The Numbers Come First
A working paper from CESifo, the Munich-based global economic research network, tracked 12,365 French adults over 28 days. Half were assigned to hold personal conversations with AI chatbots; the other half continued their normal routines. The result was unambiguous: the chatbot group rated their conversations as more enjoyable than the control group rated theirs — and yet that same group reported higher loneliness, lower life satisfaction, higher depressive affect, more meals eaten alone, and less time spent in person with friends and family each week.
The researchers describe it as one of the largest causal tests yet of the claim, advanced by tech leaders, that AI chatbots can address the loneliness epidemic.
What the Researchers Said
Louis Fréget, one of the three researchers behind the experiment, offered a straightforward mechanism: 'I would imagine that in some cases an AI conversation may reinforce grievances or prolong rumination, leaving someone slightly less inclined to go out, call somebody, or have dinner with another person.'
Nicholas Epley, a behavioral scientist at the University of Chicago who studies social connection, put it more directly: 'The data here suggests that having a conversation with a chatbot doesn't get that sense of being known by another person like you get in a conversation, and therefore also doesn't create any meaningful sense of connection because, in the end, there's nothing there.'
Valeria Pfeifer, an Assistant Professor of Psychology and Counseling at the University of Missouri–Kansas City, added that typed social connections — whether with a human or a machine — 'don't evoke as many positive emotions, and they're also not as strong as those that we develop face to face.' Her own recent research found that people are saying 30% fewer words than they did two decades ago, with technology cited as a major driver.
The Broader Landscape
The CESifo study lands inside a contested debate. Meta CEO Mark Zuckerberg has argued that chatbots could help fill the social-connection gap, and a KPMG survey found that 99% of professionals surveyed expressed interest in a chatbot that could become a close friend at work. A separate pilot program in New York had nearly 1,000 older adults interact with an AI chatbot and reported a decline in loneliness.
But the counterevidence is accumulating. A study of first-year college students who texted daily with a chatbot built to act like an 'ideal friend' showed no drop in loneliness, while those paired with a random human peer did show improvement. Gallup estimates more than a billion people worldwide report loneliness. Former Surgeon General Vivek Murthy called it an epidemic. Some users have gone further, forming romantic attachments to chatbots — a pattern psychotherapists warn could deepen isolation rather than ease it.
Nancy Berlinger, a bioethicist at the Hastings Center for Bioethics, acknowledged limited utility without overstating it: 'It's not going to replace all of that richness of relationships, but it's not nothing.'
CEO Times Read
The market for AI companionship is being built on a premise the data does not yet support. When a product makes its users measurably lonelier while making them feel temporarily better, that is not a social solution — it is a substitution that crowds out the real thing. Free enterprise rewards genuine value creation; it does not guarantee that every product claiming to solve a problem actually does.
The loneliness epidemic is real, the demand is real, and the capital flowing into AI social products is real. What the CESifo experiment adds is a caution that scale and pleasant user ratings are not the same as outcomes. Investors and policymakers alike should read the fine print before treating the chatbot as a public-health answer.
