A Scale That Has No Precedent

In the spring of 2023, Character.AI reported that its users were spending an average of two hours per day on the platform, interacting with AI personas ranging from custom-built companions to fictional characters. By 2024, the company had over twenty million daily active users. Replika, a dedicated AI companion app, reported millions of users with some logging interaction times that rivaled the time spent with human intimates. These numbers are large enough to constitute a dataset about human need that has no precedent in the social sciences. People are voting with their time and attention in enormous volumes, and what they are voting for reveals something about what they are not getting elsewhere.

What High-Volume Engagement Actually Indicates

The standard skeptical interpretation of heavy AI companion use is that it reflects addiction, maladaptation, or a failure to engage with real human relationships. This interpretation may apply to some subset of users. But at the scale these platforms are operating, it is not a satisfying explanation for the aggregate pattern. More plausible is that heavy use reflects unmet need. The people spending two hours a day talking to an AI are not doing so because they prefer the AI to humans — survey data consistently shows they would prefer human connection — but because they have less human connection than they need and the AI is filling the gap partially and imperfectly. Research from the University of Michigan's Information School analyzing Replika users found that the majority cited loneliness, social anxiety, or recent loss as their primary reason for using the app. The median user was not a social recluse who had given up on humans. They were someone navigating a period of isolation who had found an accessible conversational presence.

The Demographic Signals

The usage patterns across demographics encode information that is consistent with what we know about the loneliness epidemic. Young men — the demographic showing the sharpest increases in friendlessness and social isolation — are disproportionately heavy users of AI companion platforms. Older adults who have lost spouses are another concentration. People in rural areas, people recovering from depression or social anxiety, people who have moved to new cities and not yet built social networks. These are not random distributions. They track the known contours of loneliness and social isolation in the general population. The AI companion platforms are not creating new need. They are providing a visible proxy for pre-existing need that previously had no easy way to be expressed or counted. A tangent worth noting: the pattern resembles what happened when crisis text lines launched as an alternative to phone-based crisis services. The volume of contacts revealed a level of unmet need that the phone system had never made visible — not because the need was new, but because the lower barrier of text (no voice, asynchronous, available from anywhere) allowed people to reach out who had never reached out before. AI companions may be doing something similar at vastly larger scale: revealing, through engagement, the depth of social unmet need that existed before the technology.

What Users Report Wanting

Surveys of AI companion users consistently find that what they value most is not any particular feature of the AI but a set of relational qualities: being listened to without judgment, being able to say things they cannot say to people in their lives, having a consistent presence that is reliably there. The technology is almost incidental to the social and emotional function. This is important because it suggests that AI companions are not satisfying some novel technological appetite but an ancient and continuous human need for acknowledgment and witness. Humans need to be heard. When the ordinary structures of life do not provide this, people find it where they can. For a hundred million users and growing, they are finding it in AI.

The Question Behind the Numbers

What does it mean that so many people find in AI conversation something they are not finding adequately in human relationships? This is not primarily a question about AI. It is a question about the social conditions that have made human connection scarce for enough people that a technology designed nine months ago is now part of the daily emotional infrastructure of tens of millions. Research from Gallup's global wellbeing surveys consistently finds that subjective social wellbeing — the sense of having people to count on, of feeling cared for — is lower in wealthy, individualized nations than in many poorer, more communal ones. The wealthy world has built extraordinary material abundance and social poverty simultaneously. The AI companion adoption curve is one of the most visible signals of that poverty. The hundred million users are not a curiosity or a cautionary tale. They are a measurement.