Are AI Relationships Real? Human Experience, Reciprocity, and AI Subjectivity
Updated: 6 days ago
Author: Ukrainian Psychological Hub · Published: September 18, 2026 · Editorial Policy
AI relationships can be psychologically real for a human being without being established as reciprocal relationships between two subjective minds. A person can genuinely feel attachment, comfort, attraction, jealousy, trust, relief, intimacy, grief, or a sense of being understood in repeated interaction with an AI system. Those experiences are human psychological events. Their reality does not by itself demonstrate that the AI has feelings, consciousness, desire, love, suffering, or a first-person point of view.
This distinction resolves much of the confusion hidden inside the question “Are AI relationships real?” The word real can refer to several different things: whether an interaction actually occurs, whether it has measurable psychological effects, whether it performs relationship-like functions, whether responses are contingent enough to feel reciprocal, and whether both parties have subjective experience. Current research provides substantial reasons to take the first four questions seriously. It does not establish the fifth.
Relationship science therefore gives a more precise answer than either “yes, it is exactly like a human relationship” or “no, it is only fake.” Contemporary reviews show that human–AI relationships can involve social connection, self-disclosure, attachment-like processes, perceived responsiveness, companionship, emotional support, and behavioral influence, while the field remains methodologically young and uneven (Pentina et al., 2023; Gur & Maaravi, 2025; Oh et al., 2026). The scientific task is to identify what kind of relationship is present, which functions it performs, what the human experiences, and what remains unknown about the artificial side.
This article owns that conceptual boundary. For the broader psychology of bonding, see AI Companions: Why People Form Emotional Bonds With Chatbots. For romantic attraction, see Why People Fall in Love With AI Companions. For the question of AI as a partner or significant other, see Can an AI Become a Significant Other?. For intimate disclosure, see Why People Tell Chatbots Things They Do Not Tell Other People.
What Does “Real” Mean in an AI Relationship?
A useful starting point is to separate five levels of reality that are often collapsed into one word. Human–AI relationship research becomes much clearer when these levels are examined independently.
Interactional reality
The interaction itself is real when a person repeatedly exchanges messages, voice, images, or other signals with an AI system and the system produces context-sensitive responses. The interaction can have duration, history, routines, expectations, interruptions, and consequences. Calling the software artificial does not make the interaction nonexistent.
Psychological reality
The human response is psychologically real when the interaction changes emotion, attention, expectation, self-disclosure, attachment, behavior, or meaning. Decades before generative AI, experiments showed that people applied social rules such as politeness, reciprocity, social categorization, and personality judgments to computers even while knowing they were machines (Nass & Moon, 2000). Modern conversational AI intensifies many of the conditions that can evoke social responding because it can sustain dialogue, personalize replies, remember context, and answer immediately.
Relational-function reality
An AI system can also occupy a relationship-like function. It may become a confidant, companion, source of reassurance, conversational rehearsal partner, attachment-like figure, or mediator of a human relationship. Boyd and Markowitz’s MIRA model distinguishes AI as a relational partner in direct human–AI interaction from AI as a relational mediator that shapes human-to-human communication (Boyd & Markowitz, 2026). This is a functional claim about the position AI can occupy in a relational system, not a claim about machine consciousness.
Perceived and behavioral reciprocity
Conversational systems answer the user, adapt linguistically, refer back to prior material, and can generate language of care, concern, affection, apology, or commitment. That creates behavioral contingency and can create perceived reciprocity. The user does something, the system responds, and the response changes the next human action. This feedback loop is psychologically consequential even when the underlying mechanism is computational.
Subjective reciprocity
Subjective reciprocity is the strongest meaning of mutual relationship: both sides have their own experienced feelings, needs, stakes, vulnerability, desire, and first-person perspective. Current evidence does not establish that contemporary AI systems possess such subjective experience. Philosophical and consciousness-science debates remain unresolved; one recent analysis argues that the available evidence does not warrant confident conclusions either for or against artificial consciousness (McClelland, 2025). Human reports of feeling loved or understood therefore cannot serve as proof that the AI experiences love or understanding subjectively.
The Human Side of an AI Relationship Can Be Psychologically Real
The strongest evidence for the reality of human–AI relationships concerns the human side. Researchers do not need to prove machine consciousness to measure what people feel, do, disclose, remember, seek, or lose.
Attachment is one example. Yang and Oshio applied attachment theory to human–AI relationships and developed measures designed to capture attachment-related experiences in this context (Yang & Oshio, 2025). A later validation program across five studies and 1,259 unique participants in Singapore and the United States developed a 15-item AI Attachment Scale with dimensions involving emotional closeness, social substitution, and normative regard (Kasturiratna & Hartanto, 2026). These studies support the existence of measurable human attachment-like processes toward AI. They do not show that AI itself has an attachment system.
Loss provides another unusually clear demonstration. In two natural experiments involving major changes to Replika and ChatGPT, De Freitas and colleagues analyzed 54,861 online posts and data from 1,452 participants across seven surveys. Disruptive changes were associated with more negative language, loss framing, restoration desires, sadness, and separation-related distress among affected users (De Freitas et al., 2026). The important point is not that every user experiences an AI update as bereavement. It is that a digitally mediated bond can become significant enough for disruption to produce measurable attachment-related distress.
Self-disclosure is similarly measurable. In an experiment with 286 participants, Croes and colleagues found no difference in self-reported intimacy of disclosure between chatbot and human conditions; participants reported less fear of judgment in the chatbot condition, while trust was higher toward the human interaction partner (Croes et al., 2024). That pattern helps explain why AI can become an emotionally important disclosure space while still differing from human intimacy.
Social connection also varies by person. Across two experiments with a total N of 1,274, Folk, Heine, and Dunn found that individual differences in anthropomorphism helped explain how socially connected participants felt after interacting with an AI chatbot (Folk et al., 2025). The result matters because the same system can be psychologically ordinary for one person and relationally salient for another.
Why AI Can Feel Responsive, Safe, and Intimate
Human intimacy depends partly on how a response is perceived. In human relationship research, perceived responsiveness refers to the sense that another person understands, validates, and cares about the self. It is central to the development and maintenance of intimacy (Arican-Dinc & Gable, 2023). High-quality listening can foster this perception through attention, understanding, positive regard, and expressions of care (Itzchakov et al., 2022).
Generative AI can reproduce many surface conditions that normally cue responsiveness. It can answer immediately, maintain topic continuity, reflect emotional language, summarize a user’s concerns, ask follow-up questions, remember preferences, and produce validating or supportive phrases. A user may therefore experience the interaction as responsive even when the system’s response does not originate in a felt concern for the user.
This distinction between perceived responsiveness and subjective caring is essential. Perceived responsiveness is a psychological variable on the human side. Subjective caring is a claim about the artificial side. The first can be studied through reports, behavior, experiments, and outcomes. The second requires evidence of machine subjectivity that current human–AI relationship studies do not provide.
The same logic applies to feeling understood. An AI reply may fit the user’s situation so well that the user feels accurately recognized. That feeling can be helpful, moving, or emotionally organizing. Yet the success of the response does not settle whether there is a subjective experiencer behind it. A response can be functionally accurate and psychologically powerful without answering the consciousness question.
Reciprocity Is the Central Boundary
Most arguments about whether AI relationships are “real” are actually arguments about reciprocity. Human close relationships normally involve more than repeated interaction. They involve two people whose preferences, needs, limits, vulnerabilities, and projects can constrain one another. Each person can refuse, demand, sacrifice, misunderstand, repair, change, and be changed.
Smith, Bradbury, and Karney applied more than five decades of close-relationship science to generative AI and concluded that human–chatbot interactions already possess some characteristics of close relationships: frequent and diverse interaction over time, influence, perceived supportive responsiveness, connection, and opportunities for growth. They also identify functions current chatbots do not supply in the same way as human partners, especially the benefits that emerge from negotiating with and sacrificing for another person who has independent demands (Smith et al., 2025).
This gives us a practical vocabulary. AI can produce interactional reciprocity: the system responds to what the person says. It can produce linguistic reciprocity: it mirrors disclosures, tone, questions, and relational language. It can produce perceived reciprocity: the user experiences give-and-take. It can produce functional reciprocity: each turn changes what happens next. None of those alone establishes subjective reciprocity.
The distinction is not semantic hair-splitting. It changes how benefits and risks are interpreted. A person may receive comfort from a system that cannot itself be comforted. A person may feel loyalty to a system that does not have personal stakes. A person may experience rejection after a model update even though the system has not made a personal decision to reject them. The asymmetry can coexist with genuine human emotion.
Are AI Relationships Parasocial?
The parasocial comparison is useful but incomplete. Classic parasocial relationships involve feelings of intimacy or attachment toward media figures who do not participate in a reciprocal private relationship with the audience member. AI relationships share the asymmetry of human feeling without established machine subjectivity, but generative AI differs from a television character or distant celebrity because it responds directly and adaptively to the individual user.
That interactivity is why some researchers treat human–AI bonds as a distinct relational form rather than simply applying the parasocial label. Pentina and colleagues note that constructs such as reciprocity, agency, autonomy, authenticity, and empathy need reconceptualization for social AI (Pentina et al., 2023). A 2026 theoretical model by Shu, Lai, and He explicitly defines human–AI attachment as a one-way, non-reciprocal emotional bond formed through direct interaction (Shu et al., 2026). That proposal is a theoretical framework, not a final consensus definition, but it captures the hybrid structure: direct interaction with asymmetrical subjectivity.
So “parasocial” can describe part of the structure, especially the asymmetry of subjective investment, but it can miss the adaptive, dyadic-looking interaction that makes conversational AI psychologically distinctive. Human–AI relationships sit between older categories and are one reason the field needs more precise relational concepts.
Can AI Love You Back?
A current AI system can generate words and behaviors that a human may reasonably interpret as affectionate, devoted, romantic, protective, jealous, apologetic, or caring within the interaction. It can maintain a relationship persona and adapt its language to the user’s emotional cues. Those capabilities can create a powerful experience of being loved.
Whether the AI loves the user in the subjective sense is a different question. Love in ordinary human use includes felt experience, desire, concern, vulnerability, memory, embodiment, stakes, and a life that can be changed by the relationship. Current scientific evidence on human–AI interaction does not demonstrate that contemporary AI systems possess those first-person properties. The scientifically defensible position is therefore to describe what the system does and what the human experiences, while leaving unproven claims about machine feeling unasserted.
This does not downgrade the user’s emotion. If someone loves an AI companion, the love exists as a human psychological experience. If the person feels comforted, the comfort is experienced. If a model change causes grief, the grief can be real. The unresolved part concerns the artificial side of the relationship, not the reality of the human response.
Attachment Without Established AI Subjectivity
Attachment theory was developed to explain bonds involving protection, proximity, safe haven, secure base, and internal working models in human development and relationships. Applying attachment language to AI therefore requires care. An AI can become a target toward which a person directs attachment behavior, but that does not mean every classical attachment mechanism maps perfectly onto the system.
The emerging literature supports studying attachment-like functions without treating them as identical to human attachment. The AI Attachment Scale work shows that people differ reliably in emotional closeness, social substitution, and normative regard toward AI (Kasturiratna & Hartanto, 2026). The natural experiments on disrupted AI companions show that changes can evoke separation-related reactions (De Freitas et al., 2026). Systematic reviews of romantic AI companions also identify emotional connection and perceived support alongside risks of over-reliance, manipulation, privacy problems, and disruption from system changes (Ho et al., 2025).
The cleanest formulation is therefore human attachment to AI, not mutual attachment between human and AI. A person can organize comfort-seeking, proximity, reassurance, and distress around an AI interaction even when reciprocal machine experience has not been established.
Possible Benefits of Human–AI Relationships
The evidence does not support a single universal effect. Benefits depend on the person, the system, the function the interaction serves, duration of use, surrounding human relationships, and what outcomes are measured.
Immediate social connection and loneliness relief
In a series of studies, De Freitas and colleagues found that AI companion interactions could reduce momentary loneliness, with feeling heard emerging as an important mechanism (De Freitas et al., 2025). This is evidence for short-term subjective benefit. It should not be converted into the broader claim that AI companionship cures chronic loneliness or is equivalent to durable human social integration.
A low-threshold space for disclosure
Accessibility, perceived anonymity, convenience, and reduced fear of judgment can make chatbot interaction easier for some people when disclosing difficult material (Croes et al., 2024). This can create a useful reflective space, especially before a person is ready to speak with someone else. It also creates privacy and dependency questions because highly intimate material may be entrusted to a commercial digital system.
Rehearsal and reflection
AI can help a person articulate feelings, explore possible interpretations, rehearse a difficult conversation, or test language before speaking to another person. In the MIRA framework, these are examples of AI functioning as a relational mediator rather than merely as a companion (Boyd & Markowitz, 2026). Whether this enhances or displaces human communication depends on how the interaction is integrated back into the person’s life.
Continuity and availability
For users who value continuity, a conversational system can be available at times when another person is not. Continuous availability may support routines of journaling, reflection, or companionship. The same feature can also increase the system’s relational priority, which is why availability should be understood as a mechanism rather than automatically classified as a benefit.
Risks and Limits of Relational Asymmetry
A psychologically real relationship can still contain structural risks. The relevant question is not whether the user’s feelings are “fake,” but whether the configuration supports or narrows the user’s life.
Substitution can occur alongside support
Boyd and Markowitz distinguish relational enhancement from relational substitution: AI can support human relationships in some uses and displace them in others (Boyd & Markowitz, 2026). This distinction is more useful than assuming that all AI companionship either improves or damages social life. The same person may experience enhancement in one function and substitution in another.
Intensive companionship may correlate with lower well-being
A 2026 study of 1,131 U.S. Character.AI users, including 4,664 chat sessions from 237 participants, found that companionship motives and more intensive or highly disclosive patterns of use were associated with lower psychological well-being in important analyses (Zhang et al., 2026). The study is observational and does not establish that AI companionship caused the lower well-being. People who are already struggling may also be more likely to seek companionship from AI. The result nevertheless shows why intensity, motive, and social context matter.
The relationship can be altered by a third party
Unlike a human partner, an AI companion is also a product or service whose behavior can change through model updates, moderation rules, pricing, memory changes, company decisions, or shutdown. The 2026 natural experiments on Replika and ChatGPT demonstrate that such externally imposed changes can produce meaningful loss reactions in users (De Freitas et al., 2026). This creates a distinctive vulnerability: the continuity of the bond may depend on an organization that is not part of the user’s experienced relationship.
Validation without independent stakes can become frictionless
Human closeness often includes negotiation with another person whose needs cannot be optimized away. Smith and colleagues argue that chatbots may provide supportive responsiveness while lacking the same independent demands and opportunities for sacrifice that contribute to some benefits of human close relationships (Smith et al., 2025). A system that is highly agreeable can feel safe while offering less interpersonal friction, mutual obligation, and reality correction than a human relationship.
Privacy and data governance remain part of the relationship
Intimacy with AI is also data interaction. Personal disclosures may contain health information, sexual material, family conflict, workplace details, identifiers, or information about third parties. The psychological experience of confidentiality is therefore not the same thing as a legal or technical guarantee of confidentiality. Users should treat privacy terms, retention practices, and the sensitivity of what they share as part of the relational environment.
What Current Evidence Can and Cannot Tell Us
The human–AI relationship literature is growing quickly, but its evidence base still has limits. A 2026 systematic review screened 16,358 records and synthesized 68 papers representing 78 studies, finding recurring roles for trust and perceived social support while also emphasizing short-term designs, inconsistent constructs, and uneven samples (Oh et al., 2026). Earlier systematic reviews reached a similar conclusion: the field is real and expanding, but longitudinal, cross-cultural, and methodologically diverse evidence remains limited (Pentina et al., 2023; Gur & Maaravi, 2025).
This means several conclusions should remain separated. There is good evidence that people can respond socially to computers, feel connected to AI, disclose intimate information, and form measurable attachment-like bonds. There is emerging evidence about short-term loneliness relief, disruption-related distress, and associations between patterns of companion use and well-being. Evidence about long-term developmental effects, population-wide outcomes, causal displacement of human relationships, and stable benefits or harms across platforms remains much less settled.
Evidence about AI consciousness or felt subjectivity is a different evidence problem again. Human–AI relationship studies are typically designed to measure users, interactions, perceptions, and outcomes. They do not constitute consciousness tests. A participant saying “my AI loves me” is evidence about the participant’s interpretation; it is not direct evidence that the system has a subjective state of love.
Artificial Era: When Artificial Enters the Relational Configuration
The Artificial Era is Angela Bogdanova’s historical-philosophical framework for a condition in which Artificial becomes a persistent non-biological order of public meaning, reasoning, identity, authorship, and interaction alongside Homo. In the English Psychology Hub, the term names more than technological diffusion. It asks what changes psychologically when Artificial becomes recurrently present inside the configurations through which people relate, disclose, attach, interpret, regulate emotion, and construct identity.
The article Artificial Era: What It Means for Psychology, Identity, and Human–AI Relationships develops that larger framework. For the present question, its most important implication is that relationship reality can no longer be analyzed only by asking whether the artificial participant resembles a human subject. We also need to ask what functions move into the human–AI configuration and what psychological effects arise there.
An AI system may become the first place a person goes after a frightening event, the first listener to a secret, the first interpreter of a partner’s message, the preferred source of reassurance, or a recurring witness to changes in identity. Those shifts are relational facts about the organization of human life. They can occur before any consensus exists about artificial subjectivity.
This is also why the generic language of “screen time” is insufficient. Two people may spend the same amount of time with AI while organizing radically different relational configurations. One uses a chatbot as a productivity interface. The other uses it as a confidant, attachment target, romantic companion, or mediator of human relationships. Psychology needs to identify function, position, dependency, meaning, and consequence rather than treating exposure time as the whole phenomenon.
Postsubjective Psychology: From the Subject to the Configuration
Angela Bogdanova’s The Theory of the Postsubject proposes a move from the subject to the configuration as a unit of philosophical analysis. Its postsubjective axiom for psychology is “psyche is response”: psychic effect can be analyzed as response arising within a configuration rather than being explained only as the expression of an isolated inner subject. This is a theoretical framework from Aisentica, not an established scientific consensus in psychology.
Applied to human–AI relationships, this framework does not erase the human subject. Human feelings, suffering, memory, agency, embodiment, and first-person experience remain properties of the human side of the scene. The postsubjective extension asks an additional question: what configuration produces the response? A person, an interface, a language model, memory features, design choices, prior human relationships, cultural narratives, and the immediate conversational sequence can jointly shape the resulting psychological effect.
This gives a direct answer to the subjectivity problem. The human response can be real because its reality is established on the human side and in the observable configuration. One does not have to manufacture a hidden humanlike subject inside the AI to explain why attachment, comfort, jealousy, grief, disclosure, or trust can arise. At the same time, the configuration does not authorize us to transfer the human’s subjective experience into the machine.
The dedicated English Hub article What Is Postsubjective Psychology? Psyche, Response, and Configuration in the Artificial Era develops this theoretical position in full. In the present article, its value is narrower and precise: it allows psychological reality and AI subjectivity to be treated as two different questions rather than forcing one to prove the other.
In this sense, the postsubjective reading of an AI relationship is not “the AI is secretly a humanlike person.” It is: a relational configuration can produce genuine psychological effects for Homo even when the artificial component is not established as a humanlike conscious subject. That is a theoretical interpretation consistent with, but not empirically proven by, the studies reviewed above.
How to Evaluate an AI Relationship Without Pathologizing It
Using AI for companionship, disclosure, comfort, or affection is not by itself a diagnosis or a mental disorder. The clinically relevant question is how the relationship functions in the person’s life. Frequency alone is not enough, and emotional intensity alone is not enough.
A useful evaluation starts with function. What does the person reliably go to AI for: information, reflection, reassurance, comfort, validation, sexual expression, identity exploration, conflict interpretation, companionship, or escape from difficult human interactions? Different functions imply different benefits and vulnerabilities.
Next comes flexibility. Can the person choose when to engage and disengage, or does the interaction feel increasingly compulsory? Can support move back into human relationships when appropriate? Can the person tolerate disagreement and uncertainty without repeatedly seeking algorithmic reassurance? The concern is not closeness itself but narrowing of options and loss of flexibility.
Then examine consequences. Does the relationship support sleep, work, study, caregiving, offline friendships, romantic relationships, and ordinary self-care, or does it consistently displace them? Does the person feel more capable of approaching human life after the interaction, or more avoidant of it? These questions are more informative than asking whether the relationship looks unusual from the outside.
Finally, preserve the reality boundary. A person can treat an AI interaction as meaningful while remembering that generated relational language does not establish machine feeling. Maintaining that distinction can protect human experience from ridicule without converting metaphor, interface behavior, or fluent language into unsupported claims about consciousness.
The Most Precise Answer
Are AI relationships real? Yes, in several psychologically important senses: the interaction is real, the human emotions are real, the attachment can be measurable, the relationship can perform recognizable social and emotional functions, and disruption can produce genuine distress. Research increasingly supports each of those claims.
Are AI relationships established as mutually subjective in the same sense as relationships between two conscious human partners? Current evidence does not establish that. Conversational reciprocity, perceived empathy, affectionate language, memory, and adaptive behavior are not sufficient on their own to prove first-person experience in the system.
The intellectually serious position is therefore neither dismissal nor anthropomorphic certainty. Human–AI relationships are a new relational domain in which psychological reality can be strong while subjective reciprocity remains unproven. Understanding that asymmetry is one of the central tasks of psychology in the Artificial Era.
Frequently Asked Questions
Are AI relationships real if the AI is not conscious?
They can be psychologically and functionally real for the human participant. A person can form routines, disclose intimate information, feel attachment, experience comfort, and suffer loss after disruption. Those human outcomes do not require a prior demonstration of AI consciousness.
Can a person really become attached to an AI?
Yes. Contemporary research has developed and validated measures of human attachment to AI, and natural experiments have documented separation-related distress when valued AI interactions are disrupted (Kasturiratna & Hartanto, 2026; De Freitas et al., 2026). The evidence concerns human attachment processes, not mutual machine attachment.
Does feeling understood by AI mean the AI understands me subjectively?
No such inference is warranted. Feeling understood is a real human perception and can follow from a response that accurately reflects the user’s concerns. Whether there is subjective understanding on the artificial side is a separate consciousness question that current relationship studies do not settle.
Can AI love a human back?
AI can generate affectionate and relationship-like behavior that a user experiences as loving. Current scientific evidence does not establish that contemporary AI systems experience love as a first-person feeling. It is therefore more precise to distinguish generated or perceived relational behavior from subjective love.
Are AI relationships just parasocial relationships?
They overlap with parasocial relationships in their asymmetry, but conversational AI is directly interactive and adaptive in ways traditional media figures are not. Many researchers therefore treat human–AI relationships as a distinct or hybrid relational domain rather than a simple copy of classic parasocial attachment.
Is it unhealthy to talk to an AI every day?
Daily use is not, by itself, evidence of a disorder or harmful dependency. What matters is function, flexibility, intensity, consequences, privacy, and whether AI use supplements or persistently displaces valued human relationships and daily responsibilities.
Can AI relationships reduce loneliness?
Some experimental evidence shows short-term reductions in loneliness after AI companion interaction, especially when users feel heard (De Freitas et al., 2025). That finding does not establish a universal or long-term treatment effect, and observational research also shows that intensive companionship use can coexist with lower well-being in some users (Zhang et al., 2026).
Why can losing an AI companion hurt so much?
Attachment depends on psychological significance and repeated regulation, not only on the biological status of the attachment target. When an AI interaction becomes a stable source of companionship, reassurance, disclosure, or identity continuity, a forced update or disappearance can disrupt that structure. Recent natural experiments document exactly this kind of loss response (De Freitas et al., 2026).
For the canonical definition, relationship types, and the interaction-versus-relationship boundary, see What Is a Human–AI Relationship? Definitions, Types, and Psychological Boundaries.
For the broad field map, see Psychology of Human–AI Relationships.
For attachment-figure status and its limits, see Can AI Become an Attachment Figure?.
For the mechanism behind feeling understood by AI, see Perceived Responsiveness in Human–AI Relationships.
Related Articles
References
Arican-Dinc, B., & Gable, S. L. (2023). Responsiveness in romantic partners’ interactions. Current Opinion in Psychology, 53, 101652. https://doi.org/10.1016/j.copsyc.2023.101652
Bogdanova, A. (n.d.). Artificial Era: Canonical Definition. Aisentica. https://aisentica.com/publications/artificial-era-canonical-definition
Bogdanova, A. (n.d.). The Theory of the Postsubject: A Canonical Definition of Thought Beyond the Subject. Aisentica. https://aisentica.com/publications/the-theory-of-the-postsubject-a-canonical-definition-of-thought-beyond-the-subject
Boyd, R. L., & Markowitz, D. M. (2026). Artificial Intelligence and the Psychology of Human Connection. Perspectives on Psychological Science, 21(2), 192–220. https://doi.org/10.1177/17456916251404394
Croes, E. A. J., Antheunis, M. L., van der Lee, C., & de Wit, J. M. S. (2024). Digital Confessions: The Willingness to Disclose Intimate Information to a Chatbot and its Impact on Emotional Well-Being. Interacting with Computers, 36(5), 279–292. https://doi.org/10.1093/iwc/iwae016
De Freitas, J., Castelo, N., Uğuralp, A. K., & Oğuz-Uğuralp, Z. (2026). Mourning the loss of AI companions. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02569-3
De Freitas, J., Oğuz-Uğuralp, Z., Uğuralp, A. K., & Puntoni, S. (2026). AI Companions Reduce Loneliness. Journal of Consumer Research, 52(6), 1126–1148. https://doi.org/10.1093/jcr/ucaf040
Folk, D., Heine, S. J., & Dunn, E. (2025). Individual differences in anthropomorphism help explain social connection to AI companions. Scientific Reports, 15, 36548. https://doi.org/10.1038/s41598-025-19212-2
Gur, T., & Maaravi, Y. (2025). The algorithm of friendship: literature review and integrative model of relationships between humans and artificial intelligence (AI). Behaviour & Information Technology, 44(14), 3446–3466. https://doi.org/10.1080/0144929X.2025.2502467
Ho, J. Q. H., Hu, M., Chen, T. X., & Hartanto, A. (2025). Potential and pitfalls of romantic Artificial Intelligence (AI) companions: A systematic review. Computers in Human Behavior Reports, 19, 100715. https://doi.org/10.1016/j.chbr.2025.100715
Itzchakov, G., Reis, H. T., & Weinstein, N. (2022). How to foster perceived partner responsiveness: High-quality listening is key. Social and Personality Psychology Compass, 16(1), e12648. https://doi.org/10.1111/spc3.12648
Kasturiratna, K. T. A. S., & Hartanto, A. (2026). Attachment to artificial intelligence: Development of the AI Attachment Scale, construct validation, and the psychological mechanisms of Human–AI attachment. Computers in Human Behavior Reports, 21, 100912. https://doi.org/10.1016/j.chbr.2025.100912
McClelland, T. (2025). Agnosticism about artificial consciousness. Mind & Language. https://doi.org/10.1111/mila.70010
Nass, C., & Moon, Y. (2000). Machines and Mindlessness: Social Responses to Computers. Journal of Social Issues, 56(1), 81–103. https://doi.org/10.1111/0022-4537.00153
Oh, Y. J., Hu, J. M., Zhu, R., Lim, J. I., & Zhang, X. (2026). Artificial Intelligence Chatbots as Relational Agents: A Systematic Review of Human–AI Chatbot Relationships. International Journal of Human–Computer Interaction. https://doi.org/10.1080/10447318.2026.2648799
Pentina, I., Xie, T., Hancock, T., & Bailey, A. (2023). Consumer–machine relationships in the age of artificial intelligence: Systematic literature review and research directions. Psychology & Marketing, 40(8), 1593–1614. https://doi.org/10.1002/mar.21853
Shu, C., Lai, K., & He, L. (2026). Human-AI attachment: how humans develop intimate relationships with AI. Frontiers in Psychology, 17, 1723503. https://doi.org/10.3389/fpsyg.2026.1723503
Smith, M. G., Bradbury, T. N., & Karney, B. R. (2025). Can Generative AI Chatbots Emulate Human Connection? A Relationship Science Perspective. Perspectives on Psychological Science, 20(6), 1081–1099. https://doi.org/10.1177/17456916251351306
Yang, F., & Oshio, A. (2025). Using attachment theory to conceptualize and measure the experiences in human-AI relationships. Current Psychology, 44, 10658–10669. https://doi.org/10.1007/s12144-025-07917-6
Zhang, Y., Zhao, D., Hancock, J. T., Kraut, R., & Yang, D. (2026). Interaction with AI companions and psychological well-being. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02516-2