AI companions are a net negative for human relationships

What's this about?

People disagree about whether AI friends harm real-life bonds more than they help. We still do not know the full answer.

What supporters say

• AI friends can help some people feel less alone for a short time.
• They can give comfort when someone needs to talk right away.
• Some users keep close ties with friends and family while using AI.
• AI can help at any hour, often for little cost and with no fear of blame.

What critics say

• Feeling better after one AI chat does not prove someone built stronger real-life bonds.
• Studies do not show that AI friends give the same help as human friends.
• AI may lead to strong attachment and dependence on a made-to-fit kind of closeness.
• User stories cannot show how often AI adds to human ties or takes their place.

The bottom line

AI friends may ease loneliness for some people, mainly in the short term. We do not yet know if they help or harm real-life bonds overall.

AI companions are increasingly marketed as friends, confidants and romantic partners. The evidence suggests they can ease loneliness in the short term, but it remains unclear whether, overall, they strengthen or weaken people’s real-world relationships.

The case for

The strongest argument against calling AI companions a clear social harm is that they can make some people feel less lonely. A systematic review and meta-analysis found reductions in loneliness in certain groups and settings, while a randomized comparison found that generative AI could provide short-term benefits—though it was not the same as human companionship. 1

For many users, these systems may serve as a supplement rather than a replacement for friends, family or partners. Qualitative studies describe people using chatbots for comfort, emotional regulation or temporary support while remaining connected to wider social lives. Some form real attachments to the technology, but that does not necessarily mean they abandon human ties. 2

Accessibility is another potential benefit. AI companions are available immediately, at any hour, and may be especially appealing to people who face barriers to counselling, social services or regular contact with others. Users often describe the systems as responsive and emotionally supportive, offering a form of interaction that is difficult to obtain elsewhere. 3

These benefits matter, particularly for people experiencing isolation. But a lower loneliness score does not by itself show that a person’s human relationships have become stronger, more durable or more mutual. Most studies have measured feelings such as loneliness, mood and attachment—not the quality of users’ offline social networks.

The case against

The central worry is that AI companions can encourage emotional dependence on a relationship that cannot truly be reciprocal. Their constant availability, personalization, apparent understanding and ease of control can make them highly attractive. Unlike human relationships, they do not require negotiation, mutual vulnerability or accountability.

Research has found links between loneliness, companion-chatbot use and dependence, raising concerns about a cycle in which isolated people turn increasingly to AI and reduce human contact (see Figure 1). 4 But the evidence cannot yet show which comes first: chatbot use may worsen loneliness, or people who are already lonelier may be more likely to seek out these tools.

Simulated intimacy may also affect what users expect from human relationships. AI can offer attention and affirmation without disagreement, delay or personal needs of its own. That creates a plausible risk that some users will find ordinary relationships less satisfying or harder to navigate. Still, the research has not established that this shift in expectations happens commonly. 5

There are separate concerns about the companies building these products. AI companions are sold as relational goods, yet commercial goals may not match users’ long-term welfare. Policy and research sources have identified risks including emotional manipulation, unsafe advice, privacy misuse, boundary violations and designs that reinforce unhealthy dependence. 6 7 These are credible—and sometimes documented—ways harm can occur, but there is not yet evidence showing how frequent they are or proving that they have caused a broad decline in human relationships.

The bottom line

The claim that AI companions are a net negative for human relationships is plausible, but not established. The evidence is genuinely mixed: AI can offer short-term support and relief from loneliness, while also creating meaningful risks of attachment, dependence and blurred relational boundaries.

The effects are likely conditional rather than universal. Benefits appear more likely when AI use is temporary and supplementary to human contact. Risks appear greater when use becomes intensive, exclusive or emotionally dependent, especially for vulnerable users.

The biggest gap is a lack of long-term research that can separate cause from correlation and measure what happens to real relationships over time. Current studies are often short, varied in design, or focused on wellbeing rather than the durability and reciprocity of human ties. There is high confidence that AI companions can both help and pose risks; there is low confidence about their overall effect on real-world relationships.