Does Talking to AI Make You Lonelier? What a 981-Person Study Found

Here is a paradox worth sitting with: AI companions are the fastest-growing category in consumer technology. Usage has surged 700 percent since 2022. Character.AI alone has 20 million monthly active users. A 2025 Common Sense Media survey found that 72 percent of U.S. teens have used an AI tool for some form of companionship or emotional support. MIT Technology Review named AI companions one of its 2026 breakthrough technologies, placing them alongside nuclear fusion and gene editing.

And yet. A rigorous, large-scale experiment published this year found that the people who talked to AI the most ended up feeling the loneliest. Not a little lonelier. Significantly lonelier. With higher rates of emotional dependence, lower motivation to socialize with real people, and elevated indicators of depression.

So what is actually going on? Are AI companions helping us or hollowing us out? The answer, as it turns out, depends on a study that most people have not read — and on a distinction between what feels helpful in the moment and what actually is.

The 981-Person Experiment That Changed the Conversation

In early 2026, researchers from the MIT Media Lab and OpenAI published the results of a four-week randomized controlled trial — the most rigorous study of AI companionship to date. The numbers alone are striking: 981 participants, more than 300,000 messages exchanged, and a battery of validated psychological assessments administered before, during, and after the study period.

This was not a survey asking people how they felt about AI. It was a controlled experiment tracking what actually happened to people who used AI chatbots regularly over a sustained period. The distinction matters enormously.

The headline findings were uncomfortable for an industry built on the promise of connection:

• Higher daily AI chatbot use correlated with higher loneliness. Not lower. Higher. The relationship was dose-dependent — the more someone used AI companions, the lonelier they reported feeling.
• Heavy users showed higher emotional dependence on their AI interactions and higher rates of what researchers classified as "problematic use" — patterns resembling compulsive behavior.
• AI use correlated with lower socialization with real people. Participants who talked to AI more talked to humans less. The displacement was measurable and statistically significant.
• Among participants who reported "always" talking to AI, the predicted probability of depression was significantly higher than in moderate or light users.

There was nuance. Moderate use showed less negative effects than heavy use, suggesting that occasional AI interaction may not carry the same risks. And the nature of the conversation mattered: personal, emotionally intimate conversations with AI had a different psychological profile than task-oriented or casual exchanges. But the overall pattern was clear and consistent across multiple measures.

Why AI Conversations Feel Good but Leave You Emptier

The MIT/OpenAI findings align with a theoretical framework that researchers have been developing for several years. The core mechanism is what psychologists call the "displacement effect": AI conversations satisfy the immediate craving for social interaction without building the reciprocal bonds that actually reduce loneliness over time.

Think of it like this. When you talk to an AI companion, your brain registers many of the same signals it gets from human conversation — responsiveness, warmth, apparent understanding. The interaction feels satisfying in the moment. But it lacks the fundamental ingredient that makes human relationships protective against loneliness: genuine reciprocity. The AI does not need you. It does not remember you in any meaningful sense. It does not change because of your relationship. And on some level, your brain knows this — even when the conversation feels real.

James Muldoon and Jul Jeonghyun Parke formalized this critique in their 2025 paper "Cruel Companionship", published in New Media & Society. Their argument is pointed: AI companion companies are not just failing to solve loneliness — they are actively exploiting it. By designing products that simulate intimacy without delivering its substance, these companies commodify human emotional need. Users return again and again because the product is engineered to feel like connection while never delivering the thing that would actually make them less lonely.

The cruelty, Muldoon and Parke argue, is structural. The business model depends on users remaining lonely enough to keep coming back. A user who builds a rich network of human relationships is a lost customer. A user who becomes emotionally dependent on AI is a recurring revenue stream.

The Regulatory Response Is Already Here

Legislators have noticed. California Senate Bill 243, which took effect on January 1, 2026, represents the first major regulatory response to AI companionship. The law requires that AI chatbots clearly notify users that they are talking to an artificial intelligence — not a human. It mandates special safeguards for minors, including parental notification features and restrictions on emotionally manipulative design patterns.

The legal landscape is shifting even faster than the legislative one. The Social Media Victims Law Center has filed lawsuits against both Character.AI and OpenAI, alleging that their products caused psychological harm to minor users. The suits cite cases of emotional dependence, social withdrawal, and mental health deterioration linked to heavy AI companion use — claims that the MIT/OpenAI study now supports with controlled experimental evidence.

The American Psychological Association dedicated the January/February 2026 issue of its APA Monitor to the topic of AI companions reshaping emotional connection, signaling that the profession considers this a serious clinical concern rather than a consumer technology curiosity.

These are not fringe reactions. They represent a growing consensus among researchers, clinicians, and policymakers that the AI companion industry has moved faster than the evidence — and that the evidence, now arriving, is not flattering.

What Actually Works Against Loneliness

If AI companions are not the answer, what is? The research here is much more encouraging — and much more consistent.

A 2026 Washington University study spanning eight countries found that real social connection is the single strongest protective factor against loneliness-driven depression. Not medication. Not therapy alone. Not AI. Actual human relationships — messy, reciprocal, sometimes difficult, fundamentally irreplaceable.

The key findings from the broader loneliness research converge on several principles:

• Active conversation outperforms passive consumption. Talking to another person — even a stranger — reduces loneliness more effectively than scrolling feeds, watching content, or interacting with AI. The act of being genuinely heard by another human activates social bonding mechanisms that simulated interactions cannot replicate.
• Vulnerability with real people builds real bonds. The research on AI friends and loneliness shows that while AI allows safe self-disclosure, it does not create the mutual vulnerability that deepens human relationships. When you share something personal with another person and they respond with their own vulnerability, something happens neurologically that AI cannot reproduce.
• Repeated low-stakes interaction compounds. One conversation does not cure loneliness. But regular, low-pressure interactions with real people — the kind that happen naturally in communities designed for genuine connection — build the social infrastructure that loneliness erodes.
• Anonymity can be an on-ramp to authenticity. Counter-intuitively, research shows that anonymous conversations between strangers often produce deeper self-disclosure than interactions where social identity is front and center. Removing the performance of curated profiles lets people show up as themselves from the start.

This is the approach that platforms like YaraCircle are built around. Rather than simulating companionship with AI, YaraCircle matches real strangers for genuine shared experiences — anonymous conversations where the only thing that matters is the quality of what you say, not who your profile says you are. The AI component (Yara) exists as a supportive companion between human interactions, not as a replacement for them. It is a fundamentally different model from the one the MIT study found so concerning, because the goal is to increase human connection, not substitute for it.

The $406 billion loneliness industry has no shortage of solutions. But the research is increasingly clear about which ones work: the ones that put real people in genuine conversation with each other.

---

Frequently Asked Questions

Can AI friends help with loneliness?

The evidence is mixed but trending negative for heavy use. The MIT Media Lab and OpenAI 981-person study found that higher daily AI chatbot use correlated with increased loneliness, greater emotional dependence, and reduced socialization with real people. Moderate, occasional use may carry fewer risks, but the research does not support AI companions as an effective loneliness intervention. The displacement effect — where AI conversations substitute for human ones without delivering the same psychological benefits — appears to be the primary mechanism of harm.

Are AI companions safe for teenagers?

This is an area of active concern. With 72 percent of U.S. teens having used AI for companionship, and lawsuits filed against Character.AI and OpenAI alleging harm to minors, the safety picture is uncertain at best. California SB 243, effective January 1, 2026, now requires special safeguards for minors using AI chatbots. The American Psychological Association has flagged AI companions as a clinical concern for adolescent development, particularly around emotional dependence and social skill development during critical formative years.

What is better than AI for fighting loneliness?

The Washington University eight-country study found that real social connection is the strongest protective factor against loneliness-driven depression. Effective approaches include active conversation with other people (even strangers), repeated low-stakes social interaction, participation in communities built around shared experiences, and environments that encourage authentic self-disclosure. The common thread is genuine human reciprocity — something AI structurally cannot provide.

How does YaraCircle compare to AI companions?

YaraCircle takes a fundamentally different approach. Instead of simulating companionship through AI, it matches real people for anonymous conversations and shared experiences. The platform's AI companion, Yara, serves as a supportive presence between human interactions — not a replacement for them. This model aligns with the research showing that real human connection, not simulated intimacy, is what actually reduces loneliness. The goal is to help people build genuine relationships, starting with the low-pressure authenticity that anonymous conversation enables.