Every pitch for an AI companion begins with the same slide: a loneliness epidemic, a friendship recession, a population with no one to talk to. The researchers who run the largest ongoing study of American friendship keep finding something else. People have friends. About two percent have none. What they lack is the evening, the money, and the energy to see them. That is a different problem, and software does not solve it.
The American Friendship Project is run by Jeffrey Hall at the University of Kansas, Natalie Pennington at Colorado State, and Amanda Holmstrom at Michigan State. Its first large survey, nearly 6,000 adults across 2022 and 2023, published in PLOS ONE, found that the average American has four or five friends, that over 75 percent are satisfied with how many friends they have, and that about 2 percent report having none at all. The dissatisfaction was somewhere else: 42 percent said they were not as close to their friends as they would like.
The 2025 wave, reported in the summer of 2026, asked why. More than 2,000 people were surveyed and the results compared with public polling back to the 1990s. Time with friends is down. Fifty-seven percent said they had visited with friends or family the previous day, lower than in the older polls, and 58.6 percent see friends at least monthly, also a decline.
Then the reasons people gave for not seeing friends. Not enough money: 38.2 percent. Mental health: 31.2 percent. Too busy or too tired: 30.8 percent. Not enough time: 29.7 percent. Putting family first: 24.1 percent. Work getting in the way: 19.2 percent.
And the finding that gives the study its headline: with all that, people did not report more loneliness. “My read on the data,” Hall said, “is people are subbing a night out with friends for a night in with family.” And: “It seems people have friends, but what they really want is more time to spend with those friends.”
The story that got sold
Set that beside the story the technology industry tells. In 2025 Mark Zuckerberg said the average American has fewer than three friends and has demand for around fifteen, and offered AI as the way to close the gap. Every companion startup since has opened with a version of the same claim. The pitch requires a population of people with nobody, because a product that supplies somebody only makes sense if somebody is what is missing.
The friendship data does not describe that population. It describes people with four or five friends, a full calendar, and a tight budget. Their problem is logistics. A night out costs money. A babysitter costs money. The friend lives forty minutes away and both of them work. By the time the dishes are done it is nine o’clock, and the couch is right there.
This matters because the two diagnoses lead to opposite treatments. If people have no one, a conversational product is at least pointed at the problem. If people have someone and cannot get to them, then a conversational product is pointed at the symptom, and it makes the cause worse. The hour spent talking to an app at nine o’clock is the hour that might have been a phone call. Americans already spend twice as much time with AI companions as on dating apps, and that time came from somewhere.
It is fair to say that this publication has leaned on the epidemic framing too. Surveys in which people describe themselves as lonely are real, and so is the distress in them. But “many people feel lonely sometimes” and “people have no one” are different claims, and the second is the one the business case depends on. The best available evidence does not support it.
What 38 percent means
The top reason people do not see their friends is money. Not phones, not social anxiety, not the death of the bowling league. Money.
That should change how the whole subject is discussed. Friendship has a cost of entry that has been rising for a long time: the price of a meal out, a ticket, a drink, the gas, the sitter. There are fewer places to be together for free. When Hall says more people are finding connection at home, part of what he is describing is that home is what they can afford.
The second and third reasons are mental health and exhaustion. Those feed each other, and they feed the first. A person working long hours for not quite enough is tired, and a tired person cancels, and a person who has cancelled three times stops being invited.
None of these has a technological fix, and it is worth noticing how convenient the technological framing is for the people offering it. An epidemic of loneliness is a market. A shortage of time and money is a political and economic condition, and no one can sell a subscription against it. So the first story gets told at product launches and the second gets told in a university press release in July.
The use that fits and the use that does not
There is still a real question about whether talking to an AI helps anyone, and the research on whether companions help or hurt loneliness is mixed for good reasons. Some people are isolated: the housebound, the recently bereaved, the person in a new city. Two percent of a country is millions of people, and their situation is real.
But a product built for the two percent and marketed to the ninety-eight is doing something else to the ninety-eight. It is offering the tired person with four friends an option that requires no money, no travel, no scheduling, and no energy, at exactly the moment they lack all four. It will win that contest most nights. And each night it wins, the forty-minute drive gets a little harder to justify.
The study found one more thing, almost in passing. People who made time for both friends and family reported higher life satisfaction and less loneliness than those who chose one. The benefit came from the making time. That is the part with no shortcut.
So the useful advice is unglamorous. Text the friend. Suggest the free thing: a walk, a kitchen table, an errand run together. Accept that it will be a little inconvenient and go anyway. Everybody needs someone, and most people, it turns out, already have them. They are forty minutes away and just as tired.
The loneliness epidemic may be the wrong name for what is happening. A better one would be the Tuesday problem: there is nothing wrong with our friendships that a free evening would not fix, and a free evening is the one thing nobody is selling.
Sources: University of Kansas, “Research shows people aren’t lonelier, they’re changing how they spend their time” (July 28, 2026; American Friendship Project 2025 survey, more than 2,000 participants; 57 percent, 58.6 percent; reasons for not spending time with friends; Hall quotations). Hall, J. A., Pennington, N., and Holmstrom, A. J., American Friendship Project, PLOS ONE (2024; nearly 6,000 adults surveyed 2022 and 2023; four to five friends on average; about 2 percent with none; over 75 percent satisfied with number of friends; 42 percent not as close as they would like). GCN (November 8, 2026). Mark Zuckerberg, interview with Dwarkesh Patel (April 2025; “fewer than three friends,” demand for fifteen).






