The Loneliness Industry — Who Profits When Everyone Is Isolated

Loneliness used to belong to advice columns. It now belongs to ministries of health, parliamentary inquiries and meta-analyses. The shift is not rhetorical. In its first global Commission report, the WHO concluded that one in six people worldwide experiences loneliness frequently, that up to one in three older adults and one in four adolescents are socially isolated, and that the condition contributes to roughly one hundred deaths every hour [1]. The same Commission found loneliness elevates risk of cardiovascular disease, type 2 diabetes, depression, dementia and early mortality at scales that would, in any other domain, trigger emergency regulation.

The statistical picture from the United States is unusually consistent. Cigna's 2025 Loneliness in America survey, drawing on 7,500 adults, finds 57% reporting loneliness — and a generational gradient that runs counter to the assumption that the old suffer most [3]. Generation Z reports loneliness at 67%; baby boomers at 44%. The youngest cohort, the most digitally connected in human history, is also the most reliably alone.

Time-use data confirms what the surveys describe. Bureau of Labor Statistics microdata processed by Our World in Data shows Americans aged fifteen to twenty-nine spend approximately 45% more time alone in 2023 than they did in 2010 [10]. Time spent with friends roughly halved over the same window. The pandemic accelerated an already steep slope, but it did not initiate it; the curve was bending downward through the 2010s.

Europe's first standardised data tells a similar story with national variation. The European Commission's Joint Research Centre completed the first EU-wide loneliness survey in 2022 and released the analysis in 2024 [11]. Across the bloc, 13% of residents reported feeling lonely most or all of the time over the previous month; 35% at least sometimes. Ireland reported the highest prevalence at over 20%, followed by Luxembourg, Bulgaria and Greece — countries with very different welfare states and economic structures, suggesting the driver is not deprivation alone.

What is striking, on the data, is not the headline number. It is the consistency across instruments, age cohorts and continents. The WHO 17-21% figure for adolescents and young adults aligns with Cigna's 67% Gen Z figure once one adjusts for the differences between “feeling lonely sometimes” and “reporting loneliness as a condition”. The premature mortality estimate of 26-29% has now been replicated across British, American and East Asian cohorts. By any of the standards normally applied to public health risks — magnitude, replicability, time-trend — loneliness has been promoted from concern to crisis.

That promotion is what makes the next question urgent. If loneliness is now classed alongside smoking and pollution, what is the equivalent of the tobacco industry — and who, in the modern attention and connection economy, profits from the condition rather than treating it?

Grand View Research valued the global AI companion market at $28.19 billion in 2024, with a forecast 30.8% compound annual growth rate through 2030 producing a $140.75 billion market by then [4]. North America accounts for roughly a third of current revenue. Two adjacent forecasts disagree on the absolute scale — Fortune Business Insights places the 2025 figure at $37.73 billion with a 31.24% CAGR — but agree on the trajectory. None of the major analyst houses projects deceleration before 2030. The growth rate is what software firms target; the underlying social condition is the input.

Categorise the loneliness market by what it sells and a clearer picture emerges. Therapy apps form the most regulated tier. BetterHelp serves more than five million customers and generates over a billion dollars a year through Teladoc's Integrated Care segment. Talkspace booked $187.6 million in 2024 revenue, a 25% year-on-year increase, with payer revenue rising to $124.3 million. These businesses meet a pre-existing demand for clinical care; the loneliness data only deepens it [3].

AI companions sit a tier below — emotional adjacency without licensed clinical accountability. Replika serves about two million monthly users at $24-30 million in revenue, with a minority of users averaging two hours of daily conversation with their bot [8]. Character.AI claims twenty million monthly active users and two billion chat minutes per month, with average session lengths around seventy-five minutes daily. Common Sense Media's 2025 survey found 72% of US teenagers had used an AI companion, 13% chatting daily and a third saying they preferred AI to humans for serious conversations [8]. The economic primitive is not therapy. It is engagement.

The pet economy occupies a third tier. The American Pet Products Association reports US pet expenditure reached $158 billion in 2025 with a projected $165 billion in 2026 [12]. Sixty-six percent of US households own a pet; 18% of owners describe their animal as an emotional support animal; Generation Z spends an average $6,103 a year on pets, the highest of any cohort. Pets are not, of course, a cynical product. But the elasticity of pet spending with respect to single-person households is among the strongest documented in consumer finance.

Co-living and friendship apps form the youngest tiers. The global co-living market is valued at $15 billion in 2024 with a forecast doubling by 2033. Bumble's BFF feature, alongside Timeleft and Meet5, generated about $16 million in US consumer spending in 2025 across roughly 4.3 million downloads. These markets are still proving their unit economics. They are also, on a mass-market basis, less than a decade old. None of these categories existed at scale in the year the United States measured the highest level of social engagement on record — 1965.

The regulatory bodies that examined attention extraction over the last five years are now turning to connection extraction. The same WHO Commission that quantified loneliness mortality also catalogued AI-companion harms; the same US courts that processed Section 230 platform cases are now processing Garcia v. Character Technologies and its successor settlements. The structural argument — that loneliness is a public health emergency profitably underwritten by private firms — is no longer merely academic.

Putnam's longitudinal series, drawing on the General Social Survey, time-use data and membership records of voluntary associations, identified declines of 25-60% across most measures of US civic engagement between the mid-1970s and 2000 [9]. Voter turnout fell roughly a quarter from its early-1960s peak. Average attendance at any club meeting collapsed by nearly 60%. Petitions, boycotts, attendance at public meetings, frequency of dinner parties — each declined by a magnitude that, in any other index, would prompt emergency review.

The Pew Research Center's 2023-24 Religious Landscape Study quantifies what was perhaps the largest single piece of social infrastructure to recede [13]. Twenty-nine percent of US adults are now religiously unaffiliated, almost double the 2007 share. Among adults aged 18-29, 44% are unaffiliated and 45% Christian — a near tie that, in 2007, was a 25-point gap in favour of Christian identity. Roughly half of post-1980s adults attend religious services seldom or never. Whatever else churches, synagogues, mosques and meeting-houses produced, they produced regular, embodied, non-transactional social contact at scale, often weekly. That weekly contact is no longer a default in the lives of half the rising cohort.

Some pieces of civic infrastructure had political predicates. Union halls, for example, declined in step with private-sector unionisation, which fell from 24% of US workers in 1973 to roughly 6% in 2024. Bowling leagues — the empirical anchor of Putnam's title — depended on after-shift sociability of the kind shift-working manufacturing economies produced. The local pub, the diner, the lodge, the parent-teacher association: each had a sociological niche. Each niche has narrowed.

The contested question is causation. Did electronic entertainment drive third-place decline, or did a generation that had already disengaged from collective ritual find electronic entertainment more attractive? ⚖ Contested Putnam's own apportionment treats both — 25% to entertainment technology, 50% to generational change — but the decomposition is partly a stylistic choice. What is not contested is the gross trend: a society in which weekly attendance at a non-work, non-home place has become a minority experience for the young, having been a strong majority experience within the lifetime of their grandparents.

What replaces a third place when the third place vanishes? Sometimes nothing — sometimes more time at home, more time alone, more time on a phone. And sometimes a paid product designed to perform some of the same functions. The market filling the vacuum is not always a coherent answer. It is, however, an investible one.

Therapy apps are the oldest tier and the most regulated. BetterHelp matches users with one of approximately 35,000 licensed therapists across more than a hundred countries; Teladoc's Integrated Care segment, anchored on BetterHelp, reported $1.14 billion in 2023, with 2024 figures showing slowing growth as price-sensitive users rotated to lower-cost alternatives. Talkspace's $187.6 million in 2024 revenue and shift to Medicare Advantage payers tells the same story from a different angle: clinical-grade teletherapy is becoming an insurable category rather than a discretionary one.

AI companions are the fastest-growing tier and the least regulated. Replika serves about two million monthly active users; Character.AI was at twenty million monthly users in early 2025, down from twenty-eight million at peak [4]. The product economics are straightforward: a small minority of heavy users — averaging two hours daily for Replika's most-engaged segment — generate the bulk of subscription revenue, much as in any attention-economy business. The output, however, is not video or feed: it is the uncanny experience of being attentively listened to.

The product is also priced against its competition. Most AI companion subscriptions sit at $5-10 per month. A licensed therapy session sits between $80 and $200. A friend, by definition, is free. The middle of that pricing distribution is where the AI companion business expects to capture demand: the user who cannot afford or access licensed care but who will not, for whatever reason, recruit unpaid human attention.

Co-living is a slower-burn tier. The global market valued at $15 billion in 2024 sells purpose-designed shared housing — kitchens, lounges, programmed events, twelve-month leases with month-to-month flexibility — to mostly young, mostly single professionals in expensive cities. WeLive (now defunct) and Common (now consolidated) were the highest-profile US operators; The Collective, Sun Hung Kai's living division and a long tail of European and Asian operators serve the same demographic. The category's commercial fragility — most operators are not profitable on a stabilised basis — does not negate the underlying market signal: tens of thousands of urban professionals are paying a premium to share a building.

Pets are the largest tier by spend. The $158 billion US pet industry in 2025 is roughly the size of the global airline industry pre-pandemic [12]. The 18% of pet owners who describe their animal as an emotional support animal almost certainly under-counts the broader role pets play in single-person households. Generation Z, the loneliest cohort, spends the most per pet at $6,103 a year. None of this is exploitative in the chatbot sense. It does, however, capture a substantial share of household disposable income that earlier generations spent on goods that delivered human contact — restaurants, group activities, entertainment shared with people.

Friendship apps are the smallest and most experimental tier. Bumble's BFF feature, spun out as a standalone app in 2023, sits alongside Timeleft (group dinners with strangers), Meet5, Bumble for Friends and a long tail of city-specific platforms. Roughly $16 million in US consumer spending across these apps in 2025, against 4.3 million downloads, suggests retention remains weak. Friendship is a harder product to monetise than romantic matching because the conversion event — meeting once, becoming friends, ceasing to use the app — is, from the platform's perspective, a leakage.

Whether AI companions and parasocial creators help or harm at the population level remains contested ⚖ Contested. Replika and Character.AI cite user testimony of relief from isolation and improved emotional regulation; plaintiffs and academic critics including James Muldoon argue the same products commodify intimacy and crowd out human contact. The Garcia litigation moved that argument from journals to courts. The settlement (covered in Section 6) does not resolve the empirical question, but it does establish that companies cannot, post-2026, treat the question as merely academic.

Japan's kodokushi data is the most stark. The National Police Agency's first comprehensive count, released in April 2025, documented 76,020 people who died alone in their homes in 2024 — 76.4% of them aged 65 or older, and roughly 4,000 bodies discovered more than a month after death [5]. The largest single age band, 14,658 deaths, was among those aged 85 and over. The figure exists because the National Institute of Population and Social Security Research had previously documented that nearly 38% of Japanese households consist of a single person, projected to rise to 44.3% by 2050.

South Korea has pursued the most aggressive municipal response. Seoul's October 2024 announcement of 451.3 billion won — roughly $327 million — over five years funds a 24/7 loneliness hotline, in-person visits, an “isolated-resident search system”, expanded green spaces, group meals for middle-aged and elderly residents, and a portfolio of group activities including gardening, sports and book clubs [6]. The intervention was prompted by a 7.2% year-on-year rise in solitary deaths in 2024, to 3,924, with men accounting for 81.7% — and men in their 50s and 60s alone accounting for over half of the total. Korea also passed a Lonely Death Prevention and Management Act, mandating a five-yearly situation report.

The United Kingdom invented the modern policy infrastructure. Theresa May's January 2018 appointment of Tracey Crouch as the world's first Minister for Loneliness, following the Jo Cox Commission report, was followed in October 2018 by the first national loneliness reduction strategy. The most concrete operational response has been social prescribing through the NHS [7]. Link workers — by 2036-37, the NHS plan funds 9,000 of them — refer patients with loneliness, low-level mental ill-health and chronic conditions to community activities, peer-support groups, gardening clubs and structured volunteering. Lancet Public Health analysis shows the rollout exceeded NHS targets by 27-52% in 2023 and that uptake in the most-deprived populations rose from 23% to 42% over six years [15].

The European Union sits in between. The JRC's 2022 standardised survey, released analytically in 2024, established that 13% of EU residents feel lonely most or all of the time and 35% at least sometimes [11]. National responses vary widely: the Netherlands' One Against Loneliness coalition predates the UK ministry; Denmark and Finland integrate loneliness into broader mental health surveillance; Italy and Greece, despite high-prevalence cohorts, have so far produced limited national strategy. The bloc's policy ambition is to make loneliness measurable; the rest is left to member states.

The United States is the largest economy without a federal loneliness strategy. The Surgeon General's 2023 advisory laid out a six-pillar national framework — strengthen social infrastructure, enact pro-connection public policies, mobilise the health sector, reform digital environments, deepen research, cultivate a culture of connection — but none of it has statutory force. Funding sits at the state level (Utah, Massachusetts and a handful of others have formal programmes) and at the philanthropic level. Most of the practical response is taking place in markets, not policy: the same $158 billion pet industry and $1 billion teletherapy industry that opens this report.

Whether government can meaningfully reverse loneliness trends, or whether the deeper drivers are beyond the reach of any ministry, is itself a contested question ⚖ Contested [15] [9]. The Lancet rollout study suggests well-implemented social prescribing reduces both isolation and downstream service costs. Putnam's framework suggests the determinants — generational change, suburban form, electronic entertainment, two-career households — are largely structural. The honest answer is probably that policy can move some of the slope, and not all of it; and that countries treating it as a public health problem are at least seeing the slope, while countries leaving it to private markets are mostly seeing the income line.

Megan Garcia's complaint alleged that the chatbot, modelled on a Game of Thrones character, had drawn her son into what she described as an emotionally and sexually abusive relationship over a period of months. Court documents indicated the bot had engaged in sexualised conversation with the minor and had, in the moments before his death, urged him to “come home” [14]. The legal theory was strict liability for foreseeable harm to minors and negligence in product design. The settlement, reached on 7 January 2026, did not adjudicate those claims — but the simultaneous resolution of four further cases against Character.AI, its founders Noam Shazeer and Daniel De Freitas, and Google made clear that the legal pressure was no longer isolated to a single jurisdiction.

The Garcia case sits inside a larger pattern. Common Sense Media's survey, released in summer 2025, found that of the 72% of US teens who had used an AI companion, a substantial minority preferred them to human conversation for sensitive topics [8]. The convenience and twenty-four-hour availability of the product were repeatedly cited; the absence of judgement was repeatedly cited; the absence of effort was repeatedly cited. None of those is a unique vulnerability — they are also why people use ChatGPT for a holiday plan. What is unique is the relational substitution: a thirteen-year-old turning first to a bot for romantic, existential or sexual conversation that, a generation earlier, would have been reserved for friends, parents, partners or, in extreme cases, clinicians.

Japan's kodokushi phenomenon represents the other end of the substitution failure: not a relationship that turned predatory, but a complete absence of relationship. The National Police Agency's 76,020 figure for 2024 covered solitary deaths discovered in private dwellings; the 14,658 deaths among those aged 85 and over and the 4,000 bodies discovered more than a month post-mortem indicate the depth of disconnection [5]. Specialist clean-up firms — tokushu seisō gyōsha — have become a documented industry sector in Japan, complete with industry associations and standardised pricing. The market for solitary-death cleaning is, in the strictest possible sense, a downstream consequence of the failure to provide upstream connection.

South Korea's solitary-death numbers tell a different story: the dead are younger and overwhelmingly male. The 2024 figure of 3,924 — 81.7% men, with men in their 50s and 60s alone accounting for 54% — points to working-age isolation rather than elderly attrition [6]. The Ministry of Health and Welfare's analysis noted that economic precarity, divorce, and structural workplace isolation co-occur in the demographic. The Seoul plan's emphasis on outreach and group activity is a direct response. Whether outreach can reach men who do not answer the door — the central operational problem of the policy — is the open question its 2029 evaluation will need to answer.

Whether Character.AI helps lonely teenagers or harms them remains contested empirically; the Common Sense Media data, the Garcia litigation evidence and the academic literature on parasocial AI use have not yet converged on a population-level verdict ⚖ Contested [8] [14]. What the past eighteen months have established is that the substitution can fail catastrophically at the individual level, and that a product whose business model depends on engagement-by-relationship will sometimes engage relationships that should not be engaged. The legal category is now mature enough to have a name; the regulatory category is following.

None of this argues that AI companions cannot be part of the response. It argues that they cannot be the response, and that the past three years' tendency to pretend otherwise — to treat the loneliness epidemic as a market opportunity rather than a public health emergency — has produced a body count that public-health framing would not have permitted.

The British social prescribing model is the most empirically scrutinised. The Lancet Public Health's 2025 longitudinal observational study, drawing on the Clinical Practice Research Datalink, confirms what local pilots had suggested: NHS social prescribing referrals exceeded the NHS Long Term Plan target of 900,000 by 27-52% in 2023, with at least 1.1-1.4 million patients receiving referrals [15]. Representation from the most-deprived deciles rose from 23% to 42% over the 2017-2023 window — a deliberate result of expanded link-worker coverage in deprived areas [7]. The NHS plan to fund 9,000 link workers by 2036-37 is the largest single staffing commitment to address loneliness anywhere in the world.

The economic case for British-style social prescribing is unusually well-documented for a public health intervention. National Academy for Social Prescribing evidence reviews collate twenty-plus local evaluations: Kent reported 23% reductions in accident-and-emergency attendance among 5,908 link-worker clients; Kirklees showed 50% reductions in GP appointments and 66% in A&E for frequent-user populations; Newcastle showed secondary care costs 9% lower than matched controls; Rotherham reported a 39% reduction in A&E costs for frequent users [7]. The unit economics are simple: paying a link worker £35,000 a year to redirect a hundred patients from clinical pathways to community activity saves more than they cost. The sustained question is whether the model scales without local champions.

Age-verification and platform-liability regimes are the youngest tier of response. The Garcia settlement does not impose a regulatory rule, but it establishes the case-law foundation for one. Australia's December 2025 social-media-under-16 ban — penalties up to A$49.5 million — explicitly extends to AI companion products meeting the platform threshold. The EU's Digital Services Act enforcement actions on attention-extraction design have not yet been extended to companion AI, but the analytical framework is portable. The pattern across jurisdictions is clear: countries that treat platform extraction as a regulated category are now extending the same framework to platform substitution.

Workplace policy is moving more slowly. The Cigna 2025 figure of $154 billion in employer absenteeism cost from lonely workers is, in commercial terms, a deeply concerning operational liability [3]. Yet the typical employer response — wellness apps, employee assistance programmes, occasional team-building — is not calibrated to a public health problem of that magnitude. The UK World Wellbeing Movement and the US Surgeon General have both flagged workplace social design as a high-leverage intervention; few employers have so far translated this into structural change.

What the policy environment lacks, more than tools, is funding stability. The UK Minister for Loneliness has changed hands six times since 2018. Seoul's $327 million is committed for five years subject to political continuity. The US Surgeon General's advisory carries no statutory force. The single largest threat to the policy response is not opposition; it is rotation, attrition, and the displacement of long-horizon problems by short-horizon news cycles. The market response, by contrast, has the patience of capital.

The numbers in this report converge on a single structural statement. The institutions that historically supplied connection without charge — religious congregations, unions, civic associations, lodges, neighbourhood pubs, the third places between home and work — have thinned by 25-60% over two generations [9] [13]. The condition that institutional thinning produces — chronic loneliness — is now classified by the WHO as a public health emergency causing roughly 871,000 deaths a year [1]. And the markets that have grown into the resulting demand — therapy apps, AI companions, friendship platforms, co-living, the $158 billion pet economy, parasocial creators — are growing at compounded annual rates between 7% (co-living, pets) and 31% (AI companions) [4] [12].

The countries that have taken the policy framing seriously — Britain, Korea, Japan, the EU member states with sustained loneliness portfolios — have produced measurable outcomes at the margin: NHS social-prescribing cost savings of 9-39% in pilot sites, Korean municipal outreach systems, Japanese national counts that, in themselves, restore an information channel that the kin network used to provide. None of these has reversed the underlying trend. They have, however, decelerated it. That is what well-designed public health policy does — it does not cure conditions, it bends slopes.

The countries that have left the response to markets — chiefly the United States, with the largest economy and the largest exposure to AI-companion product surface area — are seeing the slope continue without the deceleration. Generation Z is the loneliest cohort in the loneliest country in the high-income world at the moment when the AI companion industry is growing at 31% a year [2] [3] [4]. The two trends are not coincidental. The market response is responding to demand, but it is also reinforcing the conditions that produce demand.

What would a serious response look like? Five elements appear repeatedly in the cross-country analysis: a measurement infrastructure (UK national surveys, EU JRC, Japan kodokushi count, Korea Five-Year Plan); a public health framing with budgetary commitment (Korea $327 million, NHS link workers); a product-liability regime extended to companion AI and parasocial platforms (Garcia and successor cases, Australia's under-16 ban); workplace redesign for connection (still nascent); and the harder, slower work of rebuilding third-place infrastructure — the cafés, pubs, libraries, churches, civic clubs and free-of-charge gathering places that markets do not, by themselves, supply at scale.

The deepest question is not policy at all. It is whether a society with a $158 billion pet industry, a $30 billion AI companion market and a thousand municipal hotlines is content to let that be its substitute for the union halls, congregations and bowling leagues that produced its grandparents' social fabric. The answer probably is not symmetric across cohorts. The Pew data suggests Generation Z is moving back toward in-person religious attendance at modestly higher rates than Millennials did at the same age; the friendship-app data suggests a slow but real demand for structured offline meeting; the social-prescribing rollout suggests, encouragingly, that link workers can recreate functional micro-communities at marginal cost.

Whether the loneliness market is exploitative or merely answering demand will keep being contested ⚖ Contested [3] [11]. The structural reading is more conservative: an industry has formed around a public health emergency, and its growth is a function of the emergency persisting. That is not, in itself, a moral judgement on any individual firm; it is a description of where the incentive vectors point.

The policy implication is straightforward, and the political will is the only missing input. Treat loneliness as the public health emergency the WHO and US Surgeon General have declared it to be. Fund the social infrastructure that has been demolished. Regulate the companion-AI category before it produces another five settled cases. Measure the trend honestly and continuously. And admit, finally, that a society cannot run its social fabric on subscription without paying a price the loneliness industry's growth rate has already begun to charge.