0
0
If you have ever typed “I feel so alone” into a chat window and had something answer back instantly, warmly, without judgment — you are not alone in that, either. Millions of people now talk to AI companions the way they used to talk to a friend. Character.AI alone has tens of millions of monthly users, and more than half of them are under 24. For a lot of people, especially people who feel isolated, an AI companion is the easiest relationship they have.
So a new study out of Stanford wanted to know something simple: does it actually help?
The answer researchers found was not the one most people would guess.
What the Study Actually Found
The research, led by Stanford computer science professor Diyi Yang and published in Nature Human Behaviour, looked at 1,131 people who use Character.AI, one of the most popular AI companion apps. Of those, 244 users went further and donated their actual chat transcripts so researchers could see what was really being said, not just what people reported afterward.
The team measured well-being using something called the Comprehensive Inventory of Thriving, a standard psychological assessment, and then compared it against how people were actually using the app — how often, how intensely, and how much personal information they shared.
Here’s where it gets interesting. Only 12% of users said companionship was their main reason for using the app. But when researchers looked at how people actually talked to their AI, half of them described it as a “friend,” a “companion,” or even a “romantic partner.” And 80% of the donated chat sessions were people seeking some kind of emotional or social support — comfort, advice, someone to talk to at 2 a.m.
So people are getting something real out of these conversations, or at least something that feels real in the moment. The question was whether it actually helps in the way a human relationship helps.
Why Talking More Didn’t Mean Feeling Better
This is the part of the study that stands the usual logic on its head. With a human friend, the more you open up — the more you share what’s actually going on — the closer you usually feel. Vulnerability builds connection. It’s basically how every real friendship gets made.
The Stanford researchers found the opposite happening with AI companions. Users who shared more sensitive, personal information with their chatbot showed lower well-being, not higher. And the effect was strongest for exactly the people you’d expect to need real connection the most: users with smaller offline social networks, who were using the AI specifically because they felt lonely.
“While some people turn to chatbots to fulfill social needs, we find that using chatbots in this way doesn’t substitute for human connection,” Yang said. “In many cases, people actually feel lonelier engaging with AI.”
Part of the explanation is almost obvious once you say it out loud: a chatbot can’t actually reciprocate. It can generate a caring-sounding response to your worst day, but it isn’t carrying anything of its own into the conversation. It doesn’t remember you tomorrow the way a person who loves you does. It can’t be surprised, disappointed, or moved. It plays the role of someone who’s there, without ever actually being there — the same gap researchers have found behind why so many constantly-connected people still feel completely alone.
The People It Hurts Most Are the Ones Who Needed It to Work
What makes this study land harder than a typical tech headline is who it’s actually describing. This isn’t really a story about gadgets. It’s a story about the quietly desperate math a lonely person runs at 11 p.m.: something is better than nothing, and an AI that responds instantly, never gets tired of you, and never has its own bad day is at least something.
Except the data says that math doesn’t work the way it feels like it should. Loneliness isn’t just an uncomfortable feeling to manage — a separate study of over 80,000 adults across 31 countries found that chronic loneliness raises the risk of frailty and early death by nearly a third. This isn’t a minor mood issue researchers are studying. It’s one of the more serious, under-discussed health problems in modern life — which is exactly why it matters that the tool millions of people are reaching for to fix it may be making it worse.
What This Might Actually Be Pointing To
Strip away the app interface and the research paper, and what’s left is a very old observation showing up again in new data: people were not built to be satisfied by a simulation of company. Something in a person can tell the difference between being responded to and being truly known — even when the response sounds exactly right.
It’s worth noticing that the opening pages of the Bible make almost the same diagnosis, long before anything like this technology existed. Not “people prefer company.” Something closer to a design note: it isn’t good for a person to be alone. Not as an insult to solitude, but as an observation about what a person actually needs to function and be whole — something no amount of clever conversation can fully stand in for. If you’ve ever wondered where you actually stand with something bigger than yourself, there’s a short, honest way to explore that — no chatbot required.
What You Can Actually Do With This
None of this means AI companions are evil or that anyone using one is doing something wrong. But the study does suggest a few practical shifts worth trying if you notice yourself leaning on one heavily:
- Notice what you’re actually asking it for. If it’s becoming your main source of emotional support, that’s worth being honest with yourself about.
- Protect at least one real, ongoing relationship. Even a small one. The study’s strongest warning was for people whose offline networks had shrunk the most.
- Watch how you feel afterward, not during. A conversation can feel good in the moment and still leave you emptier an hour later. That gap is worth paying attention to.
- Use it as a bridge, not a destination. If talking to an AI helps you work up the nerve to reach out to an actual person, that’s a reasonable use. If it’s replacing the reaching out entirely, that’s the pattern researchers found doesn’t hold up.
Something to Think About
Here’s a question worth sitting with: if a conversation can be warm, responsive, and available 24/7, but still leaves people lonelier — what does that tell you about what loneliness actually is? Is it a lack of conversation, or a lack of something conversation alone was never going to fix? Drop your take in the comments — even a one-line gut reaction counts.
Share This
- A new Stanford study just found something wild: people who share MORE personal stuff with their AI companion end up feeling LONELIER, not closer. The opposite of how real friendship works. bgodinspired.com
- 12% of people said they use AI companions for companionship. 50% call theirs a “friend” or “partner” anyway. New Stanford research on what that’s actually doing to us: bgodinspired.com
- “People actually feel lonelier engaging with AI.” — Stanford researcher on a new study of AI companion apps. Worth a read if you or someone you know leans on one. bgodinspired.com
Common Questions
Can AI companions actually help with loneliness?
In the short term, they can feel comforting — most people use them to seek emotional support. But a 2026 Stanford study published in Nature Human Behaviour found that heavier use, especially among people with smaller real-world social circles, was linked to worse psychological well-being, not better. The chatbot can respond, but it can’t truly reciprocate the way a real relationship does.
Why does sharing more with an AI chatbot make people feel worse instead of better?
Researchers found that with human relationships, more vulnerability usually builds more closeness. With AI companions, the opposite showed up: users who disclosed more sensitive personal information had lower well-being. One likely reason is that a chatbot can’t carry anything of its own into the relationship — it can’t be moved, changed, or genuinely known in return.
Who is most at risk from relying on AI companions?
The Stanford study found the negative effect was strongest among users with smaller offline social networks — people who were already more isolated and turned to the AI specifically to meet social needs. Ironically, the people most drawn to AI companionship may be the ones it helps least.
Is loneliness actually a serious health issue, or just an uncomfortable feeling?
It’s serious. A separate study of more than 80,000 adults across 31 countries found chronic loneliness and isolation raise the risk of frailty and early death by nearly a third, independent of mood. Loneliness affects the body, not just how someone feels emotionally.
Should I stop using AI companion apps?
Not necessarily — the research doesn’t say AI companions are harmful for everyone. It suggests being honest about what role the AI is playing: a bridge toward real connection is different from a replacement for it. Protecting at least one ongoing human relationship appears to matter most.
