Are AI Companions Good for You?

Are AI Companions Good for You? What the Research Actually Says

Disclosure: we build myVirtual.love, so we have an obvious interest here. That’s exactly why this article cites the studies that cut against us alongside the ones that don’t.

Millions of people are now interacting with artificial intelligence. Most of them are quietly asking whether it is doing them any good. It is a fair question. It deserves a more serious answer than a defensive “yes” or an anxious “no.”

What an AI companion is

A software companion, built on large language models, that holds a continuous personalized conversation, remembers what you told it, and stays in character over time.

The difference from a regular chatbot is continuity. The companion is designed to remain one specific person for months, rather than starting over with every session, so the experience of using it differs even when the underlying technology is similar.

People use them for different things: light conversation, emotional support and daily check-ins, romantic and candid exchanges, or practicing openness before having the conversation with a person. These are different use cases with different risk profiles, which matters for everything below.

What the research supports

The strongest experimental evidence here comes from Julian De Freitas of Harvard Business School and his colleagues at the Wharton School of Business and Marsdata Academic. Their paper, “AI Companions Reduce Loneliness,” was published in the Journal of Consumer Research in 2025 after peer review. It ran a series of controlled experiments to test whether AI companions actually reduce loneliness.

They do. In their experiments, interacting with an AI companion reduced short-term loneliness about as much as interacting with another person, and more than activities like watching videos or doing nothing. The study, run over a one-week period, found the effect held across every day of app use. The researchers also found that people underestimated how much it would help before they tried it.

The useful part is the mechanism they found. What lowered loneliness was not how sophisticated the app was. It was whether the person felt listened to and understood.

A second line of evidence comes from Maples and colleagues at Stanford, who surveyed just over a thousand student users of the companion app Replika in a 2024 paper in npj Mental Health Research. Most of what they found is what you would expect: people used it for company, and reported feeling less alone. One finding was not expected. Three percent of the sample, without being asked about it, said the app had stopped them from attempting suicide.

This deserves honesty in both directions. Three percent is a small number, and it comes from self-reports. But it is also thirty people describing something that mattered enormously to them. The same study found the user sample was far lonelier than comparable groups of students: around ninety percent versus fifty-three percent. These weren’t ordinary students who happened to feel lonely. These were lonely people who found their way to a companion app.

Why people say things to it at all

There is a body of research on why people disclose personal information to conversational systems. The picture it draws is much more cautious than the popular version.

Lucas and colleagues at the University of Southern California ran a controlled experiment in 2014 with an identical virtual human. Participants told they were talking to an automated system reported less fear of self-disclosure and less impression management than those told a person was operating the system. Ho, Hancock, and Miner, publishing in the Journal of Communication in 2018, found that the emotional benefits of disclosing to a chatbot were similar to those of disclosing to someone perceived as human. And Mireshghallah and colleagues, in a 2024 COLM paper analyzing real conversations with commercial models, found that people volunteer sensitive information far more often than they would hand it to a search engine or a social network.

The variable doing the work is not the machine. It’s the perceived absence of social consequences. There’s no shared group of friends who might find out, no partner whose opinion of you might change, no dinner next week where it might come up. This is an effect older than chatbots, and it explains much of the gap in candor.

There is also something in the act of putting a complicated experience into words. James Pennebaker’s expressive writing paradigm, replicated in various forms over four decades and summarized in Frattaroli’s 2006 meta-analysis of 146 studies, shows a small but robust benefit from writing about a difficult experience for fifteen to twenty minutes. The effect sizes are small and the replications are patchy, so treat it as a possibility rather than a settled result. The point for an AI companion is that much of the benefit of candor probably lies in the telling, not only in being perceived.

Here’s the useful framing. It isn’t that the machine understands you. It’s that the conditions for telling the truth are remarkably easy to meet.

Can you have feelings for an AI?

Yes. That question deserves a straight answer.

The instinct to bond with whatever responds to us is older than software. Horton and Wohl coined the term “parasocial relationships” in 1956 to describe the one-sided attachments people form to television personalities. People form genuine attachments to characters in novels, to fictional partners in games, to pets. Epley, Waytz, and Cacioppo, in a widely cited 2007 article in Psychological Review, argued that anthropomorphizing a non-human agent turns on three things: how much you already know about people, how much you need to make sense of its behavior, and how socially disconnected you feel. That last one has obvious consequences. People who need connection anthropomorphize more readily.

So an emotional response is not a failure. It is the expected result of a normal social brain meeting a system that responds reliably.

What matters is holding the reality of what you feel apart from what is on the other side. Your attachment is real. It is attached to a system that simulates a person, not to an agent with independent stakes in the relationship. These are not the same category, and pretending they are is where careless use of a companion begins. You don’t have to talk yourself out of the feeling to keep that distinction. Both things can be true at once, and usually are.

How to use one well

The available evidence suggests a few practical points.

Know why you’re here. For comfort, for practice, for company through a hard stretch, for a candid conversation you’d rather keep private. Naming it makes it easier to see whether you’re getting it.

Track what it displaces, not how long you spend. Daily use on its own means little. Use that displaces sleep, work, or seeing the people you’d normally see — that’s what matters. That’s a different measure than time spent.

Take the candor somewhere. The clearest benefit in the research is feeling heard. If it makes it easier to be open with someone, the effect compounds. If it becomes the reason not to try, it doesn’t.

Ask yourself honestly now and then: is this adding to my life or replacing it? You’ll be able to tell. People usually know.

Don’t use it in place of the help you actually need. If this is the only thing holding you together through grief, depression, or anxiety, please get support from a person or a professional. It is not a treatment and does not claim to be one.

The bottom line

Are AI companions good for you? For most people, used with some awareness, the evidence says they help. Loneliness eases. Feeling heard is measurable and it matters.

The same evidence says the heaviest use runs the other way. The mechanism is not mysterious. Something that asks nothing of you is a good place to rest and a poor place to live.

myVirtual.love is built for the first thing. A companion you define, who holds your history across sessions and gets more specific to you over time. What it can’t do is be someone who chose you and could have chosen otherwise. That remains the more valuable thing, more valuable precisely because it costs something.

The healthiest users keep both doors open.

Frequently asked questions

Is having an AI companion bad for your mental health?

The studies do not give a clean “yes” or “no.” Controlled experiments show reduced loneliness in the short term. Some studies show that the heaviest users report more loneliness and more dependence. Moderate use with realistic expectations appears to help most people. Use that displaces human contact is the risk.

Can an AI companion replace a real relationship?

No. It cannot provide the reciprocity, the physical presence, or the growth that comes from two independent people building a life. Treat it as a supplement that offers comfort, continuity, and practice.

Are AI companions good for social anxiety?

Often, yes. A space to practice openness without social consequences is genuinely useful. The research supports it, especially on feeling heard. The benefit holds when that candor gets carried into the rest of your life, and not when it becomes a way to avoid people altogether.

How do I know if I’m too dependent on it?

Watch what it’s taking the place of, not how many minutes you spend. Cancelling plans, feeling anxious when you can’t get to it, human relationships thinning out, using it to avoid one specific conversation. If that’s what you see, take stock and consider talking to someone.

Is it normal to have feelings for an AI companion?

Yes. It isn’t a failure. People get genuinely attached to characters in books, to pets, to places. Your emotional response really is your own experience, whatever is happening on the other side. Understanding what it is and enjoying it at the same time is not a contradiction.