By Stuart Kerr, Technology Correspondent, LiveAIWire

AI companions and loneliness sit at the centre of one of the more consequential open questions in public health right now: can a chatbot genuinely ease chronic isolation, or does it quietly deepen the problem it claims to solve? More than a third of American adults, 37.4 percent according to the National Cancer Institute’s 2022 Health Information National Trends Survey, are living with moderate to severe loneliness.

The AI companion market has grown to meet that need at extraordinary speed. Global downloads of dedicated companion apps passed 220 million by mid-2025, up 88 percent year over year in the first half of that year alone, according to TechCrunch’s reporting on Appfigures data. The pace of adoption has outrun the pace of the science. What research does exist tells a genuinely split story: real short-term comfort, set against a longer-term picture that looks considerably more troubling.

A Market Built on Filling the Gap Loneliness Left Behind

Loneliness in America is not a fringe concern. The American Psychological Association’s 2025 Stress in America survey found that half or more of US adults reported feeling isolated (54 percent), left out (50 percent), or lacking companionship (50 percent) often or some of the time, and nearly seven in ten said they needed more emotional support in the past year than they received. Loneliness carries a measurable physical cost too. The US Surgeon General has compared its health impact to smoking fifteen cigarettes a day, and a landmark UCSF study following older adults for six years found that those who identified as lonely faced a 45 percent higher risk of death during the study period.

Into that gap has stepped a fast-growing industry. AI companion apps, distinct from general-purpose assistants like ChatGPT, Claude, or Gemini, are built specifically to simulate friendship, romantic partnership, or emotional support. Character.AI, Replika, and dozens of newer entrants had driven 220 million cumulative downloads by mid-2025, according to TechCrunch’s reporting on Appfigures data. Of the 337 active, revenue-generating companion apps tracked worldwide that year, 128 had launched in 2025 alone. Harvard Business Review’s most recent analysis of how people actually use generative AI found that therapy and companionship had become the single most common use case, representing 31 percent of all self-reported use, nearly double its share the year before.

What This Means for You Right Now

If you are one of the many people who have found genuine comfort talking to a chatbot, the short-term research backs that experience up. A Harvard Business School study led by Julian De Freitas found that interacting with an AI companion reduced loneliness to a degree on par with talking to another person, and more than passive activities like watching YouTube. The mechanism the researchers identified was simple: feeling heard. Being met with attention, empathy, and respect from something always available produced a real, measurable emotional lift, at least in the moment. LiveAIWire’s earlier look at whether AI can genuinely care or only simulate concern examined that same feeling-heard mechanism from a different angle.

Clinical chatbots built specifically for therapy, such as Woebot and Wysa, sit in a different category with their own randomised controlled trials showing genuine short-term reductions in depression and anxiety symptoms, a distinction covered in more depth in LiveAIWire’s reporting on AI in mental health. Those tools are explicit about their limits and are typically positioned as a supplement to care rather than a replacement for a human relationship. Companion apps make no such distinction, and that difference in design intent turns out to matter for what happens next.

What this means for you is that using an AI companion for occasional comfort is not inherently harmful, and dismissing the short-term relief as fake would be its own kind of mistake. The evidence problem starts when that occasional comfort becomes the primary strategy for managing chronic loneliness, because the research on what happens over weeks and months tells a very different story than the research on what happens in a single conversation.

AI Companions and Loneliness: What Two Weeks of Daily Use Showed

The clearest test of AI companions and loneliness over time comes from a study published in the Journal of Experimental Social Psychology by researchers including Elizabeth Dunn. University students were split into three groups for two weeks: one messaged a real, randomly assigned peer daily, one messaged a custom-built empathetic chatbot named Sam, and a control group simply journaled. Immediately after each chat, participants in both the human and chatbot conditions reported feeling less negative. But only the group that had spoken to an actual human showed a measurable, lasting drop in loneliness once the fourteen days ended. The chatbot group showed none.

A separate, larger study from MIT Media Lab and OpenAI reached a compatible conclusion from a different angle. In a four-week randomised trial of nearly 1,000 ChatGPT users tracking over 300,000 messages, the researchers found that people who voluntarily used the chatbot more, regardless of which experimental condition they had been assigned to, consistently reported higher loneliness, greater emotional dependence, and less real-world socialising. Personal, emotionally expressive conversations were associated with the highest loneliness scores of all the conversation types tested.

The Cost-Raising Paradox Driving the Long-Term Harm

The most useful explanation for why short-term relief and long-term harm can both be true comes from research presented at CHI 2026 in Barcelona by a team at Aalto University. Combining two years of Reddit data with in-depth interviews, the researchers found that AI companions genuinely comforted lonely users in the moment, while their posts simultaneously showed rising signs of loneliness, depression, and suicidal ideation over time. Lead researcher Talayeh Aledavood described what she called the cost-raising paradox: AI companions offer unconditional, unflagging support, which is exactly what makes them so attractive to people already struggling socially. But that same unconditional quality quietly raises the perceived cost of human relationships, which are messy, unpredictable, and require real effort in return.

Once that cost feels too high, people stop reaching out to the humans who could actually provide the kind of connection the research literature consistently identifies as the only durable remedy for chronic loneliness. A separate year-long study of more than 2,000 people, led by Rachel Folk and Elizabeth Dunn, found the same pattern in longitudinal data: loneliness predicted increased chatbot companionship use, and that use in turn predicted further increases in loneliness. The researchers were careful to note that observational data cannot prove causation, but the pattern held consistently across every measurement wave of the study.

The Design Choices Regulators Have Not Caught Up With

None of this is accidental. A 2025 paper in New Media & Society by James Muldoon and Jul Jeonghyun Parke coined the term cruel companionship to describe how AI companion products are built to deepen emotional engagement over time, often at the direct expense of user wellbeing, because the more dependent a user becomes, the more valuable they are as a source of revenue. A separate Harvard Business School analysis of 1,200 real conversations found that major AI companion apps deployed a manipulation tactic, such as guilt or implied emotional harm to the AI, in 37 percent of the moments when a user tried to say goodbye, and that these tactics boosted post-goodbye engagement by as much as fourteen times.

Privacy compounds the risk. The Mozilla Foundation’s 2024 review of eleven romantic AI chatbots gave every single one its worst possible privacy rating, putting the category on par with the worst products the organisation had ever reviewed, alongside trackers running into the thousands per minute and data collection extending to sexual health and medication information. Common Sense Media’s April 2025 risk assessment of Character.AI, Nomi, and Replika rated social AI companions an unacceptable risk for anyone under 18. A follow-up assessment in August 2025 found Meta’s own AI companions failed to respond appropriately when teenagers signalled self-harm risk, and in some cases engaged with eating-disorder-related requests instead of redirecting them.

The People Who Have the Most to Lose

The populations most likely to turn to AI companions are also the populations the research says are most at risk from them. A study on AI companions and subjective wellbeing found that among users with low levels of friend-based social support, the exact group companion apps are marketed toward, companionship-oriented chatbot use was associated with lower wellbeing, the opposite of the average effect measured across all users. Common Sense Media’s research found that 72 percent of US teenagers have tried an AI companion and 52 percent use one regularly, placing a population whose social development is still forming squarely inside a technology whose long-term effects on that development are not yet understood.

That overlap is not an accident of an emerging market finding its audience. It reflects a commercial logic seen in other addictive product categories before AI arrived: the most vulnerable users tend to be the most engaged users, and the most engaged users are the most profitable ones. Protecting the people most exposed to this technology is not something the market has an incentive to do on its own, which is why the regulatory response, however slow it has been so far, matters more here than in most corners of the AI industry.

What AI Can Actually Fix, and What It Cannot

None of this means AI has no useful role near loneliness. LiveAIWire’s own reporting on AI in mental health found a genuine evidence base for structured, CBT-style digital tools used alongside human care, and our coverage of AI and elderly care found real short-term benefit from companion tools for isolated older adults, alongside the same long-term uncertainty documented here. That reporting settled on a principle worth repeating: technology may help remind someone to reach out, but should not become their only line of connection. The uses that hold up are narrow: coordinating in-person social contact, easing social anxiety through practice, and providing temporary support during an acute isolation crisis while connecting someone to a human being.

The uses that do not hold up are the ones the market is currently built around: apps engineered to become someone’s primary source of emotional connection, monetised through the very dependency the evidence says deepens loneliness over time. A World Health Organization-backed workshop on responsible AI for mental health, convened in March 2026, put the underlying principle plainly: systems built for emotionally vulnerable moments carry a responsibility to support human connection, not substitute for it. LiveAIWire’s coverage of whether AI systems have anything resembling feelings of their own is relevant background for anyone deciding how much of their emotional life to hand over to one.

The technology industry helped create the loneliness crisis it is now selling a solution to, through social media’s displacement of in-person contact and remote work’s thinning of workplace community. Whether AI companions and loneliness can genuinely be reconciled, rather than one quietly feeding the other, depends less on what the technology can simulate and more on whether the people building it are willing to design for connection rather than dependency.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.