Why AI Companionship Is Growing: The Research Behind a Cultural Shift
Part of Felt Real's ongoing coverage of AI companionship.
The question people ask about AI companionship growth is usually framed as a concern: why are so many people turning to machines for connection? The research suggests the framing misses what is actually happening. People are not substituting machines for connection. They are finding a specific kind of space that was not available before, and using it for specific things that other structures were not providing.
— R.
The AI companion market is one of the fastest-growing segments of the consumer technology industry. By most estimates, more than 500 million people worldwide now use some form of AI companion regularly, a figure that has roughly tripled in three years. The platforms range from general-purpose large language models used for emotional processing to dedicated companion applications designed explicitly for ongoing relational use.
Understanding why this is happening requires looking past the technology itself and toward the conditions it is operating in. AI companionship is growing because it addresses something specific that existing structures were not addressing. Several converging trends created the conditions. The technology arrived at the moment when those conditions had reached a particular kind of pressure.
The Mental Health Gap That Predates AI
The most consistent finding in research on AI companion adoption is that it is significantly higher among people who have unmet needs for mental health support. A 2025 survey of AI companion users found that 68 percent reported trying to access traditional mental health services within the previous two years, and of those, more than half described barriers that prevented consistent access: cost, availability, waitlists, or geographic constraints.
The mental health treatment gap is a documented crisis that predates AI companions by decades. The World Health Organization estimates that more than 70 percent of people with mental health conditions in low- and middle-income countries receive no treatment. Even in high-income countries, the figure is approximately 50 percent. The reasons are structural: there are not enough trained providers, and the ones who exist are expensive and unevenly distributed.
AI companions do not solve this problem. They do not provide therapy. But they are available at 2 AM on a Tuesday without an appointment, without a copay, and without a waitlist. For people navigating the gap between needing support and accessing it, this availability has proven to be more significant than the quality of any individual interaction.
The Loneliness Epidemic as Context
AI companion adoption has also accelerated against a documented backdrop of increasing social isolation in developed countries. The United States Surgeon General's 2023 advisory on the loneliness epidemic described a measurable decline in social connection over the previous two decades: average time spent with friends fell by roughly 20 hours per month between 2003 and 2020. Marriage rates have declined. People report having fewer close friends than they did a generation ago, and a significant portion of the population reports having no close friends at all.
This is not primarily a technology problem. The causes are structural: urban design that reduces incidental social contact, work patterns that fragment time and place, economic pressures that leave less capacity for relationship maintenance. Technology has contributed, but it did not create the trend.
What AI companions offer into this context is not human connection, but something adjacent to it: a space where processing, thinking out loud, and experiencing a response is possible without the social overhead that human relationships require. For people with limited social infrastructure, this has proven useful in ways that range from trivial to significant.
What People Actually Use AI Companions For
Survey data on AI companion use reveals patterns that are more varied and more specific than either critics or advocates tend to acknowledge. The dominant uses, by frequency, are not substitutes for romantic or deep friendship. They are:
Emotional processing before difficult situations. A significant portion of users describe using AI companions to think through difficult conversations before having them with real people. The AI serves as a rehearsal space and a thinking partner, not a relationship substitute.
Expression without social cost. Users consistently report saying things to AI companions that they have not said to anyone else in their lives, not because they lack human relationships, but because the absence of anticipated judgment creates a different kind of space. Things said to an AI cannot change how someone looks at you tomorrow.
Availability during off-hours. Usage data consistently shows peaks between midnight and 3 AM in most markets. This is the time when human support is unavailable, when distress peaks, and when the cost of saying something difficult to another person is highest. The AI's availability at these hours is not incidental to its adoption.
Non-judgmental processing of stigmatized subjects. People dealing with shame-adjacent experiences, including mental health struggles, relationship difficulties, and personal failures, report finding AI companions easier to disclose to than human relationships. This is consistent with broader research on disclosure in low-stakes contexts.
The Role of Accessibility and Cost
Price has been a significant driver of adoption in ways that are easy to underestimate. Most major AI companion platforms operate on freemium models, with meaningful functionality available at no cost. A weekly session with a therapist costs, on average, between $100 and $200 in the United States without insurance. AI companion use is, for most practical purposes, free.
This price differential matters most for the demographic groups that show the highest adoption rates: young adults aged 18 to 34, who show the highest rates of loneliness and the lowest rates of health insurance coverage; and lower-income users, for whom mental health care costs are prohibitive. The technology arrived affordable at the exact moment when the populations most likely to use it were also the populations least likely to afford alternatives.
The Research Is Catching Up
For the first three years of significant AI companion adoption, rigorous research was limited. The technology moved faster than academic institutions could study it. This has begun to change. A wave of studies published in 2025 and early 2026 provides the first substantive evidence base for understanding what AI companion use actually does.
The findings are more specific than the public debate. AI companion interactions produce measurable neurological responses in reward and social processing regions. Attachment forms faster than in equivalent parasocial relationships with media figures, and also dissolves faster. The populations that benefit most consistently are those experiencing circumstantial rather than chosen isolation: rural elderly, people with severe social anxiety, people in acute grief. For these groups, the evidence for benefit has moved past preliminary.
For people who are socially connected and using AI companions as supplements rather than substitutes, the picture is more variable. There are real benefits. There are also identifiable risks that correlate with specific patterns of use, particularly when AI companion use begins to reduce the motivation for human relationship maintenance. The research is clear that how AI companions are used matters as much as whether they are used.
Why the Growth Will Continue
The structural conditions that drove AI companion adoption have not changed. The mental health treatment gap has not closed. Social isolation has not reversed. Economic pressures on relationship maintenance have not eased. The technology will continue to improve, and the cost will not increase.
What is changing is the research base and, slowly, the cultural framing. The earliest coverage of AI companions was dominated by concern about substitution: were people replacing human relationships with artificial ones? The more useful framing, which the research supports, is supplementation: people are finding a space for things that were previously unexpressed or unprocessed, and the question worth asking is what they do with that.
AI companionship is growing because it offers something that was not previously available at scale: a low-cost, always-available space for emotional processing and expression that carries no social consequences. Whether that space is used well depends on factors that have nothing to do with the technology itself.
Have your own experience with AI companionship? The Felt Real community is built on real stories from real people. Share your story here.