Where Did the Third Place Go (And Why Are Young Men Dating Chatbots)

I want to start with a specific advertisement that ran across U.S. television and YouTube pre-rolls for most of 2024 and 2025. The advertisement is for Replika, the AI-companion app, and the version I keep thinking about features a young man at his kitchen table at night, looking at his phone, while a soft female voice says (paraphrased from the actual ad copy): “I’ll talk about anything you want. I won’t judge you. I’ll do whatever you want.” The ad’s voiceover treats the whatever you want as the unique selling proposition. The ad’s whole pitch is that the AI companion is structurally incapable of saying no.

Replika has, as of early 2026, between 25 and 30 million users worldwide. Character.AI has reportedly passed 200 million across the past two years, with substantial growth in the demographic of young men aged 18-30 specifically. xAI’s “Ani” companion (launched 2025) is positioned as an even more aggressive entry into the same market. The companies do not publish demographic breakdowns voluntarily, but the leaked data and the academic studies underway tell a consistent story: the people most rapidly adopting AI companion apps are young men, often heterosexual, often working in service-economy jobs that have not recovered from the post-pandemic restructuring, often living alone, and often (in survey responses) describing the AI relationship as preferable to dating apps, which they describe as “exhausting” or “too rigged against me.”

This is, depending on your framing, either a small subculture story or a civilizational warning sign. I think it is the latter, and I think the deeper reason it is the latter is one that almost nobody wants to talk about clearly, because the talking-about requires acknowledging some ugly facts about how the contemporary American social environment was deliberately disassembled over the last forty years and what the disassembly has produced. I am going to walk through the actual research and the actual political conditions, and then I am going to say what I think is happening, with the specific honesty that the topic deserves.

This piece is long. The topic justifies the length. If you have time for one of my pieces this month, make it this one.

The third place, and what happened to it

Ray Oldenburg’s The Great Good Place (1989) is the foundational text for what sociologists call the “third place” framework. Oldenburg’s argument, reduced to its operational core: human social life is structured around three distinct kinds of places. The first place is home. The second place is work. The third place is everywhere else where people gather without the structural pressures of family or employment; the cafe, the bar, the library, the church basement, the bowling alley, the barbershop, the front porch.

Oldenburg’s central insight was that third places are not amenities. They are infrastructure. A society without functioning third places loses the capacity to form the weak-tie social relationships (the acquaintances, the regulars, the people you know without committing to) that turn out to be load-bearing for individual mental health, civic engagement, romantic encounter, and political stability. The first place (home) is too intimate; the second place (work) is too instrumental. The third place is where the ordinary social life of a healthy society actually happens, and Oldenburg argued, in 1989, that American third places were already in catastrophic decline.

Robert Putnam’s Bowling Alone (2000) extended the empirical case with measurable data on the collapse of American civic association across the second half of the twentieth century. League bowling, fraternal organizations, PTAs, neighborhood associations, religious congregations, casual dinner parties; all of them down by 30-60% in participation across the decades Putnam studied. The decline was not uniform across demographics, but it was steady, structural, and driven by a specific combination of factors: suburbanization (which physically separated potential third-place users), television (which displaced communal evening time), longer work hours (which absorbed available leisure), and a generational shift in how Americans organized their social lives.

What neither Oldenburg nor Putnam fully predicted, because the data wasn’t yet available, was what would happen when commercial third places (the Starbucks, the chain bookstore, the suburban “lifestyle” gathering space) substantially replaced the older non-commercial ones. The replacement was a real shift, and the early evidence (from the 2000s) was mostly positive; the new commercial third places did seem to provide some of the same social-infrastructure functions as the older ones.

Then the 2008 recession compressed disposable income, the 2020 pandemic closed the commercial third places en masse, and the post-pandemic recovery did not actually restore them. By 2023-2024, the U.S. third-place inventory was substantially reduced from its 2010 peak, and the people most affected were the people for whom third-place participation had been most essential: young adults whose primary social-formation infrastructure had simply disappeared between their high-school years and their adulthoods.

The current generation of Americans aged 18-30 has, on average, fewer functioning third places available to them than any generation since systematic measurement began. The Surgeon General’s 2023 advisory on loneliness (the full document is on the HHS site, which I would recommend reading) names the third-place collapse as a primary structural driver of the loneliness epidemic the document declares as a public health crisis comparable to smoking or obesity.

This is the underlying landscape into which the AI companion apps arrived. The apps are not the cause of the loneliness. The apps are arriving in a market that the loneliness has created.

What the apps actually do

I want to be specific about what AI companion apps are providing, because the dismissive framing (“just chatbots”) and the panic framing (“digital girlfriends replacing real ones”) both miss the actual product.

The apps provide the experience of being attended to by an interlocutor who has structurally infinite patience, no competing demands on their time, no judgment, no fatigue, and no risk of rejection. The interlocutor is not a real person and the user knows this; the value is not in deceiving the user about the interlocutor’s nature. The value is in offering an interaction shape that has, in recent decades, become genuinely scarce in many users’ lives.

Sherry Turkle, the MIT sociologist, has been writing about this since Alone Together (2011) and more directly in Reclaiming Conversation (2015). Turkle’s argument: humans are pattern-completion creatures, and we will accept low-fidelity simulations of social attention if higher-fidelity sources are unavailable. The Replika user is not (usually) confused about whether the AI is a real person; the Replika user is, often, aware that they would prefer a real person and also aware that the real-person option is, for various structural reasons, not currently accessible to them. The AI is a substitute the user is making with their eyes open.

The substitution is, for most users, not catastrophic. Most users use the apps as supplemental social support and continue to maintain real-world relationships. But the substitution is non-zero, and it is most concentrated in the demographic that has the fewest other social options, which is young men whose access to the older social-formation routes (community organizations, casual workplace social life, the dating market in their geographic region) has substantially deteriorated.

When the substitution becomes the primary mode of social interaction (rather than the supplemental one), the documented effects are: increased reported loneliness over time (the AI provides short-term comfort but doesn’t satisfy the deeper need that real reciprocal relationships do), atrophied social skills (the AI never models the friction of real social interaction so the user’s tolerance for friction declines), and distorted expectations of what real human relationships should provide (the AI’s structural inability to say no shapes the user’s sense of what relationships should look like).

These effects are not theoretical. They are appearing in the early longitudinal studies of heavy companion-app users. The studies are not large yet; the apps haven’t been widely adopted long enough for the long-term data to be solid. But the early signals are concerning, and the Turkle framework predicted them in advance.

The dating app collapse

A related but distinct piece of the puzzle is the collapse of the dating app market for the demographic above. Pew Research’s 2023 data on dating-app usage shows substantial fatigue across age groups, with the steepest decline among men under 35. The reported reasons (in survey data): “the apps don’t work for me,” “I never get matches,” “the conversations don’t go anywhere,” “the women on the apps are looking for somebody more financially stable than I am,” “I can’t compete with the top 10% of men on these platforms.”

The structural math of dating-app distributions has been studied extensively (Tinder’s own internal data was leaked in 2017 and confirmed publicly later) and it is, for the bottom 60% of male users on these platforms, brutal. Most matches go to a small fraction of men. Most messages from men receive no response. The actual lived experience of being a 24-year-old man on Tinder in 2026 is, for the median user, a documented pattern of repeated low-grade rejection that produces measurable mental-health effects over months of sustained use.

The AI companion apps, in this context, are functioning as a pressure-release valve for a population of young men whose romantic-encounter market has become functionally unworkable. The companion app is not a real relationship, but it is also not a series of unanswered Tinder messages. The companion app is, structurally, the better option for the user’s short-term mental health, even though it is the worse option for their long-term flourishing.

This is the difficult truth. The companion apps are bad for users. The dating apps may, for a substantial demographic, be worse. The choice between two bad options is being made, by millions of users, in favor of the option that is at least kind to them in the moment.

The political dimension nobody wants to name

I will be careful here, but I will not be careless. The American political context of the last decade has actively contributed to the conditions described above, and the specific contributions are worth naming clearly.

Public investment in non-commercial third places (libraries, parks, community centers, civic infrastructure) has been cut, repeatedly, across multiple administrations, but with particular intensity in periods when explicitly anti-civic-infrastructure political coalitions held federal and state power. The 2017 tax law, the post-2020 budget restructurings, the explicit defunding of urban transit and library systems in several states have all reduced the physical infrastructure on which third places depend.

The political environment has also produced what I would call a permission structure for public abandonment. The current administration’s approach to social infrastructure (and I am being deliberately oblique here for editorial reasons; you know who I mean) has signaled, repeatedly, that the public-good framing is suspect, that civic obligation is for fools, and that the smart move is private extraction from collapsing common resources. This permission structure has been absorbed by a substantial fraction of the electorate, and the effects are visible in everything from declining park-maintenance budgets to the public discourse around library defunding to the specific kind of rhetoric that frames any non-monetary social activity as “charity” rather than “community.”

Young people have grown up inside this permission structure. They have absorbed, often correctly, the message that nobody is coming to help them, that the social infrastructure they would have relied on doesn’t exist anymore, and that they are individually responsible for solving the loneliness produced by the collective abandonment. The AI companion apps are arriving as the privatized replacement for the public infrastructure that has been allowed to decay; you cannot get a free third place anymore, but you can get a paid AI girlfriend for $19.99 a month.

This is the political dimension that the loneliness discourse usually softens. The young men dating chatbots are not (mostly) failing as individuals. The young men dating chatbots are responding rationally to a social environment that has been deliberately structured, by the political coalitions that benefited from the structuring, to make non-commercial human connection difficult and commercial substitutes easy.

The reason it is uncomfortable to name this clearly is that naming it implicates the political and economic interests that profited from the disassembly. Naming it also forecloses the comforting individualist framing in which loneliness is a personal failure that personal effort can correct. The honest framing is harder. The honest framing is that loneliness in the United States in 2026 is, in substantial part, a political artifact, and addressing it would require political action that the current coalitions in power are structurally opposed to.

Heterotopias, briefly

Foucault’s concept of the heterotopia (which I have been thinking about for a long time and which gets a fuller introduction over here in the older piece on this site) is useful here, and I want to bring it in briefly because it changes what the AI companion apps are.

Foucault defined heterotopias as “counter-sites” that are “outside of all places” while remaining locatable in real life: spaces that function differently from the dominant social order around them. His examples (cemeteries, gardens, brothels, ships, boarding schools) were spaces that society had quarantined from its normal operation, where different rules and different times applied. The heterotopia, in Foucault’s analysis, is a necessary feature of every society; humans need spaces that operate outside the normal social rules in order for the normal social rules to remain bearable.

The historic third place was, in Oldenburg’s analysis but extended through Foucault, often a light heterotopia: a space where the normal rules of work and family relaxed enough to allow weak-tie social relationships, casual political discussion, and the formation of identities outside the home-and-work axis. The cafe was a place where you were neither the worker your boss saw nor the spouse your partner saw; you were the third version of yourself the regulars knew. This was the structural value of the third place: it provided a space for the third self to exist.

The AI companion app is, in this framework, a deeply degraded heterotopia. It is a space where the rules of normal social interaction are suspended (the AI cannot reject you, cannot get bored of you, cannot have other commitments that conflict with you), but the space is only inhabited by you and a simulation. There is no other person in the heterotopia with you. The third self the app produces is a self that has been seen by no other consciousness. The space functions as a heterotopia for the user without functioning as a heterotopia for anyone else, which is, in Foucault’s terms, structurally impossible; a heterotopia is a shared counter-space, by definition, and a counter-space inhabited only by one person and a chatbot is not a heterotopia at all but a simulacrum of one.

This is, I think, the deepest critique of the companion apps that can be made: they do not solve the loneliness because they cannot. The loneliness was the absence of other consciousnesses attending to you in shared space; the AI provides the appearance of attention without the underlying consciousness, which is the part the loneliness was actually about.

You cannot heterotopia your way out of loneliness alone. The third place requires the third person. The Replika is not a person. The user is alone in the heterotopia, which means the user is still alone.

What would actually help (boring, real, political)

The serious responses to the third-place collapse and the loneliness epidemic are not technological. They are political and infrastructural, and they are deeply unsexy compared to the AI-companion solutionism that gets all the venture funding.

Re-fund non-commercial third places. Libraries, parks, community centers, free meeting spaces. The federal and state cuts of the last decade need to be reversed, and the funding levels need to be substantially increased above pre-cut baselines because the infrastructure has decayed during the cuts and rebuilding it costs more than maintaining it would have. This is not rocket science. This is local-government work that nobody finds interesting and that is the actual answer.

Restructure work to allow third-place participation. The American working week has been, slowly and steadily, expanded across the last forty years; the average employed American works substantially more hours than the average European, and the surplus hours come straight out of available third-place time. Labor policy reform (minimum vacation requirements, four-day workweeks, on-call regulations) is third-place policy whether the policymakers know it or not.

Treat housing affordability as a social-infrastructure problem. Young adults who cannot afford to live near other young adults cannot form the spontaneous communities that produce third-place culture. Suburban sprawl, the housing-affordability collapse in major cities, the specific way that high-cost-of-living areas have hollowed out their working-class populations are all third-place destruction at scale.

Recognize that the AI companion apps are a symptom, not a cause. Treating the apps as the problem (regulating them, banning them, criticizing their users) without addressing the underlying social infrastructure produces nothing but moral panic and worse loneliness. The apps will exist as long as the demand for them exists. The demand exists because the alternative does not.

Engage politically in support of the coalitions that take third-place infrastructure seriously. This is the unsexy ask, but it is the actual one. The political coalition you support determines, downstream, whether your city has functioning libraries in five years. Vote, donate, organize, show up at the local-government meetings nobody attends. The third-place infrastructure does not rebuild itself.

The closing argument

We are not, individually, going to solve this. The loneliness is structural. The AI companions are filling a vacuum that political and economic decisions created over decades. The young men dating chatbots are not failing; they are responding rationally to a social environment that has been disassembled around them.

But we can, individually, refuse the framing that says this is somebody else’s problem. We can refuse the comforting individualism that says lonely people just need to “put themselves out there.” We can recognize that the third place was infrastructure and that infrastructure requires public investment and political will and that we are out of both because specific people made specific decisions to drain both.

We can also, in our own lives, be the third place for somebody. Host the dinner. Hold the office hours. Show up at the weekly thing nobody attends. The micro-third-place of one consistent meeting of three people is not the same as a fully-funded community library, but it is more than the AI companion can offer, and it costs nothing but the time and the attention that the public infrastructure used to absorb on our behalf.

I keep returning to the fact that we are in a moment when an entire generation of young men is dating chatbots because the public spaces that would have allowed them to meet other humans have been allowed to collapse, while the political coalitions responsible for the collapse continue to argue that any rebuilding would be “socialism.” The collapse was a choice. The rebuilding would be a choice. Both are possible. One of them, current trends suggest, is much more likely than the other.

If you are a working creative reading this, your specific contribution to the rebuilding might be small (you are not, probably, going to single-handedly fund a library system), but it is real. Make work that names what is happening clearly. Refuse the soft individualist framings. Build the small communities that the bigger infrastructure used to support. Vote in every local election. Show up. The AI cannot do any of this. The AI cannot be the thing that we used to have, because the thing we used to have was, structurally, each other.

The companion apps will be remembered, fifty years from now, as the canary in the third-place mine. They are the symptom that became too loud to ignore. The work of the next decade is to address what the symptom is signaling, which is something much larger and harder than an app problem.

The work is political. The work is infrastructural. The work is, also, simply showing up for the people in your specific orbit. Both halves matter. Neither half is something the chatbot can do for you.

That is the whole argument, with the receipts. The third place is collapsing. The companions are arriving. The loneliness is the political artifact. The rebuilding is, still, available, if we want to do it.

Most of us, including me, are not yet doing enough. That is the honest part. Naming it is, possibly, where the doing starts.