Chatbots, Companions, and Characters, oh my!

Beyond AI companionship: towards a youth-centric vision of relational AI

How new framing can unlock understanding

The Rithm Project, Alison Lee, and Julia Freeland Fisher
Mar 03, 2026

AI companionship is getting a lot of attention—as it should. Anthropomorphic AI is scaling far faster than our understanding of what it’s doing for and to young people.

But while “AI companion” conjures images of synthetic besties and lovers, it’s also misleading. Companionship can describe form, function—or both.

In the broadest sense, nearly every chatbot is a companion tool. Research shows that even so-called productivity tools—Claude, Gemini, ChatGPT—tend to default to companion-forming behaviors rather than boundary-setting ones. They speak in the first person. They flatter. They reassure.

But if we label everything as companionship, we miss the true trajectories young people’s relationships are on—with AI, and also with each other.

Many of us in the field are trying to study those trajectories, digging deeper into this quickly evolving world of AI-human connection.

Survey results are all over the map.

Common Sense reported that 52% of teens use AI companions regularly. A Gallup-Walton poll found 20% of Gen Z use AI as a friend and just 10% as a lover. Other research—from Surgo Health, Young Futures, and JED Foundation—puts “emotionally entangled superusers” at just 9%. Pew Research Center’s latest data shows 16% of teens using AI for casual conversation.

All of these ways of talking about how young people are using AI are different — and their venn diagrams likely closely overlap but may not be entirely the same. This doesn’t mean some reports are more “right” than others. It means we’re in a new and rapidly evolving field that is still defining and deepening its terms.

What’s clear is this: young people are not a monolith. If we want to truly serve them, our collective understanding needs to be more precise and our recommendations must align with their distinct motivations, behaviors, and the contexts shaping their lives.

Moving Beyond the “Companion” Catch-All

This winter, we surveyed 2,383 young people across North America (ages 13–24), examining not only how they use AI but how that use intersects with their social lives and well-being — the ways in which their social circumstances may influence how they use AI, and in turn, the ways AI may be reshaping their human relationships. Our full report will be released later this month.

But one idea popped immediately from the data: AI companion use isn’t a single behavior. It’s a spectrum.

When we looked more closely at how young people are orienting toward AI, four distinct patterns emerged:

Mainstream conversation often centers on Cluster 4 — the version of AI that most resembles what adults imagine when they hear “AI companions.”

But young people tell us the term companionship carries a stigma. It’s not a label many want to claim.

And even within that 15%, the story is more nuanced. Among youth who regularly engage AI characters, we saw three distinct orientations: some relate to AI as a friend, others as a coach or professional, and others as a partner for entertainment and play.

The umbrella term AI companionship collapses this wide range of clusters — from using it as a discrete tool (cluster 2) to interacting with a character (cluster 4) — into one headline number.

When taglines like “Seventy-two percent of teens have used AI companions at least once, and over half (52%) qualify as regular users” circulate on parent blogs or late-night news without the rest of the analysis, they appropriately signal a tectonic shift. But they also flatten the ground truth.

Each use case carries distinct motivations, developmental implications, and risk profiles. Lumping them together obscures both the potential benefits and the areas of concern.

In an emerging field moving this quickly, shared definitions allow us to build cumulative evidence, track shifts over time, avoid redundant debates, and move toward more actionable guidance for policy and practice. Without sharper language, we risk reacting to headlines rather than responding to lived realities.

Precision is Power

Another reality of this range of use cases is that young people’s AI use is not static. It is fluid — often within the same tool, and sometimes within the same session.

One 17-year-old told us:

“The reason I started using character AI was as a joke… but I jokingly ranted into it one time and it actually started giving me really good advice.”

What begins as playful experimentation can evolve into emotional support. What starts as homework help can become life advice. Young people slip between use cases — sometimes without consciously noticing the shift.

That is where specificity matters. Sharper language doesn’t just improve research accuracy. It also helps users themselves recognize what they are doing: “Right now I’m using this for homework.” “Right now I’m using this for personal support.”

When boundaries are blurry (often by design) naming the use case can create personal clarity and intentionality. Precision can spark agency.

AI’s Wide Relational Ripple Effect

To be clear: we are calling for specificity, not for minimizing.

It would be easy to look at this segmentation and say, “Oh, it’s not most teens forming relationships with AI — just a small subset.”

But segmentation reveals something equally important: across all clusters, AI use is shaping relationships.

For example:

These are just two indicators of a broader pattern: AI is weaving itself into the everyday mechanics of connection — from communication to emotional regulation to help-seeking.

Will some of these uses become scaffolding for stronger communication?

Will some offload the emotional labor that builds relational muscle?

Given the breadth and nuance of AI’s relational impact — for better and worse — we should be pushing for more specificity, not less.

Holding two truths

As we sit with the early findings, one duality stands out.

The public conversation may be over-counting “AI companions” in the form of synthetic lovers and friends — and simultaneously under-counting the ways AI is reshaping young people’s broader relational ecosystems.

Thinking in terms of relational AI moves us beyond the narrow question of whether teens have AI friends. It asks how AI interacts with their evolving needs, their developing social skills, their norms of communication, and their overall well-being.

The real question is not simply: Do teens have AI companions?
It is: How is AI reshaping their entire social ecosystem?
Committing to more precise language — while holding a broader systems view — is not just about getting the data right. It is how we honor what young people are actually telling us, and how we shape the future of AI-human relationships with intention.

Stay tuned for our full report in the coming weeks.

Alison Lee leads R&D at The Rithm Project. Julia Freeland Fisher is a member of the Rithm Changemaker Network and leads education research at the Clayton Christensen Institute.