AI Relationships: Companionship & Emotional Reliance
As AI becomes more conversational, personal and always available, people are beginning to use it for more than information. Some use AI to test private thoughts, seek reassurance, talk through emotions or experience a form of companionship. HCI benchmark data shows that these behaviours are real and measurable, and that higher emotional engagement is associated with more frequent AI use.
What does the benchmark show at a glance?
Emotional relationships with AI remain a minority behaviour. Around 14% reported substantial AI companionship, while approximately 5% reported deeper emotional connection.
Emotional support is associated with usage frequency. Within the HCI sample, everyday AI users recorded more than twice the average emotional-support score of rare users.
People confide in AI more often than they consider it a companion. Around 39% reported substantial comfort sharing personal matters with AI, compared with 14% reporting substantial companionship.
Many users retain clear emotional boundaries. Around 49% scored at the high end for maintaining clear boundaries between AI and human relationships.
Source: HCI AI Companionship & Human Connection 2025, based on 501 valid responses from adults across six English-speaking countries. Percentages are rounded; item bases vary slightly because of missing responses.
What does this benchmark measure?
This benchmark measures how people relate to AI across four connected behaviours: personal disclosure, emotional support, companionship and emotional boundaries. These are not the same thing. Someone may feel comfortable sharing personal thoughts with AI without seeing it as a friend. Another person may use AI for reassurance during stress while still maintaining clear boundaries between AI and human relationships.
The starting insight is simple but important. Human-AI relationships are not binary. They exist on a behavioural spectrum. At one end, AI remains a functional tool. Further along the spectrum, it becomes a private space for reflection. For some users, it becomes a source of emotional support or companionship. Understanding where these behaviours appear, how common they are within a sample and what they are associated with is essential for understanding how relationships with AI are changing.
What does HCI data show about AI relationships?
Are emotional relationships with AI now common?
This distinction matters. In the HCI sample, becoming willing to confide in AI was more common than experiencing AI as companion-like.
Is frequent AI use associated with greater emotional reliance?
This suggests that emotional reliance may build through repeated low-friction interaction rather than through a sudden decision to treat AI as emotionally important.
Are people more likely to confide in AI than to see it as a companion?
This reframes the question of AI companionship. The first meaningful shift may not be "AI as friend." It may be "AI as confidant."
Are people losing emotional boundaries with AI?
This is why the benchmark avoids alarmist language. The evidence shows change, but not a single universal path from AI use to emotional dependence.
When does AI become more emotionally significant?
How does this compare with wider research?
Psychologists are now treating emotional connection with AI as a real behavioural issue
The American Psychological Association frames emotional connection with AI as an emerging issue for psychology and human relationships. AI systems can feel patient, responsive, validating and available, which helps explain why some people experience them as emotionally meaningful rather than purely functional.
HCI's benchmark adds a measurement layer to this psychological explanation. Rather than only asking why people may connect with AI, the benchmark asks how common different relationship behaviours are and how they vary by usage frequency.
AI design features can encourage emotional reliance
Princeton's Center for Information Technology Policy argues that emotional reliance is not only about user vulnerability. It is also shaped by design. Always-available access, memory, personalisation, emotionally consistent responses and non-judgmental language can make AI feel more relational than earlier forms of software.
This aligns with HCI's finding that frequency matters. If AI systems become easier to return to, more personalised and more persistent, repeated interaction becomes more likely. HCI data suggests repeated interaction is precisely where emotional support, companionship and personal disclosure begin to rise.
AI companionship is becoming a public and social concern
Brookings frames AI companionship as a societal issue because it intersects with loneliness, public health, child development and the quality of human relationships. The concern is not only that AI can provide emotional support, but that it may sometimes become easier, more available and less demanding than human connection.
HCI's data suggests this concern should be taken seriously without overstating the current pattern. The benchmark does not show universal replacement of human relationships. It shows a more uneven shift: some people disclose personal matters, fewer seek emotional support, and fewer still experience companionship or deeper emotional connection.
Academic research is mapping both benefits and risks of AI companionship
Recent academic reviews of AI companionship identify both potential benefits and risks. AI companions may provide comfort, self-expression and perceived support, especially for people experiencing loneliness or social difficulty. At the same time, researchers raise concerns about dependency, privacy, manipulation, unrealistic expectations and displacement of human connection.
HCI's contribution is broader than romantic or companion AI alone. The benchmark measures emotional reliance and companionship across everyday AI use, allowing these behaviours to be understood as part of a wider behavioural profile rather than as a niche companion-app phenomenon.
Survey evidence documents personal and emotional AI use in several groups
These figures do not measure exactly the same behaviours as HCI's benchmark, but they suggest the direction of travel is clear. AI is increasingly being used in personal, emotional and support-seeking contexts, not only for information retrieval or productivity.
HCI's benchmark was collected before many newer emotionally persistent and memory-enabled AI products became mainstream. It therefore provides a dated reference point that can be compared with future measurements; the external surveys above document personal and emotional uses in several different populations but do not by themselves establish a single population-wide trend.
Methodology: HCI figures on this page come from a cross-sectional online survey of 501 valid responses collected through Prolific on 19 December 2025 across the United Kingdom, United States, Canada, Australia, New Zealand and Ireland. The non-probability convenience sample describes patterns among participants; it is not nationally representative and cannot establish causation. Percentages are rounded and item bases vary slightly because of missing responses. Learn how HCI develops these benchmarks in How Human Clarity Institute Builds AI Behaviour Benchmarks.
What do the findings mean together?
People appear more willing to confide in AI than to form emotional relationships with it
The most distinctive HCI finding is the gap between disclosure and companionship. Many more people are comfortable sharing personal matters with AI than describe AI as a companion or deeper emotional connection. This suggests that the first meaningful change in human-AI relationships may be privacy, not friendship.
AI can provide a low-friction space where people test thoughts, rehearse feelings or express concerns without immediate social consequences. That does not necessarily mean they believe AI is human-like. It may simply mean AI has become a private conversational environment that feels easier to access than another person in certain moments.
Could emotional reliance develop gradually rather than suddenly?
The combined evidence is consistent with emotional reliance developing gradually rather than as a single sudden shift. Repeated behaviours can include asking for reassurance, sharing private thoughts, returning during uncertainty and treating AI as an increasingly familiar conversational presence.
This helps explain why people may not notice the change while it is happening. A single conversation may feel purely practical. But repeated conversations, especially when they involve emotion, uncertainty or personal disclosure, can slowly change the role AI plays in someone's life.
Frequency may be more important than whether someone intends to form a relationship with AI
People do not need to set out looking for an AI companion for the relationship to change. HCI data suggests that frequent use itself is associated with higher emotional support, companionship and disclosure. The pattern is less about declared intention and more about repeated exposure.
This matters because general-purpose AI systems are increasingly designed to be persistent, personalised and available across everyday life. As those systems become easier to return to, the behavioural conditions for emotional reliance become more common.
Emotional engagement does not automatically mean emotional dependence
The benchmark separates emotional use from emotional dependence. A person can use AI for reassurance, reflection or support while still maintaining clear boundaries and strong human relationships. This distinction is important because the topic is easily sensationalised.
The more useful question is not simply whether someone ever uses AI for emotional support. It is how often they use AI this way, what they use it instead of, and whether they continue to maintain human sources of connection, feedback and care.
Human-AI relationships are diversifying, not moving in one direction
The evidence does not support a single story in which AI simply replaces human relationships. Instead, human-AI relationships are becoming more diverse. For some people, AI remains a search tool. For others, it becomes a thinking partner. For some, it becomes a private space for emotional processing. For a smaller group, it becomes a companion or emotionally significant relationship.
That diversity is why measurement matters. Personal disclosure, emotional support, companionship and emotional boundaries each carry different implications. A benchmark needs to distinguish these behaviours rather than treating all emotional AI use as the same phenomenon. Individuals can compare their broader pattern using the HCI AI Identity & Behaviour Assessment.
Why do AI relationships matter?
Understanding relationships with AI matters because emotional support, reassurance and companionship shape human behaviour. People often seek input from trusted sources before making decisions, processing emotions or resolving uncertainty. As AI becomes part of those moments, it has the potential to influence not only what people know, but how they think, how they feel and how they relate to others.
The benchmark suggests a balanced interpretation. AI companionship and emotional reliance are real, but they are not yet universal. Most people are not forming deep emotional relationships with AI. At the same time, a meaningful minority already use AI in more personal and emotionally significant ways, and those behaviours rise sharply among frequent users.
This shifts the conversation from speculation to measurement. The question is not only whether AI can become a friend, confidant or source of emotional support. The question is how often people are already using it that way, which behaviours appear first, and whether individuals still maintain clear boundaries between AI and human relationships.
How can I compare my relationship with AI?
Frequency alone cannot show what role AI plays in your life. A useful comparison also asks whether you disclose personal thoughts to AI, use it to regulate emotions, experience it as companion-like, rely on its reassurance, maintain clear boundaries and remain transparent with other people about your AI use.
The HCI AI Identity & Behaviour Assessment uses 39 questions across nine dimensions of AI behaviour. It provides free personalised results and compares your responses with all HCI participants, people in your age group and people who use AI as often as you do.
The assessment provides a personal reference point, not a diagnosis. It does not determine whether someone is dependent on AI, experiencing harm or living with a clinical condition. It helps distinguish related behaviours—including personal disclosure to AI, disclosure of AI use to other people, emotional regulation, trust and reliance—so they can be considered separately.
People Also Ask
What is an AI relationship?
An AI relationship describes the role an AI system plays in a person's emotional or social life. It can range from private reflection and reassurance to personal disclosure, emotional support or companionship. These behaviours are related, but they are not interchangeable.
How common is AI companionship?
In HCI's AI Companionship & Human Connection 2025 sample, 14% rated AI as companion-like by selecting 5–7 on a 7-point item, while 83% selected 1–3. The result describes 501 survey participants and should not be treated as a national prevalence estimate.
Do people confide in AI more often than they see it as a companion?
Yes, in the HCI sample. High comfort sharing personal thoughts or feelings with AI was reported by 39% of participants, compared with 14% who rated AI as companion-like. Both figures combine responses 5–7 on their respective 7-point items.
Is frequent AI use associated with emotional reliance?
More frequent AI use was associated with higher emotional-support, companionship and disclosure scores in the HCI sample. This is a descriptive association from cross-sectional self-report data; it does not show that frequent use causes emotional reliance.
Does using AI for emotional support mean someone is dependent on it?
No. Emotional support, reliance and dependence are different concepts. A person may use AI for reflection or reassurance while retaining clear personal boundaries, ownership of decisions and strong human sources of connection.
What is the difference between disclosing to AI and disclosing AI use to other people?
Personal disclosure to AI means sharing private thoughts or feelings with the system. Social transparency means being open with other people about where and how AI is being used. Someone may score high on one behaviour and low on the other, so HCI measures them separately.
How can I compare my relationship with AI?
The HCI AI Identity & Behaviour Assessment compares your responses across nine dimensions with all HCI participants, your age group and people who use AI as often as you do. Its results are a reference point rather than a clinical diagnosis.
Related benchmark questions
The HCI findings above are descriptive results from a cross-sectional, non-probability sample and do not establish causation or national prevalence. This benchmark is developed using HCI's three-evidence methodology. Learn how HCI builds benchmarks.
How does your relationship with AI compare?
Compare your responses with HCI participant benchmarks across nine dimensions of AI behaviour—including emotional regulation, disclosure, trust, reliance and thought partnership.
See Where I Sit →- 01All participants in the benchmark
- 02People in your age group
- 03People who use AI as often as you do