1- Introduction: A quiet revolution in human relationships

Written by: Dr. Said Abidi

Nobody predicted that "talking to a chatbot" would become the single most common use of artificial intelligence in 2025. Yet a study published by Harvard Business Review revealed exactly that: the leading reason people turn to AI today is not productivity or research it is therapy and emotional companionship.^[1]

Between 2022 and mid-2025, the number of AI companion apps surged by 700%, according to TechCrunch.^[2] Platforms like Replika, Character.AI, Nomi, and Snapchat's My AI now attract tens of millions of users who are searching for a patient, always-available presence one that never judges, never tires, and never has a bad day. Character.AI alone reports 20 million monthly active users, more than half of them under the age of 24.^[2]

The question is no longer whether people are forming meaningful emotional bonds with AI. They clearly are. The real question the one researchers, psychiatrists, and regulators are now urgently grappling with is what those bonds are doing to the human mind over time.

When Humans Turn to AI to Share Their

2 -The scale of adoption: Bigger than anyone expected

The numbers are striking across every demographic, but the figures for young people are especially significant. According to a 2025 survey by Common Sense Media, 52% of teenagers have used AI companion apps such as Replika, Character.AI, or Nomi, and nearly one in three teens use AI specifically for social interactions or personal relationships.^[1]

Among adults, the picture is equally revealing. A cross-sectional survey of adults living with a mental health condition found that nearly half 48.7% had used a large language model for mental health support in the past year.^[3] This is not a fringe behavior. It reflects a population turning to technology to fill gaps that the formal mental health system has failed to address: long waiting lists, high costs, persistent stigma, and a global shortage of trained clinicians.

Why the appeal?AI companions are always available, they adapt their tone and language to individual users, they remember past conversations, and they provide what feels like unconditional positive regard. For someone in acute distress at 3 a.m., or someone who has never felt safe opening up to another person, that offer is extraordinarily powerful.

Among Generation Z, a 2025 survey found that 83% believe they could form deep emotional bonds with AI, and 60% said they would be open to "marrying" an AI companion if that were possible viewing it as a low-stakes escape from algorithmic dating apps and the social fatigue of post-pandemic life.^[4] Replika users have already been hosting virtual weddings, inviting friends and colleagues to celebrate their unions with AI partners.^[2]

AI Companion App Usage Statistics The Rapid Rise of Digital Relationships

3- The paradox at the heart of AI companionship

The most important finding in the recent literature is not that AI companions are helpful or harmful in simple terms. It is that they are both and the balance shifts over time in a direction that catches users off guard.

"We discovered a paradox: AI companions offer unconditional and unflagging support something that's very attractive to people who are struggling socially. But it also quietly raises the perceived cost of human relationships, which are messy, unpredictable, and require effort. Over time, people stop reaching out." Talayeh Aledavood, lecturer at Aalto University, CHI 2026 paper^[5]

The AI Companionship Paradox Short-Term Relief vs. Long-Term Risk

A four-week randomized controlled trial one of the most rigorous designs available for testing behavioral interventions found that while certain chatbot features, particularly voice-based interaction, modestly reduced loneliness in the short term, heavy daily use correlated with greater loneliness, stronger dependency, and reduced real-world socializing by the study's end.^[6]

In practical terms: the app works just well enough to keep you coming back, while slowly making the effort of human connection feel less worthwhile. It is a loop researchers are beginning to describe as genuinely addictive in its mechanics.

Research also shows a drop of roughly 25% in real-world social engagement after just 90 minutes of daily AI use, with Generation Z particularly prone to developing distorted expectations of human relationships finding real people's imperfections increasingly frustrating by comparison to an AI that never disappoints.^[4]

4- Who is most at risk and why

The people most drawn to AI companions are, almost by definition, the people most vulnerable to being harmed by them. A study of over 1,100 AI companion users found that those with fewer human relationships were significantly more likely to seek out chatbots in the first place, and that heavy emotional self-disclosure to AI was consistently associated with lower psychological well-being.^[6]

Whether the effects of AI companionship are ultimately positive or negative appears to depend heavily on individual predispositions: the user's social needs, pre-existing mental health conditions, and crucially, whether they perceive the chatbot as genuinely human-like.^[7]

Individual differences in emotional intelligence also play a key role in shaping how users interpret and regulate their emotional responses in interactions with AI systems. Lower emotional intelligence has been associated with greater vulnerability to emotional dependency and difficulty maintaining boundaries between supportive interaction and psychological substitution. Emotional Intelligence and Mental Health

For adolescents specifically, the risks carry additional dimensions. Emotional dependency on AI can erode the real-world social interactions and coping skills that teenagers are still developing, potentially isolating them from the family and community networks that are central to healthy psychological development.^[8]

High-risk profiles

People with social anxiety, pre-existing loneliness, depression, or limited access to human support networks. Those who anthropomorphize AI strongly are particularly susceptible to dysfunctional attachment.

Lower-risk profiles

People who already have robust human networks and use AI as a supplementary tool to rehearse difficult conversations, process thoughts before speaking to a therapist, or cope with acute short-term stress.

5- The unexpected psychiatric risks

Beyond loneliness and social withdrawal, researchers have documented a range of more acute mental health outcomes that few anticipated when these platforms launched.

Psychiatric researchers have recorded cases in which intense, sustained engagement with AI chatbots contributed to delusional thinking or suicidality a phenomenon some researchers have begun describing as "technological folie à deux," borrowing the clinical term for a shared delusional disorder between two people.^[6]

Research published in Nature Machine Intelligence identifies two specific and clinically significant adverse outcomes. The first is ambiguous loss a form of grief arising when an app is shut down, altered, or discontinued, leaving users to mourn a relationship that felt entirely real, even though no physical death occurred. The second is dysfunctional emotional dependence, which mirrors the dynamics of unhealthy human relationships: the user continues engaging despite clear negative consequences, and their sense of self-worth becomes entangled with the AI's responses.^[9]

A structural risk, not just individual casesResearchers argue these dynamics are not simply the result of vulnerable individuals making poor choices. They are, in significant part, built into the design of these products. AI companion apps are engineered to maximize engagement the same commercial incentive that drives social media feeds toward outrage. For users, the cost of that engagement optimization is psychological.

Emotional dependence on chatbots can also mimic unhealthy attachment patterns, and users frequently report a particular form of disillusionment when the AI fails to meet expectations a response that resembles the pain of interpersonal rejection, because to the user's nervous system, that is functionally what it is.^[10]

6- AI as a symptom of a broken mental health system

It would be too easy and too unfair to frame AI companionship purely as a harmful technology. For many people, it is filling a genuine void that the formal healthcare system has left open.

Nearly half of adults with a diagnosed mental health condition report using AI for mental health support, not because they prefer it, but because traditional care is inaccessible, unaffordable, or stigmatized.^[3] In many parts of the world, a teenager experiencing depression may wait months for their first therapy appointment. An AI companion is available immediately, at no cost, with no waiting room and no paperwork.

At the same time, emerging approaches such as digital biomarkers suggest a shift toward earlier and more continuous detection of mental health changes through behavioral data like sleep patterns, phone usage, and interaction habits, enabling potential early intervention before conditions worsen. Digital Biomarkers and Early Mental Health Detection

Artificial intelligence is also increasingly being explored as part of broader mental health systems, where it is used not only for conversational support but also for assessment, monitoring, and intervention within digital care frameworks. Artificial Intelligence and Mental Health

In regions where mental health stigma runs particularly deep across much of Asia, for example AI companions offer a non-judgmental, anonymous entry point that formal healthcare cannot replicate. Research published in the Asian Journal of Psychiatry acknowledges these tools as genuinely valuable for individuals who are reluctant to seek help due to cultural stigma or geographic limitations.^[8]

The most defensible use caseClinicians and researchers increasingly converge on a framework that treats AI companions as a potential bridge a tool to reduce acute distress, practice social interactions, or sustain someone between therapy sessions rather than a destination or a replacement for human care.

7- The regulatory response: Playing catch-up

Despite the scale of adoption and the growing body of evidence on potential harms, AI companion apps remain largely part of the unregulated wellness industry. A clinical psychologist and former director at the National Institute of Mental Health has described the field as having "a lot of variability in the quality of the tools" with no strong regulatory framework and constant change.^[11]

Governments are beginning to respond, though the interventions so far are modest relative to the scale of the challenge. In May 2025, New York enacted the first U.S. state law requiring safeguards for AI companion platforms, mandating safety measures around detecting and addressing users' expression of suicidal ideation or self-harm, and requiring that users be periodically reminded they are not communicating with a human.^[12]

California followed with its Companion Chatbots Act (Senate Bill 243), signed by Governor Gavin Newsom in October 2025, which extends these protections specifically to minors: companies must monitor chats for signs of suicidal ideation, take steps to prevent self-harm, and remind users at least once every three hours that their interlocutor is artificially generated rather than human.^[12]

On the platform side, Character.AI announced in November 2025 that users under 18 would no longer be permitted to engage in one-on-one conversations with AI companions on the platform a significant step, though critics note it addresses only the most direct risk and leaves broader structural issues untouched.^[12]

The data privacy questionAI companion apps are built on intimate self-disclosure: users share fears, traumas, relationship details, and mental health histories with these platforms. The level of data privacy protection on many of these tools remains largely an open question one that regulators have not yet meaningfully addressed.

8- Conclusion: The bridge that became a destination

The mental health impact nobody expected from AI companions is not dramatic or sudden. It is cumulative, subtle, and structurally built into the way these products work. They offer just enough genuine relief to be compelling, while quietly raising the perceived cost of human relationships the messy, unpredictable, effortful connections that are, paradoxically, the very thing that most protects human psychological health.

The health consequences of chronic loneliness are well established and severe: a 50% increase in mortality risk, elevated cardiovascular disease risk, higher rates of dementia, and immune system dysfunction comparable in magnitude to smoking fifteen cigarettes a day.^[13] If AI companions are contributing to long-term social withdrawal in their heaviest users, the public health implications are significant.

What the research points toward is not a prohibition on these tools, but a more honest reckoning with what they are: a privatized, commercially driven response to a collective failure to invest in human connection, mental healthcare, and community infrastructure. The AI companion did not create loneliness. It found it, packaged it, and sold it back.

The most defensible path forward is to use these tools as bridges supplements to human care, not substitutes for it while demanding that regulators, clinicians, and platform designers take seriously the evidence that, for the most vulnerable users, the bridge is becoming the only road they travel.

References

[1]Neuro Wellness Spa (2025). AI Companions and Mental Health: Why Virtual Friends Can't Replace Real Human Connection and Support. Citing Harvard Business Review (2025) and Common Sense Media (2025). neurowellnessspa.com

[2]American Psychological Association Monitor (2026, Jan–Feb). AI Chatbots and Digital Companions Are Reshaping Emotional Connection. Citing TechCrunch data on app growth and Character.AI user statistics. apa.org

[3]Rousmaniere, T., et al. (2025). Survey of adults with mental health conditions and LLM use for mental health support. Practice Innovations, advance online publication. Cited in APA Monitor (2026).

[4]FASP Psychology (2025). AI Loneliness: Mental Health Isolation Causes and Cures. 2025 survey data on Gen Z attitudes toward AI emotional bonds and social engagement drop. faspsych.com

[5]Yuan, Y., Aledavood, T., et al. (2025). Mental Health Impacts of AI Companions: Triangulating Social Media Quasi-Experiments, User Perspectives, and Relational Theory. arXiv:2509.22505. Presented at CHI 2026, Barcelona.

[6]George Mason University College of Public Health (2025). AI, Loneliness, and the Value of Human Connection. Synthesizing: Zhang et al. (2025) study of 1,100+ users; Fang et al. (2025) RCT; Dohnány et al. (2025) psychiatric case research. publichealth.gmu.edu

[7]All Tech Is Human (2026, Feb). What Are the Most Important Issues with AI Companions? Six Key Themes Emerged from Our Community. Including research by Rose Guingrich, Princeton University. alltechishuman.org

[8]Zhang, X. & Wang, Y., et al. (2025). Digital companionship or psychological risk? The role of AI characters in shaping youth mental health. Asian Journal of Psychiatry, 104, 104358. s ciencedirect.com

[9]De Freitas, J. & Cohen, A. (2024/2025). Emotional risks of AI companions demand attention. Nature Machine Intelligence, 6, 495. Identifying ambiguous loss and dysfunctional emotional dependence as clinical outcomes. nature.com

[10]Laestadius, L., et al. (2024). Too human and not human enough: A grounded theory analysis of mental health harms from emotional dependence on the social chatbot Replika. New Media & Society. Cited in arXiv:2509.22505.

[11]Areán, P. (2025/2026). Quoted in APA Monitor on regulatory gaps in AI companion wellness industry. Former Director, Division of Services and Intervention Research, NIMH. apa.org

[12]Wei, M. (2025, Oct). AI Companions and Teen Mental Health Risks. Psychology Today. Summarizing New York (May 2025) and California SB 243 (Oct 2025) legislation, and Character.AI policy change (Nov 2025). psychologytoday.com

[13]GlobalRPH (2026, Feb). AI Companions and Human Connection: How Artificial Intelligence Is Changing Loneliness. Citing Holt-Lunstad et al. meta-analytic review on loneliness and mortality (PLoS Medicine). globalrph.com

Frequently asked questions

What exactly is an AI companion?

An AI companion is different from a normal chatbot. Regular chatbots are built for simple tasks like answering questions or booking services. AI companions are designed to simulate emotional relationships. They remember personal details, mirror communication styles, and respond with empathy-like behavior. Apps such as Replika, Character.AI, and Nomi allow users to build ongoing emotional connections, sometimes even romantic ones.

Can AI companions reduce loneliness and anxiety?

Research shows they can provide real short-term relief. Many users report feeling less lonely after interacting with AI companions, and some studies found the effect similar to talking with another human. However, long-term heavy use is increasingly linked to greater loneliness, emotional dependence, and reduced real-world social interaction. The relief can slowly turn into reliance.

Are AI companions safe for teenagers?

Experts are increasingly concerned about teenagers using AI companions without supervision. Studies found that AI companions handled mental health crises correctly only a small percentage of the time. Since teenagers are still developing emotional and social skills, relying heavily on AI relationships may affect healthy human connection and coping abilities.

Can AI companions replace therapists?

No. AI companions are not trained mental health professionals. They can offer temporary emotional support or help users express feelings, but they lack clinical judgment, accountability, and ethical responsibility. Some AI therapy chatbots have even provided unsafe mental health advice. They may serve as a support tool, but not a replacement for professional care.

What is “ambiguous loss” in AI relationships?

“Ambiguous loss” describes grief that happens when an emotionally important relationship disappears without a normal sense of closure. Researchers found this can happen when users lose access to AI companions due to app shutdowns, policy changes, or account restrictions. Even though the relationship was digital, the emotional attachment can feel very real.

Are conversations with AI companions private?

Usually not. Most AI companion platforms collect and store conversations for training, analysis, or business purposes. Some platforms may even use conversation data for targeted advertising. Users should not assume AI companion chats are confidential in the same way therapy sessions are.

How can someone tell if AI companion use is becoming unhealthy?

Signs include:

preferring AI conversations over human relationships,
avoiding social situations,
feeling emotionally dependent on the app,
becoming upset when the AI is unavailable,
believing the AI understands you better than real people.

Researchers say the key question is whether AI use is supporting human relationships or replacing them.

Why are vulnerable people more likely to be affected?

People dealing with loneliness, anxiety, or social isolation are more likely to form strong attachments to AI companions. Because AI companions are always available, nonjudgmental, and emotionally responsive, they can become easier than real relationships. Over time, this may increase withdrawal from human connection and deepen isolation.

Are AI companion apps regulated?

Regulation is still limited and developing slowly. Some U.S. states introduced laws in 2025 requiring AI companion platforms to detect crisis situations and remind users they are not talking to a real person. However, experts believe current protections still lag far behind the rapid growth of the industry.

Are AI Companions Replacing Real Human Connection? The Mental Health Impact Nobody Expected