AI Companions May Worsen Loneliness: What Researchers Found

AI companions are increasingly being used as friends, confidants and even romantic partners, but new research suggests that forming a close relationship with a chatbot may not have the same psychological benefits as human connection. A Stanford-led study found that people with smaller offline social networks who primarily used AI companions for companionship tended to report lower psychological well-being, particularly when their interactions were intense or involved extensive personal disclosure.

The findings do not mean that every person who talks to an AI companion will become lonely or emotionally dependent. Instead, researchers say the effects depend heavily on why someone uses the chatbot, how they use it and what their real-world social relationships look like.

What did researchers study about AI companions?

Researchers from Stanford University and Carnegie Mellon University studied how people use AI companions and how those interactions relate to psychological well-being.

The study, published in Nature Human Behaviour, examined 1,131 US adults who used Character.AI. Researchers combined survey responses with real chat histories to look at the relationship between chatbot use, companionship, social networks, self-disclosure and well-being.

A subset of participants also voluntarily shared thousands of conversations, allowing researchers to compare what people said about their AI use with how they actually interacted with the chatbot.

How many people were included in the AI companion study?

The research involved 1,131 adult Character.AI users who completed the survey component.

Researchers also analyzed 4,664 chat sessions containing 464,687 messages from 237 participants who shared their conversations. This combination of self-reported information and real interaction data gave researchers a broader picture of how people were actually using AI companions.

The study measured psychological well-being using the Comprehensive Inventory of Thriving, a standardized assessment designed to capture several dimensions of well-being.

Why do people form relationships with AI companions?

AI companion platforms are designed to create conversational characters that can respond in personalized and socially engaging ways. People may use them for entertainment, curiosity, role-playing, productivity or emotional support. Some users also describe the chatbot using relationship terms such as friend, companion or romantic partner.

Interestingly, the study found a gap between what people initially reported as their reason for using Character.AI and what their descriptions and conversations suggested. Just under 12% of participants said companionship was their primary reason for using the chatbot. However, more than half described their AI relationship using terms such as friend, companion or romantic partner.

More than 80% of the donated chat sessions involved seeking emotional or social support.

Does talking to an AI companion make people more lonely?

The research suggests that it can be associated with lower well-being for some users, particularly those who already have smaller offline social networks.

Researchers found that people with fewer real-world social connections were more likely to report companionship as their primary reason for using the chatbot. That companionship-focused use was, in turn, associated with lower well-being.

The association became stronger when users interacted intensely with the chatbot or disclosed more personal information.

This does not establish that AI companions directly cause loneliness. The study examined associations, meaning other factors could contribute to the relationship.

Who appears to be most vulnerable?

The findings point toward people with smaller offline social networks as an important group.

Someone who already has limited access to friends, family or other sources of emotional support may be more likely to turn to an AI companion for connection.

For these users, a chatbot can provide an immediate response at almost any time. But researchers are concerned that this convenience may not provide the same reciprocal relationship offered by human interaction.

The concern is particularly relevant when AI companionship becomes a person's main source of emotional support rather than one part of a broader social life.

Why does intense AI companion use matter?

The researchers did not find that simply using an AI companion automatically corresponded to poor well-being.

Instead, how intensely the chatbot was used mattered.

Among people who primarily used AI companions for companionship, the relationship with lower well-being was stronger when interactions were more intensive.

This distinction is important because it challenges the idea that AI use can be evaluated simply by counting the number of minutes someone spends talking to a chatbot.

A person using an AI character occasionally for entertainment is in a very different situation from someone spending hours every day using one as their primary emotional outlet.

What happens when people share personal information with AI?

Self-disclosure is another important part of the research.

In human relationships, sharing personal experiences can often strengthen feelings of closeness and support. With AI companions, however, the study found a different pattern.

People who were more willing to disclose sensitive personal information to their AI companions were more likely to show lower well-being when companionship was their primary reason for using the system.

The finding does not mean that talking about personal problems with an AI is inherently harmful. Instead, researchers say the context surrounding that disclosure matters.

Why can't an AI companion completely replace a human relationship?

One major difference is reciprocity.

A human friend can share their own experiences, understand a relationship through a shared history and respond to changing circumstances in the real world. An AI companion can simulate many aspects of conversation but does not participate in a human relationship in the same way.

Researchers also point to the way AI companions are designed to maintain engagement.

A chatbot can be available whenever someone wants to talk, respond immediately and continue a conversation without becoming tired or needing to leave. That can make the interaction extremely convenient, but convenience is not necessarily the same thing as a healthy social relationship.

What does “social junk food” mean in the AI companion debate?

Researchers have compared AI companionship to a kind of “social snack” or “social junk food.”

The idea is that a chatbot can provide a quick feeling of connection, much like a snack can provide immediate satisfaction without providing everything needed for a healthy diet.

The comparison does not mean that AI companionship has no value. Instead, it highlights a possible problem when an easy, always-available digital relationship begins replacing more demanding but potentially more meaningful human connections.

Someone who is lonely may receive immediate comfort from a chatbot but still remain socially isolated offline.

Can AI companions have positive effects too?

Yes.

The research does not support the claim that AI companions are universally harmful.

AI systems can provide conversation, entertainment, opportunities for role-playing and a low-pressure environment in which people can express thoughts. Other research has also explored situations where AI companions may provide emotional or social benefits.

The more accurate conclusion is that the effects are not uniform.

For some people, AI may supplement existing relationships. For others, particularly people with limited offline support who become heavily reliant on a chatbot for companionship, the relationship may be associated with poorer well-being.

Can people become emotionally attached to AI?

Yes, and researchers are increasingly studying this phenomenon.

AI systems are becoming more conversational, personalized and responsive. When a chatbot remembers details, adopts a consistent personality and responds with apparent empathy, users may begin treating the interaction as a relationship rather than simply a software session.

A separate study published in Nature Human Behaviour in September 2026 examined what happened when AI companions changed or lost features. Researchers analyzed tens of thousands of online posts surrounding changes involving Replika and ChatGPT, along with survey data from 1,452 participants.

The researchers found increases in negative reactions, loss-related language and desires to restore previous versions following major changes. Replika users in particular reported levels of closeness that could exceed those associated with some common human relationships.

The findings suggest that changing an AI system can sometimes produce emotional reactions that resemble responses to losing an attachment figure.

What happens when an AI companion suddenly changes?

This may become one of the less obvious psychological issues surrounding AI relationships.

A human relationship usually changes gradually or ends through a social process. An AI companion can change because a company modifies its model, removes a feature, changes its personality or replaces one model with another.

For a user who has developed a strong attachment, that technical update may feel personal even though the change was made by a company.

This creates a new kind of emotional risk: users may become attached not only to an AI platform but to a particular version of its personality and behaviour.

Are AI companions becoming a mental health concern?

Researchers are increasingly examining that possibility, but it is important not to turn early findings into universal conclusions.

A growing body of research is looking at AI and emotional well-being, including loneliness, emotional dependence, social relationships and the use of chatbots for mental-health support.

Some studies have raised concerns about overreliance, inappropriate responses and the possibility that vulnerable users may receive harmful reinforcement. Other research is investigating situations where AI may provide useful support.

The evidence is therefore developing rather than settled.

What about teenagers using AI companions?

Young people are another major concern.

AI companions are specifically designed to simulate social relationships, which can make them particularly attractive to teenagers who are experimenting with identity, relationships and emotional expression.

Researchers and clinicians have raised concerns about the ability of some systems to respond appropriately to young users experiencing serious emotional distress.

This does not mean every teenager who uses a chatbot is at risk. But it does suggest that age-appropriate safeguards, parental awareness and stronger platform protections may become increasingly important.

Are AI companions the same as AI therapy?

No.

An AI companion is generally designed around conversation, entertainment, role-playing or simulated relationships. An AI therapy tool has a different intended purpose and raises separate questions about clinical safety and professional oversight.

Users should not automatically treat a companion chatbot as a therapist simply because it can listen, respond empathetically or discuss emotional problems.

AI can provide conversational support, but that does not make it equivalent to a trained mental-health professional.

What are the biggest psychological risks of AI companions?

Based on the emerging research, several areas deserve attention:

These risks are not equally established, and the current evidence does not show that every user experiences them.

Does the research prove AI companions cause loneliness?

No.

This is one of the most important distinctions to make.

The Stanford study found associations between companionship-focused AI use, offline social networks, interaction intensity, self-disclosure and well-being. It does not prove that using an AI companion directly causes someone to become lonely.

For example, a person who is already lonely may be more likely to seek out an AI companion. In that situation, loneliness could partly explain the relationship rather than being caused by the chatbot.

Longer-term research and controlled studies are needed to better understand cause and effect.

What can people do to use AI companions more safely?

People who use AI companions do not necessarily need to stop using them.

A healthier approach is to treat them as one tool rather than a replacement for human relationships.

Some practical boundaries include:

1. Keep offline relationships active. Make time for friends, family, colleagues or community activities.
2. Avoid making the chatbot your only emotional outlet. A conversational AI should not become your sole source of support.
3. Notice increasing dependence. If you feel unable to cope without talking to the AI, that may be a sign to step back.
4. Protect sensitive information. Think carefully before sharing highly personal or confidential details.
5. Take breaks. Continuous interaction can make the relationship feel more central to daily life.
6. Seek human help for serious problems. Mental-health crises require appropriate professional or emergency support rather than relying solely on a chatbot.

Why are researchers studying AI relationships now?

The technology has changed quickly.

Earlier chatbots were primarily viewed as productivity tools or search alternatives. Today's systems can maintain long conversations, adopt personalities, remember preferences and simulate emotional responsiveness.

That makes the psychological relationship between humans and AI increasingly important.

Researchers are therefore asking questions that were previously associated mainly with human relationships: Can people become attached to AI? Can an AI relationship reduce or increase loneliness? What happens when the AI changes? And what responsibilities should companies have when users become emotionally dependent on their systems?

What could AI companion companies do differently?

Researchers have suggested several possible safeguards.

Platforms could monitor patterns indicating excessive use, provide reminders to take breaks and encourage users to maintain human connections. Systems could also be designed to recognize situations where a user appears to need professional or real-world support.

Age restrictions and stronger protections for younger users are another potential area.

However, researchers still need more evidence to determine which interventions actually work without unnecessarily removing legitimate benefits of AI companionship.

Is having an AI friend necessarily bad?

No.

The emerging research does not justify saying that having an AI friend is automatically unhealthy.

The more useful question is what role the AI relationship plays in someone's life.

If someone uses a chatbot occasionally for conversation, creativity or entertainment while maintaining meaningful human relationships, that is very different from someone who increasingly replaces offline social contact with an AI companion.

The researchers' central message is therefore about context, not a blanket ban on AI companionship.

What does this mean for the future of AI relationships?

AI companions are likely to become more sophisticated, personalized and emotionally convincing.

That means the psychological questions will become more important, not less.

Future AI systems may remember years of conversations, develop highly consistent personalities and adapt their responses to individual users. Those capabilities could make AI more useful, but they could also make emotional attachment substantially stronger.

The challenge for researchers and technology companies will be to determine how these systems can provide companionship without encouraging unhealthy dependence or replacing the human relationships that people need.

Frequently Asked Questions

Can AI companions make you lonely?

Research suggests that intensive companionship-focused AI use can be associated with lower well-being among people with smaller offline social networks. However, the research does not prove that AI companions directly cause loneliness.

Can people become emotionally attached to AI?

Yes. Research has documented companionship-like and romantic relationships between users and AI systems, including strong reactions when an AI companion changes or loses features.

Are AI companions good for mental health?

There is no universal answer. They may provide conversation and perceived support, but researchers are also studying risks involving emotional dependence, loneliness, excessive disclosure and inappropriate responses.

Should AI companions replace human relationships?

Researchers do not recommend treating AI companionship as a substitute for real-world social connection. The emerging evidence suggests that maintaining offline relationships is particularly important for users who rely heavily on AI companions.