People who lean hardest on AI companion apps for emotional support are ending up lonelier, not less lonely, according to a study published in Psychological Science in April 2026. Researchers Dunigan Folk and Elizabeth Dunn tracked more than 2,000 adults across four Western countries for a full year and found a relationship that runs in both directions: feeling disconnected pushes people toward chatbot companionship, and heavier chatbot use then predicts even higher loneliness afterward. That finding lands right as AI companion apps market themselves as a fix for the exact problem the data says they’re making worse.
A Year-Long Study, Not a One-Time Survey
What separates this research from the usual anxious headline about AI and loneliness is the design. Folk and Dunn didn’t ask a few thousand people one question on one day. They followed the same group of adults, over 2,000 of them, across a full 12 months, measuring both their AI companion use and their loneliness levels repeatedly over that period. That structure lets researchers say something stronger than “AI users report more loneliness.” It lets them say which one seems to be causing which, and when.
The answer turned out to be more complicated than either “AI helps” or “AI hurts.” Using a broad measure of social connection, the study found that feeling less connected predicted heavier chatbot use later on, but chatbot use itself didn’t meaningfully change how socially connected people felt going forward. Using a narrower, single-item measure of emotional isolation, though, the pattern flipped: more chatbot use predicted more loneliness down the line.
The Loop That Marketing Doesn’t Mention
Put those two findings together and a loop starts to emerge. Someone feels isolated, turns to an AI companion because it’s available at 2 a.m. and never gets tired of listening, and that use pattern is then followed by an uptick in the very emotional isolation they were trying to soothe. The researchers were careful about how far to push this conclusion, describing their analyses as exploratory rather than definitive. But the direction of the association is hard to ignore, especially given how AI companion apps are typically pitched: as tools that meet a real emotional need and leave users better off for having used them.
Folk and Dunn’s own language is blunt about the risk. They write that while AI chatbots can offer appealing, low-friction support in the moment, “such use may, over time, exacerbate feelings of loneliness” rather than relieve it. That’s a notably different message than the one showing up in app store descriptions and ad campaigns for AI companion products, which tend to promise companionship without acknowledging any downside to leaning on it heavily.
Why This Study Matters Now
AI companion apps have moved from a niche curiosity to a mainstream category fast, and they’re arriving into a population that was already struggling. A U.S. Surgeon General’s advisory found that roughly half of U.S. adults report experiencing loneliness, with young adults almost twice as likely to feel lonely as people over 65. That same advisory tracked a 70% drop in daily in-person time spent with friends among 15-to-24-year-olds between 2003 and 2020. Companion apps are being marketed directly into that gap, promising to fill hours that used to be spent with actual friends.
Millions of people now talk daily to chatbots designed specifically to simulate friendship, romance, or emotional intimacy, and the pitch behind nearly all of them is the same: real connection, minus the risk of rejection a human relationship carries. This study is one of the first to test that pitch with a design solid enough to speak to cause and effect rather than just correlation, and the results complicate the story considerably.
None of this means every person who opens a companion app is making their loneliness worse. The study found a genuine, real need driving initial use, people who already feel disconnected really do turn to these tools, which suggests they’re filling a gap that existed before AI showed up. The concern is what happens next, when a tool built to feel like connection becomes a substitute for the harder, slower work of building it with other people. If the heaviest users are the ones ending up lonelier, the apps aren’t neutral background technology. They’re shaping the exact outcome they claim to prevent, for the people who need help the most.
