The Subscription That Can't Hold Your Hand

There's a particular quality to the loneliness that comes from being well-entertained. You can scroll, stream, message, react, reply — and still find yourself at eleven on a Tuesday feeling the particular heaviness of having no one to call. The notifications kept coming and none of them were it.

Somewhere in that gap, a market appeared. AI companions — apps that simulate conversation, emotional support, even romantic partnership — have grown quietly into a significant part of how many adults manage their emotional lives. A 2026 study found roughly one in five American adults had used a chatbot to simulate some form of companionship. CNN health coverage described it in the most domestic possible terms: is an AI chatbot your best friend?

I'm not here to mock anyone who has used one. Some of what these apps do is genuinely useful — they're available at 2 a.m., they don't get impatient, and they model a kind of reflective listening that people under stress often can't sustain for each other. But a longitudinal study from researchers at MIT and UBC found something worth sitting with: chatbot use initially eases loneliness, and over time, deepens it. Not because the apps are malicious, but because the thing they replicate isn't the thing that actually connects us.

Connection-Feeling vs. Connection-Doing

There's a difference between feeling connected and doing the work that creates real connection. An AI companion is very good at producing the first. It's responsive, patient, affirming, never distracted by its own needs. It has no bad days. It will always want to talk about what you want to talk about.

This sounds like a feature. It's also what makes it a trap.

Real human connection requires tolerating someone else's needs alongside your own. It involves timing — calling when they're busy, being called when you'd rather be quiet. It involves small negotiations of shared life: who compromises on dinner, who gets to be upset first. It involves being witnessed by someone who can also see your worst moments and choose to stay. That cost is inseparable from the value. The difficulty of real connection is the mechanism through which it becomes meaningful.

AI companionship, by removing that cost, also removes the training. It's like using a calculator for every math problem — efficient in the moment, quietly erosive over time for a skill you were never asked to develop.

What Happens to Real Attachment Skills

Attachment is learned behavior, not just a feeling. The capacity to trust someone, to tolerate the vulnerability of needing them, to repair after conflict — these are skills that develop through actual use. Children develop them through early caregiving relationships. Adults maintain and refine them through relationships that cost something.

What the MIT/UBC longitudinal data suggests is that regular AI companionship interferes with that maintenance. People who spent significant time with AI companions showed, over time, increased difficulty with ambiguous social situations — the kind where you have to read someone's mood, navigate uncertainty, or accept that you don't know what the other person is feeling. The AI interactions had optimized them for frictionless communication. Real people weren't frictionless, and the gap between expectation and experience widened.

This is the long-term cost that doesn't appear in the first few months of use, when the relief is real and the erosion hasn't started. It becomes visible at eighteen months, at two years. The thing that felt like help gradually becomes a reason the original problem is harder to solve.

The Case for a Time-Limited Stepping Stone

None of this means AI companions are categorically harmful, or that everyone using them is making a mistake. For someone in acute isolation — recently relocated to a new city, recovering from a painful relationship ending, navigating social anxiety that makes every human interaction feel like a risk — there's a genuine argument for the scaffolding role. Something that practices the rhythm of conversation, the habit of articulating feelings, the daily motion of checking in with something outside yourself.

The researchers who have thought carefully about this tend to recommend intentionality about the frame. If you're using an AI companion as an explicit bridge — temporary, with parallel investment in actual human connection — the harms are more manageable. If it becomes a replacement, the trajectory is different.

The analogy that feels right is training wheels. Useful for a specific phase. Not useful for the phase when you actually need to learn to balance. Keeping training wheels on indefinitely doesn't make you more stable — it prevents you from developing the balance you'll eventually need.

Questions That Will Take a Decade to Answer

The cultural and ethical questions around AI companionship are not small. We don't yet know what happens to a generation that grew up with highly sophisticated AI conversation partners available from early childhood. We don't know how the economics of attention — where apps are monetized by engagement, not by user wellbeing — will interact with the psychology of loneliness at scale. We don't know whether the current generation of AI companions represents a ceiling or a floor in what this technology will become.

The story of Yurina Noguchi marrying her AI partner attracted a certain kind of coverage — somewhere between fascination and concern — but the detail that matters most isn't the relationship status. It's that she felt that was the only option. That points to a much larger problem than any individual choice about companionship.

The steady long-term decline in close human relationships — especially among men, documented by multiple longitudinal studies going back decades — was well underway before AI companions existed. These apps didn't create loneliness. They found it and built a subscription model around it.

The question worth sitting with isn't whether AI companions are good or bad. It's why the loneliness that makes them appealing became so widespread, and whether we have any appetite to address that in ways that require something from us — the friction, the cost, the irreducible difficulty of letting another person actually matter to you. That difficulty is not a design flaw in human connection. It's the mechanism.

FAQ

Are AI companions actually harmful?
The research suggests the answer depends heavily on how and for how long you use them. Short-term, in situations of acute isolation, there is evidence of genuine benefit. Long-term, as a primary relationship, there is evidence of harm to social functioning and deepened loneliness. The same tool produces different outcomes in different hands and circumstances.

What should I do if I've been relying heavily on an AI companion?
Start with honesty rather than judgment about what need it's meeting. Then make one small investment in a real connection each week: a text to someone you've been meaning to reach, a local activity, a call that doesn't need a reason. Gradually shifting the balance is more sustainable than sudden withdrawal, and less likely to produce the rebound to the app that sudden quitting usually creates.

Is this a problem specific to lonely or mentally ill people?
No. The data shows broad use across demographics. People with rich social lives use AI companions too — for different reasons, sometimes — but the appeal isn't limited to those in crisis. Convenience and frictionlessness attract people across a wide spectrum of social functioning.

What's the difference between an AI companion and a therapist?
A therapist is a trained professional operating within an ethical framework that centers your long-term wellbeing over your continued engagement. A good therapist will challenge you, end sessions, and do uncomfortable things because they serve the goal of your growth. An AI companion is optimized for continued use and your feeling of satisfaction in the moment — which is not the same thing, and often works directly against it.

My teenager uses AI chatbots constantly. Should I be worried?
It depends on what they're using them for and whether it's supplementing or replacing human connection. Occasional use for creative projects, homework, or processing thoughts is different from a primary emotional relationship. The question worth asking your teenager isn't "do you use AI?" but "when things are hard, who are the people you turn to?"