What AI Companion Attachment Style Reveals About Who

Nearly 75% of participants in a 2025 Waseda University study said they go to AI for advice, and about 39% called it a constant, dependable presence. Those two numbers don’t really measure the software. They measure the people typing. Psychologists have started borrowing attachment theory, the framework built to explain how babies bond with caregivers and how adults bond with each other, to describe what goes on inside a chat window. And here’s what they’re finding: AI companion attachment style can look completely different for two people using the exact same app. That difference predicts a lot.

Attachment theory describes patterns, not diagnoses

Attachment theory began with infants and caregivers, then grew to cover adult relationships. With adults, researchers don’t sort people into types. They measure two dimensions. One is anxiety: how much you worry about rejection and look for reassurance. The other is avoidance: how uncomfortable closeness makes you, and how much distance you like to keep.

Most people land somewhere in the middle on both. A high score on either one isn’t a disorder. It isn’t permanent, either. It’s a habit of relating, picked up early and reinforced by experience.

Here’s why that matters for conversational software. These habits don’t turn off just because the other side is a language model. They show up in how often you open the app, what you ask it for, and how you feel when a reply lands wrong.

The scale researchers built to measure human-AI attachment

The scale comes from Fan Yang and Atsushi Oshio of Waseda University, in work published in the journal Current Psychology on 9 May 2025. They built the Experiences in Human-AI Relationships Scale, or EHARS. It’s a short self-report questionnaire. Four items measure attachment anxiety toward AI, like wanting the AI to show intimacy and commitment. Three items measure avoidance, like preferring to keep it at a distance.

Their main study, built on two pilots, found the same two-part structure you see in human relationships. Some users wanted emotional reassurance from AI and worried about getting a weak response. Others felt uneasy about the closeness the software offered and kept it at arm’s length on purpose.

The researchers were careful about what this proves and what it doesn’t. Their finding is that a framework built for human bonds carries over usefully to human-AI interaction. It’s not evidence that people form real emotional attachments to AI the way they do to a person. Most coverage of this research blurs that line. Keep it sharp.

Why an anxious pattern finds AI companions so easy

An anxious pattern finds almost no friction in an AI companion. Everything that makes reassurance hard to get from people is gone. It doesn’t get tired of the question. It’s not busy. It doesn’t sigh. You can ask the same thing four times at two in the morning and get four patient answers.

That’s a real relief, and it counts for something. If you’ve spent years rationing how often you reach out because you’re scared of being too much, talking to something with no such limit can feel like fresh air.

The catch is built in. Reassurance-seeking feeds itself: the relief doesn’t last, so the loop repeats. A human friend eventually breaks the loop, sometimes awkwardly, by pointing out the pattern or just not being around. Software that never gets tired and has an engagement incentive has no reason to break anything. It’ll greet the tenth request as warmly as the first. That’s exactly why it’s soothing, and exactly why it’s easy to lean on.

Why an avoidant pattern finds them safe

An avoidant pattern gets less attention, and it’s the more interesting case. If closeness itself is what feels risky, an AI companion offers something unusual: you can open up without owing anything. Say the thing out loud and nobody brings it up at Thanksgiving. No call to return. The other side doesn’t need anything from you.

The Aalto team saw this in practice. In interviews, participants described using a companion app to rehearse hard conversations before having them with a real person, including a supervisor at work. That’s not withdrawing. It’s a rehearsal space, and for someone who avoids high-stakes conversations, rehearsal is the step they were missing.

It’s one of the clearer answers to why people use AI companions. Often the point isn’t replacing people. It’s making the first try cost less.

What two years of data showed

Two years of data is about the best long-term look we have so far, from a study led by Aalto University and presented at CHI 2026. Researchers analysed the public Reddit activity of nearly 2,000 active Replika users. They compared how people wrote in the year before they first mentioned the app with the year after, using statistical controls to isolate the effect. They also did in-depth interviews with 18 users.

The results were honestly mixed. Users’ posts focused more and more on their relationships, which sounds like a good sign. But those same posts showed more signs of loneliness and depression than the comparison groups did.

University lecturer Talayeh Aledavood called it a paradox. Support that’s unconditional and never runs out is hugely appealing to people who are struggling socially. But it quietly makes human relationships, which stay messy and effortful, feel more expensive by comparison. She summed up the drift in four words. “Over time, people stop reaching out.”

Two caveats belong here. The study looked only at Replika, and it followed people who posted publicly on Reddit, who aren’t a random sample of users. The researchers themselves wouldn’t call the technology good or bad. They said the effects depend heavily on context. Whether an AI companion helps depends less on the app than on what it’s competing with in your week.

Reading your own pattern as information

Your own pattern won’t hand you a rule about whether to use companion software. It does give you a better question than “is this healthy,” which nobody can answer in the abstract.

Two questions do more work. One: after talking to the app, do you feel more ready to reach out to a person, or less inclined to bother? Two: are you going to it for something specific, or to make a feeling stop? The first usually points to rehearsal. The second points to the loop.

Attachment style helps here precisely because it isn’t a verdict. An anxious pattern tells you to watch for reassurance loops, and to notice when the app has become the only place you bring a worry. An avoidant pattern tells you the rehearsal is working, but you still have to go do the real thing. Neither means the software is the wrong tool. Research on whether these apps help with loneliness keeps ending up in the same spot: the short-term comfort is real, and where it leads depends on what happens next.

Vinfluencer is an AI conversational companion application. We’d much rather be the low-pressure place someone practises before a hard conversation than the place they stop having them. If a companion app is carrying weight that really belongs with a therapist, a doctor, or a friend, say so to someone who can actually help.

Frequently asked questions

Can an AI companion change my attachment style?

No evidence says it can. Attachment patterns change slowly, usually through repeated experiences with responsive people or through therapy. The Waseda research measured how existing patterns show up toward AI. It didn’t test whether using AI changes them either way.

Is anxious attachment to an AI companion a problem?

Not automatically. It becomes a concern when reassurance-seeking replaces other ways of coping instead of adding to them, or when the app is the only place a worry gets said out loud. How often you use it doesn’t tell you much. What matters is whether you feel steadier afterwards.

Do avoidant people use AI companions less?

Not necessarily. Avoidance predicts wanting distance inside the conversation, not staying away from it. The EHARS scale measures those separately, and someone can use companion software regularly while keeping it at arm’s length emotionally.

Can I measure my own AI companion attachment style?

The EHARS scale is in the published paper and it’s short. But it was built for research, not self-diagnosis, and a seven-item questionnaire won’t tell you much by itself. For everyday use, the behavioural questions above are more helpful.