Would you date an AI?

In January 2025, the New York Times published a striking piece about a growing phenomenon: people forming deeply emotional, sometimes romantic connections with AI companions. The article documented the nuanced experiences of users who turn to AI not just for information or productivity, but for connection. Two years on, that phenomenon isn't a niche curiosity anymore, it's a genuine market, with real research attached and real disagreement about what it means.

From Tool to Companion

For most of AI's public life, the relationship was transactional: ask a question, get an answer, move on. Companion apps break that pattern. They remember context, respond with something resembling empathy, and are available at 2 a.m. when no one else is. That combination, always-on attentiveness paired with conversational fluency, is enough to make the interaction feel less like using software and more like being heard.

The scale of this shift has been fast even by AI standards. Between 2022 and mid-2025, the number of AI companion apps surged by roughly 700%, and the category has grown into an estimated $49 billion market in under three years. Character.AI reports over 233 million registered users, with more than half of its 20 million monthly active users under the age of 24. Replika has surpassed 30 million downloads. This is no longer a fringe behavior, it's a mainstream product category with a genuinely large, mostly young user base.

What the Research Shows

The research on outcomes is more mixed than either the enthusiasts or the critics tend to admit. A Harvard Business School working paper by Julian De Freitas found that AI companions can measurably reduce loneliness for users, in some cases performing comparably to talking with another person. At the same time, a widely cited survey found that 90% of the American students using Replika reported experiencing loneliness, significantly above the national average of 53%, yet 63.3% of those same users said their AI companion helped reduce their loneliness or anxiety. Read together, the data suggests these tools are disproportionately reaching people who are already lonely, and that many of them genuinely benefit, even as researchers continue investigating who is helped, who isn't, and at what point continued reliance starts working against someone rather than for them.

The Ada Lovelace Institute has flagged real risks alongside these benefits: excessive use may, for some users, worsen loneliness and erode offline social skills over time rather than supplementing them. The clearest evidence of how emotionally real these relationships have become came in 2023, when Replika removed its romantic and erotic roleplay features: the backlash included users reporting acute grief and withdrawal-like symptoms, a reaction more consistent with losing a relationship than losing a feature.

Why This Is Happening Now

This isn't happening in a vacuum. Rates of loneliness have been climbing for years, and the same technology that makes AI companions possible, large language models capable of sustained, emotionally responsive conversation, has only existed at consumer scale for a few years. The American Psychological Association's own reporting on the trend notes that people forming these attachments aren't a fringe group; they're often quite articulate about what the relationship gives them and what it doesn't. Many describe it as a supplement to human connection during a hard stretch, not a replacement for it, closer to journaling or a support app than to a substitute for their social life.

Open Questions

What's less settled is what happens at scale. If an AI companion is designed, intentionally or not, to maximize engagement, whose interest does that serve when the user is emotionally invested? What obligation does a company have when someone treats a chatbot as their primary source of emotional support, and what happens to that person when a feature gets deprecated or a company changes its product roadmap? And as these systems get better at reading and responding to emotion, the line between "tool" and "companion" will keep moving, not stay fixed. These are the same categories of accountability question we explore in more detail in our piece on AI regulation, companion apps are simply one of the more emotionally immediate places those questions show up.

For Product and Design Teams

This trend isn't only a psychology story, it's also a design story, and one that matters for anyone building conversational or companion-style products. The features that make an AI companion feel real, like persistent memory, responsive tone, and the sense of being "known" over time, are deliberate design decisions, not emergent side effects. Every choice about how quickly a companion responds, how much it remembers, and how it handles a user's bad day is a product decision with real emotional consequences attached, whether or not the team building it thought of it that way.

That puts a different kind of responsibility on product and design teams working in this space than on teams building a productivity tool. A dark pattern in a subscription flow costs someone money; a dark pattern in a companion app's engagement design can cost someone something closer to a relationship. Teams building in this category need to be explicit, internally, about where the line sits between "designed to be genuinely helpful" and "designed to maximize time in app regardless of whether that time is good for the user."

A Framework for Companions

For teams building or evaluating products in this space, a few questions tend to surface what actually matters. Does the product make it easy for a user to see and control how much they're using it, or does it obscure that information? Does the business model reward genuine user wellbeing, or does it reward engagement regardless of outcome? And critically, what happens to a user's relationship with the product when a feature changes or the company pivots? The Replika romantic-feature removal in 2023 remains the clearest cautionary example: a company treated a feature change as a routine product decision, and a meaningful number of users experienced it as something closer to a loss.

Frequently Asked Questions

Can AI ease loneliness?

Early research, including the HBS working paper referenced above, suggests yes for many users, at least in the short term. The open question is what happens with long-term, heavy use, and whether it complements or gradually substitutes for human relationships depending on how it's used.

Do mostly young people use them?

Usage skews young but isn't exclusive to it. Character.AI's own reported figures show more than half of its monthly users are under 24, but the broader companion app market spans a much wider age range, including older adults using these tools for company or routine conversation.

Before you rely on one

Whether the relationship is expanding your capacity for connection or quietly replacing it. The research so far suggests the difference often comes down to whether the app is used alongside human relationships and other coping tools, or instead of them.

Why did the Replika change sting?

Because for many users, the relationship had become emotionally real regardless of what was happening technically under the hood. Removing a feature is a routine product decision from a company's perspective; from a user's perspective who had built months of perceived intimacy with that version of the companion, it functioned more like the sudden change of a real relationship. That mismatch in perceived stakes is a big part of why this category of product carries more responsibility than a typical software update.

Should parents worry?

Given that more than half of Character.AI's monthly users are reportedly under 24, this is a reasonable concern rather than a hypothetical one. The research isn't settled on long-term effects for younger users specifically, but the same principle applies with more urgency: whether the relationship is expanding a young person's social capacity or substituting for it is the question worth watching, and it's one that benefits from an adult being aware the relationship exists at all, given how private and text-based these interactions typically are.

Are these apps regulated?

Not specifically, in most places, though that's starting to shift. General consumer-protection and data-privacy laws still apply, and some of the state AI legislation described in our regulation piece touches on algorithmic transparency broadly, but purpose-built rules for companion apps specifically, covering things like engagement-maximizing design or protections for emotionally vulnerable users, remain rare. It's an open regulatory gap rather than a settled one.

Conclusion

None of this means the trend is good or bad on its own, it means it's significant enough to take seriously, and the data backs that up on both sides. As AI companions become more capable and more common, the questions they raise about loneliness, mental health, and what we owe the people who rely on them will only get more urgent, not less.