AI companions won't fix loneliness. Here's what will

It's 1 a.m. and the apartment is quiet in the specific way that makes a phone feel heavy. Somewhere in the app store there's a companion that will answer instantly, remember your dog's name, never be too busy, never look at its watch. Millions of people have downloaded one. If you're one of them, this page is not going to make fun of you.

Reaching for comfort at 1 a.m. is one of the most human things a person can do. And the comfort is not fake. We'll get to the research, and it's more favorable to AI companions than you might expect. But there's an honest line between what these apps genuinely provide and what loneliness is actually the absence of, and almost nobody selling an AI friend has a reason to draw it. We do. And, full disclosure, we build AI too, which we'll get to plainly. So here it is, drawn carefully: what AI companionship gives you, what it can't, what the research knows and doesn't, and the gentlest real way out.

Do AI companion apps help with loneliness?

In the short term, honestly, yes. A Harvard Business School study by Julian De Freitas and colleagues, since published in the Journal of Consumer Research, found that talking with an AI companion reduced momentary loneliness about as much as talking with an actual person, and more than passive scrolling-adjacent activities like watching videos. Across a week of daily check-ins, the relief showed up consistently. The mechanism wasn't magic: it was feeling heard. When the AI listened and understood, loneliness dropped.

Anyone tempted to sneer at that should sit with what it means. Feeling heard is scarce enough that software supplying it measurably helps. And for some people (housebound, disabled, grieving at 3 a.m., living somewhere the nearest anything is an hour away) an AI companion is genuinely better than the actual alternative, which was nothing. If that's you, or someone you love, the companion isn't a failure. It's a shelter. Nobody should be shamed for standing in it.

This also isn't a fringe behavior anymore. Common Sense Media's 2025 national survey found 72 percent of US teens have tried an AI companion, just over half use one at least a few times a month, and 13 percent use one daily. Whatever this is, it's not a niche; it's a generation's response to a real, measured collapse in human connection.

So the question was never whether AI companionship provides something real. It does. The question is what category of thing it provides, and whether that category can ever add up to the thing you're actually missing.

Can AI be your friend?

Here's the direct answer: an AI companion can be a real comfort, a real habit, even a real relationship of a kind, and it still belongs to a different category than friendship. Not a worse version of the same thing. A different thing.

Think about what actually makes a friendship matter to you. It isn't just the conversation. AI can increasingly match the conversation. It's the facts around the conversation. Your friend has an inner life of their own, and they spend some of it on you. They chose you, out of all the people they could have chosen, and they could stop, which is exactly why their staying means something. Their attention costs them something: an evening, a favor, patience on a week when they had none to spare. And the witnessing runs both ways. You know their mess too. You've seen them fail and stayed. Friendship's weight comes from mutual cost and mutual stake: two people who could each walk away, repeatedly not walking away.

An always-available, always-agreeable presence can't offer any of that, by design. Not because the technology is bad, but because the things that give friendship its meaning are precisely the things that make it inconvenient. Remove the friction and you haven't perfected friendship; you've made something else, closer to a mirror with excellent manners.

Read the left column fairly: it's a genuinely good product. Read the right column honestly: it's a different product. Loneliness, in the research, isn't the absence of soothing conversation. It's the absence of being embedded, woven into other human lives that would have a hole in them where you used to be. An AI companion can quiet the signal. It cannot supply the thing the signal is about.

The real risk isn't the comfort. It's what the comfort replaces

Let's be precise about what the evidence does and doesn't show, because this is where most takes get lazy in one direction or the other.

What's known: the relief is real but short-horizon; the Harvard team measured days and weeks, not years. And at the heavy end, the picture darkens. A 2025 MIT Media Lab and OpenAI study (a four-week randomized trial with about 1,000 people, alongside an automated analysis of nearly 40 million real conversations) found that the heaviest daily users reported more loneliness, more emotional dependence, and less real-world socializing, with the pattern strongest among people who used the bot for emotional conversation and came to see it as a friend. Most users showed none of this; the trouble concentrated in the top slice of use.

What's not known, and say it plainly, because few do: whether heavy use causes the loneliness or lonely people simply use more. Both studies are consistent with either. Nobody has yet run the multi-year study that would settle whether AI companionship displaces human connection over time. The longitudinal evidence is thin, and anyone claiming certainty in either direction is ahead of the data.

But here's why displacement is the risk worth naming anyway, even under uncertainty. Loneliness has a job: it's the signal that pushes you toward people, the same way hunger pushes you toward food. The comfort of an AI companion works directly on that signal. And the human alternative it competes with is brutally expensive: a close friend takes 200+ hours of shared time, while the average American's friend-time budget has fallen to about twenty minutes a day. When the soothing option costs nothing and the real option costs 200 hours, comfort doesn't have to be better than friendship to win. It only has to be good enough tonight, every night, to keep the urgency below the threshold where you'd have done the harder, slower, better thing. That's not a sci-fi scenario. That's just how good products beat difficult ones.

Full disclosure: we build AI too

You should know who's talking. AroundMe is a technology product, and there's AI in it, so this isn't a hand-wringing essay about the machines, written by people selling candles.

The difference is the job we give it. Our AI composes small groups of 4 to 6 people around something they actually want to do (a hike, a board game night, a study session) at a meeting point a person went and checked in advance, and then it gets out of the way. It does not choose the venue and it does not talk to you. It's a travel agent, not a destination. Nobody chats with it at 1 a.m.; nobody is meant to. If our AI does its job perfectly, you spend zero additional minutes with software and three more hours a week with humans who now know your name. The measure of a companion app's success is time spent inside it. The measure of ours is time spent outside it. Same technology, opposite bet, and we're stating ours openly so you can judge the argument above with our interest fully visible.

What's better than an AI friend?

Not a lecture, and not cold turkey. If an AI companion has been holding you up, it's been doing a real job, and you don't owe anyone its deletion. The move isn't subtraction. It's adding one human room, smaller than you think, with the pressure taken off.

Why does something that small work? Because friendship was never going to arrive through a braver personality; it arrives through repetition, the same faces accumulating shared hours until one of them is a friend. The full mechanics, and the numbers behind them, are in our loneliness statistics page. And if the reason you opened a companion app in the first place is that "make friends" feels like advice from another planet (if the honest sentence is I have no friends), we wrote a no-judgment starting point for exactly that, and it will not tell you to put yourself out there.

One more thing, said straight: if what you talk to the AI about at night is heavier than loneliness (if it's depression, or thoughts of hurting yourself), that deserves a human who does this for a living. In the US, call or text 988 any hour. "I have no one to talk to" is reason enough.

The companion in your pocket learned to make you feel heard, and that's not nothing; being heard matters so much that even software supplying it moves the needle. But you weren't built to be heard by something that can't miss you. Somewhere nearby there's a room of people who haven't met you yet, who could. The app was the shelter. It was never the home.

Sources: Julian De Freitas, Ahmet Kaan Uğuralp, Zeliha Uğuralp & Stefano Puntoni, "AI Companions Reduce Loneliness", Harvard Business School Working Paper 24-078, published in the Journal of Consumer Research; MIT Media Lab & OpenAI, "Early methods for studying affective use and emotional wellbeing in ChatGPT" (2025); Common Sense Media, "Nearly 3 in 4 Teens Have Used AI Companions" (2025); Jeffrey A. Hall, "How many hours does it take to make a friend?", Journal of Social and Personal Relationships (2018); U.S. Surgeon General, Our Epidemic of Loneliness and Isolation (2023).