no. 8 (2008): 1023–1036. ↩

110. N. B. Ellison, J. T. Hancock and C. L. Toma, “Profile as promise,” New Media & Society 14, no. 1 (2012): 45–62. ↩

111. Stanford Encyclopedia of Philosophy, “Feminist Perspectives on Objectification”, on Martha C. Nussbaum, “Objectification,” Philosophy & Public Affairs 24, no. 4 (1995): 249–291. ↩

112. The Collector and Rainer Funk (2017) on Erich Fromm, The Art of Loving (1956). ↩

113. Polity, Cold Intimacies; Cambridge review of Why Love Hurts; Hebrew University, The End of Love. ↩

114. The Humanist and PhilArchive review of Michael J. Sandel, What Money Can’t Buy (2012). ↩

115. The Montreal Review; The New Press, on Alain Badiou, In Praise of Love. ↩

116. Matthes & Seitz, on Byung-Chul Han, Agonie des Eros (2012). ↩

117. Jamie Ferreira, “Two Forms of Love: The Problem of Preferential Love in Kierkegaard’s Works of Love,” Journal of Religious Ethics 36, no. 4 (2008); Søren Kierkegaard, Works of Love, Hong translation (excerpt). ↩

118. Notre Dame Philosophical Reviews, on Harry G. Frankfurt, The Reasons of Love (2004). ↩

119. C. B. Ferster and B. F. Skinner, Schedules of Reinforcement (1957), B. F. Skinner Foundation reprint. ↩

120. Project MUSE, Natasha Dow Schüll, Addiction by Design (2012). ↩

121. G. Orosz et al., “Too many swipes for today,” Journal of Behavioral Addictions 5, no. 3 (2016): 518–523. ↩

122. J. A. Hutson, J. G. Taft, S. Barocas and K. Levy, “Debiasing Desire,” Proceedings of the ACM on Human-Computer Interaction 2, CSCW (2018). ↩

123. London Review of Books, Amia Srinivasan; The Nation. ↩

124. Engadget; PinkNews, 2 June 2020. ↩