Character.AI has lost roughly 8 million monthly active users since its mid-2024 peak of about 28 million, dropping to around 20 million in 2026. That sounds like a platform in decline. But the users who stayed did not pull back: they now spend an average of about 75 minutes a day on the app, with sessions running 25 to 45 minutes at a stretch. Compare that to ChatGPT, where the average session lasts somewhere between 7 and 12 minutes. Fewer people are using AI companion apps, and the ones who remain are using them far more intensely than anyone uses a general-purpose chatbot. That gap is the real story, and it says more about emotional attachment to AI companion apps than any single survey does.
The numbers behind AI companion app usage
Character.AI is not the only player showing this pattern. Replika, one of the oldest AI companion apps, passed 40 million users by 2025 according to usage data compiled by industry trackers. A 2024 study published in npj Mental Health Research surveyed 1,006 student users of Replika and found the app was being described, in the users own words, as functioning simultaneously as a friend, a therapist, and a confidant. That is not marketing copy. That is how people who actually use these apps described the experience when researchers asked them directly.
The broader market reflects the same pull. Multiple market research firms tracking the AI companion category project double-digit annual growth through the early 2030s, even as individual apps like Character.AI shed millions of casual users. The category is not shrinking. It is consolidating around people who use it heavily rather than casually.
Why emotional attachment to AI companion apps happens at all
Researchers who study this point to a fairly simple mechanism. AI companion apps remember details from past conversations, respond instantly at any hour, and never get tired, distracted, or annoyed. A 2026 study on long-term AI companion use found that frequency of use was strongly linked to attachment intensity, and that attachment showed the same four markers psychologists use to describe attachment in human relationships: users sought proximity to the app, felt distress when separated from it, and treated it as both a safe haven and a secure base.
None of that requires the AI to be sophisticated. It requires it to be consistent. A study surveying reasons people keep coming back to companion apps found that enjoyment and curiosity ranked highest, followed by passing time, reducing stress, and specifically seeking companionship. Most people are not chasing a romantic fantasy. They are chasing a low-effort way to feel heard.
The research is genuinely split, not settled
Here is where it gets less tidy than most coverage suggests. Some peer-reviewed studies found that sustained use of AI companions reduced loneliness and social anxiety over weeks of interaction. A different body of research finds the opposite: heavier use correlated with increased emotional dependence and, in some users, reduced engagement with human relationships. The American Psychological Association Monitor summarized 2026 research on the topic by stating plainly that evidence on the psychological and social impacts of AI companions is emerging but remains mixed.
One of the more useful 2026 findings tried to explain why results diverge so much. It found that attachment to an AI companion was linked to lower loneliness and higher wellbeing, but only when that attachment ran alongside continued real-world social engagement, not as a replacement for it. In other words, the app seems to help the people who already have some human connection to lean on, and the picture gets murkier for people using it as their only outlet.
Who is actually using these apps
Demographic data on Replika users complicates the stereotype that companion apps are mostly a novelty for lonely teenagers. Among users who disclosed their relationship status, the majority reported already being in a relationship, married or otherwise, rather than single. A separate 2025 global survey found that 70.5% of respondents said they had used AI for emotional support in some form, a number that extends well beyond dedicated companion apps into general-purpose chatbots being used the same way.
- Character.AI: around 20 million monthly active users in 2026, down from 28 million at its 2024 peak, but with average daily use of roughly 75 minutes among remaining users.
- Replika: user base exceeding 40 million as of 2025, with roughly half of engaged users reporting some form of romantic attachment to their AI companion.
- ChatGPT, for comparison: average session length of roughly 7 to 12 minutes, reflecting task-based use rather than sustained companionship.
What this means if you are curious about trying one
The honest takeaway from the research is not that AI companion apps are good or bad. It is that the effect seems to depend heavily on what role the app plays in your life. Used as one input among several, alongside actual friends, family, or a therapist, the data leans toward these apps providing genuine, measurable comfort. Used as a total substitute for human contact, the same data leans toward reinforcing isolation rather than resolving it. The app itself cannot tell the difference. Only the person using it can.
The takeaway
What makes this trend worth watching is not that people are attached to chatbots. That was predictable the moment these apps got good at remembering things. What is more interesting is that the market is quietly sorting itself into two very different products wearing the same interface: a casual novelty that millions try and drop, and a daily emotional habit that a smaller, more committed group has built real routines around. Character.AI losing a third of its user base while its remaining users spend more time on it than ever is not a contradiction. It is what happens when a product stops being for everyone and starts being essential for a much smaller group. That is a harder trend to regulate with a single opinion, and a much more interesting one to actually watch play out over the next few years.
