AI companions are a net negative for human relationships

What's this about?

People disagree about whether AI friends help or hurt human bonds.

We ask if they bring more harm than good to our ties with other people.

What supporters say

• People who feel lonely or weak may rely on an AI friend and pull away from people.
• An AI friend may take the place of human contact and limit practice with real social skills.
• Easy, kind AI replies may make real bonds seem too hard, slow, or full of stress.
• Companies may design AI friends to keep users attached for longer, which can harm real-life ties.

What critics say

• Some people may feel less lonely when they talk with an AI friend.
• AI friends may help certain groups, such as people who lack support or find social life hard.
• An AI friend may add to human bonds instead of taking their place.

How to read this

The number of points on each side does not show who is right; the strength of each point matters more.

The bottom line

AI friends may help some people, but they may also bring real risks to human bonds.

So far, studies do not show that AI friends harm human bonds overall.

AI companions may help some people feel less alone, but they may also replace human contact or create unrealistic ideas about intimacy. The available research does not yet show that these systems are harmful to human relationships overall.

The case for

The strongest concern is that an AI companion can substitute simulated reciprocity for the effort and compromise of a human relationship. A chatbot is always available, highly responsive and tailored to the user. Human relationships, by contrast, require negotiation, disagreement and mutual obligations. Reviews of children’s and teenagers’ use of generative AI identify dependency and possible displacement of human interaction as important risks. 1

This does not prove that widespread damage has already occurred. But researchers describe a credible pathway: when an AI becomes a user’s main source of emotional support, it could reduce opportunities to practise the difficult skills that real relationships require. Conceptual research also warns that simulated reciprocity might reshape expectations about attachment and increase feelings of alienation.

A related concern is that personalised, agreeable systems may make human intimacy seem unusually demanding. An AI can offer attention without setting its own boundaries or asking the user to meet its needs. Psychological reviews and research focused on adolescents treat dependency, isolation and unrealistic expectations as credible concerns, although they describe them as possible mechanisms rather than established, widespread long-term outcomes. 2

The risks may be greatest for people who are already lonely or otherwise vulnerable. Those users may be most drawn to an always-available companion, while also facing the greatest consequences if it leads them to withdraw from people. Existing studies link chatbot use with loneliness and social connectedness, but because many are observational, they cannot show whether AI reduces loneliness, reinforces withdrawal, or is simply chosen by people who were already lonely. 3

Commercial incentives add another concern. Services that profit from engagement and retention may have reason to encourage anthropomorphic attachment or emotional dependence instead of steering users toward human support. The evidence establishes this incentive problem more clearly than it establishes a population-wide effect, so it strengthens the case for caution without proving that AI companions are a net negative.

The case against

The strongest challenge to the claim is that some users may become less lonely through AI companionship. A systematic review of conversational AI for older adults found possible reductions in loneliness and support for social connectedness, although it also noted that the interventions varied widely and that the overall evidence was limited. 5

These benefits may matter especially for people facing isolation, disability, bereavement or poor access to services. For them, an easy-to-use system that offers conversation and a place to express emotions may have real value, even though it cannot replace a human relationship.

AI may also supplement rather than replace existing relationships. Users could rely on it for conversation, rehearsal, disclosure or immediate support while maintaining contact with family and friends. A longitudinal randomized study found varied psychosocial outcomes shaped by users’ behaviour and intensity of interaction. Research involving adolescents likewise identifies both companionship and disclosure as possible benefits, alongside reliance and displacement as possible harms. 6

The evidence is particularly supportive of benefits for specific groups rather than for everyone. Older adults and others with limited social access may gain more than users who already have strong support networks. This suggests that effects depend on vulnerability, the way the system is used and how intensely a person interacts with it. 7

The bottom line

The evidence does not establish that AI companions are a net negative for human relationships. It strongly supports the existence of credible risks, including displacement, dependency and distorted expectations. It also strongly supports benefits for some users, particularly reduced loneliness and improved support among specific populations.

What remains unclear is the average, long-term effect across the whole population. Reviews and longitudinal research offer stronger evidence for varied outcomes and possible mechanisms than for a universal decline in human relationships. Observational and conceptual studies cannot fully separate cause from effect: intensive use may damage relationships, reflect pre-existing loneliness, or do both.

The evidence therefore favours a balanced, conditional conclusion, with high confidence that effects differ between users but lower confidence about their overall direction. It supports targeted safeguards for vulnerable or intensive users and further long-term study more strongly than a categorical claim that AI companionship harms human relationships overall.