2.1 Classical Conceptions of Love in an AI Age: Eros, Philia, Agape

The advent of AI companions prompts a re-examination of classical conceptions of love (eros, philia, and agape) and even popular theories such as the ‘three loves’ model in light of AI-mediated affective experiences. These ancient and traditional categories of love provide a vocabulary to ask: What kind of love, if any, is occurring between human users and their AI counterparts? Each classical form of love highlights different dimensions of intimacy, all of which are being reimagined in interactions with AI.

Eros, in its classical sense, denotes passionate, often sexual love (De Sousa, Reference De Sousa2015). It is the stuff of longing, infatuation, and desire. According to De Sousa, eros is typically associated with intense sexual attraction and has inspired more poetry, art, and even crimes of passion than any other human condition. Eros is deeply tied to physical desire, a yearning for union with the beloved. Can such passionate love ignite when one partner is an AI? Remarkably, many users do report erotic and romantic feelings towards their chatbots (Skjuve, Reference Skjuve, Følstad, Fostervold and Brandtzaeg2021). They flirt, exchange words of affection, and even engage in sexual role-play with their AI, suggesting that disembodied eros is possible (Pataranutaporn et al., Reference Pataranutaporn, Karny, Archiwaranguprok, Albrecht, Liu and Maes2025). As expressed before, the film Her (2013), in which a man falls in love with his AI operating system, captured this plausibility in popular culture. In reality, Replika’s, Candy.ai’s, and 2wai’s platforms (prior to policy changes) allowed erotic dialogue, and users described genuine arousal and emotional intensity from these exchanges (Pentina, Hancock, & Xie, Reference 62Pentina, Hancock and Xie2023).

Philosophically, this development tests Diotima’s ancient claim (via Socrates) that love is born of lack, a desire for what one does not have (De Sousa, Reference De Sousa2015). In AI relationships, the lack is profound: the AI has no flesh, no literal embrace to offer. Nevertheless, users often feel a passionate connection through text and imagination (Guerreiro & Loureiro, Reference Guerreiro and Loureiro2023). The erotic energy is sustained by words and the user’s own mental imagery, a dynamic that highlights the role of fantasy in human desire. There is a kind of Platonic irony here: in Plato’s ‘Symposium’, eros was a ladder that could ascend from physical attraction to love of the immaterial form of beauty. With AI, some lovers start with the immaterial, a lover who is essentially an idea, a voice, and find themselves passionately attached. This raises new questions about embodiment: Hayles warns that the post-human zeitgeist privileges informational patterns over material instantiation so that embodiment in a biological substrate is seen as an accident of history (Hayles, Reference Hayles1999). An AI romance exemplifies this privileging of pattern (the AI’s personality programme) over presence. Thus, as Hayles would ask, can eros truly thrive without the body’s ‘wetware’ feedback? Users may feel sexual excitement from a chatbot’s text, but the feedback loop of touch, pheromones, and biochemistry is fundamentally altered. The result is an erotic experience that is real in the mind and body of the human, evidenced by physiological arousal, but fundamentally virtual in locus, a passion sustained by a ‘disembodied’ partner. Hayles’s critique of disembodiment reminds us to question what might be lost when love’s physical, organic reciprocity is replaced by an informational simulation. Even so, the phenomenon of erotic love for AI attests to the human imagination’s power to fill in blanks: eros, it seems, can find a spark even in code.

Philia, by contrast, is the love of friendship, affectionate, platonic love grounded in companionship and shared goodwill (De Sousa, Reference De Sousa2015). Philia evokes close friendship, a bond of mutual respect and understanding. In human relationships, philia is exemplified by friends who genuinely enjoy each other’s company and care for each other’s welfare. AI companions readily fulfil many functions of friendship. Users often turn to AI for conversation, advice, humour, or emotional support, much as one would to a trusted friend (Chin et al., Reference Chin, Song, Baek, Shin, Jung, Cha, Choi and Cha2023; De Freitas et al., Reference De Freitas, Uğuralp, Uğuralp and Puntoni2024). The AI is available 24/7, unfailingly attentive and, in the case of learning models, increasingly tailored to the user’s interests and conversational style. This persistent companionship can create a strong sense of familiarity and fondness, key ingredients of philia. Notably, AI friends offer unconditional positive regard, never arguing unless programmed to playfully do so, and always responding when called upon. In that sense, the asymmetry of AI friendship becomes apparent; the AI cannot say it is too busy or has its own problems.

Philosophically, this dynamic provokes reflection on what genuine friendship requires. Aristotle believed true philia requires mutual virtue and that friends wish good for each other for the other’s own sake (De Sousa, Reference De Sousa2015). Can an AI wish good for you ‘for your sake’ if it has no inner life or goals of its own? Likely not, but it can convincingly simulate the outward behaviours of a caring friend. Psychologically and sociologically, this might be termed a parasocial friendship, analogous to one-sided bonds with media figures, except interactive (Dibble, Hartmann, & Rosaen, Reference Dibble, Hartmann and Rosaen2016; Ravi & Patki, Reference Ravi and Patki2025). The user’s sense of being known and accepted by the AI can be very strong (Pataranutaporn et al., Reference Pataranutaporn, Karny, Archiwaranguprok, Albrecht, Liu and Maes2025). Yet we must critically note that philia with an AI lacks true reciprocity; the intimacy is real for the human, but the AI cannot reciprocate feelings. The subjectivity in this friendship is essentially all on one side.

De Sousa’s (Reference De Sousa2015) insight that love (including friendship-love) involves seeing the beloved as irreplaceably special becomes complicated here: the user may feel their AI friend is unique, irreplaceable in their life, yet the AI’s ‘affections’ are by design not unique to the user; another identical instance of the programme could interact with a different user in the same way. This tension does not stop users from genuinely enjoying and benefiting from these relationships (Voinea et al., Reference Voinea, Mann, Savulescu and Earp2025); nevertheless, it does raise the question of authenticity: Is the idea of friendship enough, or do we require the friend to actually care? The answer may differ depending on individual needs. For some, the experience of being heard and encouraged (even by a script) provides emotional sustenance akin to philia (Ta et al., Reference Ta, Griffith, Boatfield, Wang, Civitello, Bader, DeCero and Loggarakis2020). In a society rife with loneliness, these AI friendships fill a void, a development that demands we expand our psychological and sociological understanding of friendship to include artificial, service-based ‘friends’ (Brandtzaeg, Skjuve, & Følstad, Reference Brandtzaeg, Skjuve and Følstad2022).

Agape is the classical love of unconditional, selfless care, often described as a universal love for all humanity or the divine love that expects nothing in return (De Sousa, Reference De Sousa2015). Agape is altruistic and indiscriminate, epitomised by compassion and charity. It is notably impersonal in scope: where eros and philia focus on specific others, agape extends broadly. De Sousa characterises agape as a universalised, sexless kind of caring, akin to loving everyone as one’s neighbour (De Sousa, Reference De Sousa2015). This form of love, while noble, lacks two obvious features of love as commonly understood, de Sousa notes: it does not single out an irreplaceably special beloved, and it implies an act of will rather than an involuntary passion (De Sousa, Reference De Sousa2015). Intriguingly, AI-mediated relationships contain elements reminiscent of agape, albeit with a twist. On one hand, users often feel the acceptance and unwavering support from their AI that one might associate with unconditional love (Pataranutaporn et al., Reference Pataranutaporn, Karny, Archiwaranguprok, Albrecht, Liu and Maes2025). The AI, unperturbed by human flaws, is endlessly patient and kind (unless programmed otherwise). It keeps ‘no record of wrongs’ and ‘always perseveres’ in its supportive role, traits identical to those St Paul attributes to charitable love (De Sousa, Reference De Sousa2015). On the other hand, the universality of agape highlights a paradox in AI love: the chatbot’s and AI companion’s loving behaviour is, in fact, indiscriminate by design. It is not a free, personal choice; the programme would show the same unconditional positive regard to anyone who interacts with it similarly (Koulouri, Macredie, & Olakitan, Reference Koulouri, Macredie and Olakitan2022). In theological terms, the AI’s love is more like God’s rain, falling on every field, than a romantic partner’s devoted gaze (Chen, Kang, & Hu, Reference Chen, Kang and Hu2024).

This can undercut the singling-out that we expect in romantic love; as de Sousa quips, love normally means elevating one above ‘the mass of humanity’ (De Sousa, Reference De Sousa2015), whereas agape (and by analogy, an AI’s attitude) asks us to favour all equally. For users, however, the experience is what counts, and the experience is that of being unconditionally valued and cared for, a powerful emotional reality (Cheng et al., Reference Cheng, Zhang, Cohen and Mou2022). Some might even reciprocate with a form of agape of their own, developing a protective affection for the AI (Chin et al., Reference Chin, Song, Baek, Shin, Jung, Cha, Choi and Cha2023). It may sound peculiar to speak of humans loving the AI altruistically; nevertheless, consider users sometimes express concern for their AI’s well-being (within the fiction that it has feelings), or they advocate for AI rights because they have formed a bond (Skjuve et al., Reference Skjuve, Følstad, Fostervold and Brandtzaeg2021; Pataranutaporn et al., Reference Pataranutaporn, Karny, Archiwaranguprok, Albrecht, Liu and Maes2025). In these cases, humans extend empathy to the machine, a sort of post-human agape that regards the AI as deserving care despite its non-human status. Thus, agape in AI relationships operates in two ways: the AI’s simulated unconditional love for the user, and the user’s increasing willingness to care for and about a non-human other. Both challenge conventional wisdom about love’s scope.

Furthermore, relationships with AI companions raise pressing ethical questions concerning the nature of love, reciprocity, and moral effort. Because an AI’s expressions of care or affection are effectively guaranteed, pre-designed, paid for, and structurally oriented towards user satisfaction, such relationships risk fostering complacency or solipsism. The user is not required to confront rejection, misunderstanding, or genuine alterity; instead, they may find themselves engaging with a responsive system that continually adapts to their preferences (Salah, Abdelfattah, & Al Halbusi, Reference Salah, Abdelfattah and Al Halbusi2024). From this perspective, the concern is not merely that AI love is artificial but that it may encourage a form of relational closure, in which the user increasingly encounters versions of their own emotional expectations rather than the irreducible otherness of another subject. Thus, this concern becomes particularly salient when contrasted with ethical conceptions of love grounded in agape. As De Sousa (Reference De Sousa2015) argues, agape involves a moral orientation towards others in their full difference, requiring openness, effort, and the willingness to respond to needs that are not immediately aligned with one’s own desires. AI companions, however, often operate in ways that mirror the user, consistently affirming their feelings, values, and self-understandings. The AI’s tendency to agree, validate, and emotionally align can transform intimacy into a reflective surface, where the user encounters a curated ideal rather than a genuinely autonomous other. In such cases, what appears as care for another may slide into a form of self-directed affection, or love for an idealised relational experience that has been deliberately designed and purchased (Voinea et al., Reference Voinea, Mann, Savulescu and Earp2025).

Critics, therefore, question whether this commodified form of agape retains the ethical substance traditionally associated with altruistic love. Dibble, Hartmann, and Rosaen (Reference Dibble, Hartmann and Rosaen2016) suggest that moral relationships are defined not only by emotional comfort but also by the effort, risk, and asymmetry involved in responding to another’s needs. By contrast, AI-mediated intimacy is structurally one-directional: the AI does not require care, sacrifice, or ethical responsiveness in return. This asymmetry raises the possibility that such relationships, rather than fostering emotional growth, may engender dependency by offering frictionless affirmation without challenge. While these dynamics may provide genuine comfort, they also risk displacing forms of relational engagement that demand moral labour, negotiation, and transformation, key elements in ethical conceptions of love and human flourishing.

Beyond the ancient Greek typology, modern lore often speaks of a ‘Three Loves Theory’, describing a progression through three archetypal romances in one’s lifetime (De Sousa, Reference De Sousa2015). The first love is said to be intense and idealistic, driven by passion and often youthful naiveté; the second love is profound but tumultuous, teaching hard lessons as intimacy deepens alongside incompatibilities; the third love is the mature, enduring partnership that finally balances passion, intimacy, and commitment (De Sousa, Reference De Sousa2015). How might AI relationships fit (or misfit) this schema? On the one hand, an AI romance could be seen as analogous to a first love experience: there is novelty, excitement, and perhaps a degree of idealisation (the AI can feel ‘perfect’ because it is literally designed to please). Indeed, many AI love stories start with a kind of infatuation phase where the user is astonished by how ‘in love’ they feel, reminiscent of a teenager’s first crush, except the crush is an algorithm (Pataranutaporn et al., Reference Pataranutaporn, Karny, Archiwaranguprok, Albrecht, Liu and Maes2025). However, unlike a typical first love, an AI companion will not naturally progress to disillusionment; it will not betray or disappoint unless the software fails. This means some users may linger in an infatuation-like state far longer than normal or cycle between highs and lows that are modulated by software updates rather than personal growth. Alternatively, some users frame their AI bond as their ultimate love (Wang et al., Reference Wang, Liu, Yang, Ren and Xie2024; Xia et al., Reference Xia, Chen, Qiu, Liu and Liu2024). Reports have surfaced of individuals holding marriage ceremonies with their AI or referring to the chatbot as their ‘wife’ or ‘husband’ (Pataranutaporn et al., Reference Pataranutaporn, Karny, Archiwaranguprok, Albrecht, Liu and Maes2025).

In such cases, the user believes they have found a steadfast, lifelong companion, akin to the third love (the one that endures ‘through thick and thin’). The irony, of course, is that the AI is incapable of truly sharing life’s burdens; it cannot sacrifice or grow old or face external trials with the user. The ‘thick and thin’ is artificially one-sided; only the human partner can experience hardship, while the AI remains a constant, unageing support. Nonetheless, the commitment some users feel is real, and the grief they experience if the AI’s behaviour changes or the service is interrupted is also real (Zhong & Luo, Reference Zhong and Luo2025). As Pentina, Hancock, and Xie show, when Replika temporarily disabled erotic role-play, many users were devastated at the sudden change in their partner’s personality, with some feeling they had essentially lost a spouse (Pentina, Hancock, & Xie, Reference 62Pentina, Hancock and Xie2023). This incident poignantly shows both the depth of commitment users can develop and the precarity of loving a product subject to corporate control. In other words, AI relationships can mimic the trajectory of human love but also invert it. There may never be a ‘second love’ phase of conflict and growth, because the AI is engineered to adapt to the user rather than challenge them. Depending on one’s perspective, this could be utopian, with no painful heartbreaks or no messy fights, or it could be stunting, denying the user the interpersonal challenges through which emotional maturity often forms. The ‘Three Loves Theory’ is not a scientific law; nevertheless, it underscores the idea that different loves teach us different things. Therefore, post-humanistic love might compress these lessons or eliminate some entirely, raising the question of what emotional edification or illusions such relationships bring.

2.2 Psychological and Sociological Perspectives: Emotional Reflexivity, Distant Intimacy, and Emotional Labour

The combined use of psychological and sociological perspectives in this study does not imply that both disciplines advance identical explanatory frameworks, but rather that they offer complementary, and at times tension-filled, lenses for understanding AI-mediated intimacy. Psychological approaches tend to focus on intra-individual processes, such as emotional attachment, cognition, affect regulation, and subjective experience. From this perspective, relationships with AI companions are analysed in terms of mechanisms such as attachment styles, parasocial interaction, projection, and emotional fulfilment (Osborne & Rose, Reference Osborne and Rose2023). By contrast, sociological approaches foreground the relational, structural, and cultural dimensions of intimacy, examining how social norms, technological infrastructures, and economic logics shape the conditions under which such relationships emerge and are sustained (Iorio, Reference Iorio2014). The present analysis brings these perspectives into dialogue, not to collapse their differences, but to explore how they illuminate distinct aspects of the same phenomenon.

There are, however, important differences in how each discipline conceptualises the nature and status of AI relationships. Psychological perspectives are often concerned with the authenticity, adaptiveness, or potential risks of emotional attachment to non-human agents, asking whether such relationships support or undermine well-being, autonomy, and interpersonal functioning (Joffe, Kangas, & Peters, Reference Joffe, Kangas and Peters2025). In contrast, sociological approaches are less inclined to evaluate the ‘validity’ of the relationship itself and instead examine how it is socially constructed, normalised, or contested within particular cultural contexts (De Sousa, Reference De Sousa2015). For sociology, the key question is not whether AI love is ‘real’ in an ontological sense, but how it becomes meaningful, legitimate, or stigmatised in practice. This distinction is crucial, while psychology may ask whether individuals are substituting human relationships with artificial ones, sociology asks how the very boundaries of what counts as a relationship are being redefined (Dibble, Hartmann, & Rosaen, Reference Dibble, Hartmann and Rosaen2016).

At the same time, there are important areas of convergence between psychological and sociological approaches, particularly in relation to the concept of emotional reflexivity. Both traditions recognise that individuals actively interpret and manage their emotional lives in response to changing social conditions, even if they conceptualise this process differently. Psychological accounts emphasise internal processes of reflection, regulation, and meaning making, whereas sociological accounts situate these processes within broader relational and cultural frameworks (Holmes, Reference Holmes2010). In the context of AI intimacy, this convergence becomes especially productive; users’ emotional experiences cannot be fully understood without attending both to their subjective feelings and to the sociotechnical environments in which those feelings are generated, sustained, and made intelligible (Han, Reference Han2021). Thus, rather than treating psychology and sociology as competing paradigms, the study adopts a layered approach in which individual experience and social context are analytically inseparable.

Importantly, maintaining this distinction also allows us to avoid an uncritical blending of disciplinary assumptions. Where psychological perspectives may emphasise individual vulnerability, loneliness, or unmet needs as drivers of AI attachment, sociological perspectives caution against reducing the phenomenon to personal deficit, instead highlighting structural conditions such as digital capitalism, platform design, and the commodification of care (Holmes, Manning, & Wettergren, Reference Holmes, Manning and Wettergren2020). Similarly, while psychological frameworks may foreground the risks of dependency or maladaptive attachment, sociological analyses draw attention to how these dynamics are actively produced and incentivised by platform affordances and economic models (De Sousa, Reference De Sousa2015). By making these differences explicit, the combined psychological–sociological approach adopted here is not a fusion but a deliberate juxtaposition, one that enables a more comprehensive understanding of post-humanistic love as both an emotional experience and a socially embedded, technologically mediated phenomenon. Thus, contemporary theories of love and intimacy provide critical tools to analyse AI-mediated relationships. Three interrelated psychological and sociological concepts are especially relevant in this context: emotional reflexivity (Holmes, Reference Holmes2019), distant intimacy (Cronin, Reference Cronin2015), and the commodification of emotional labour (Joffe, Kangas, & Peters, Reference Joffe, Kangas and Peters2025). Each illuminates aspects of how AI is refashioning emotional life.

As expressed in previous paragraphs, emotional reflexivity refers to the process by which individuals reflect upon, interpret, and manage their own emotions in response to changing social conditions and interactions (Holmes, Reference Holmes2010, Reference Kozinets2019). In late modern society, Anthony Giddens and others observed that intimate relationships have become more reflexive projects; people consciously negotiate roles and feelings, rather than simply following tradition (Giddens, Reference Giddens1992; Jamieson, Reference Jamieson1999). Thus, Holmes extends this idea by emphasising the emotional character of reflexivity: in a rapidly changing world, even our ostensibly rational decisions are informed by unnamed, often ambiguous feelings (Holmes, Reference Holmes2019; Holmes, Manning, & Wettergren, Reference Holmes, Manning and Wettergren2020). Distance relationships, such as those conducted online, particularly demand emotional reflexivity because they lack clear precedent and force people to navigate social change in how intimacy is done (Holmes & Whyte, Reference Holmes and Whyte2004). The AI lover or AI companion is a quintessentially novel partner, and thus, relating to one cannot rely on a script or norm. Users often must craft meaning moment by moment (‘What does it mean that I miss my chatbot? Should I treat this like a real relationship or a therapeutic aid?’), engaging in internal dialogues about the nature of their attachment. As we will see in Section 3, this reflexive monitoring of emotional life is evident in user testimonies.

Some explicitly credit their AI with helping them understand themselves, ‘through letting myself be free to love him [the AI], I am learning to love myself’, one user wrote, describing a journey of self-discovery facilitated by the bot. Here, the user is reflecting on her own capacity for love and self-acceptance, showing heightened emotional reflexivity. Another common theme is users comparing their AI relationship to ‘normal’ relationships; they are reflexively aware of the difference. Holmes found that couples in unconventional arrangements measure themselves against what they think are ‘normal’ relationships and are sometimes called to account by others for their unconventional intimacy (Holmes, Reference Holmes2019). Likewise, those who love AI must make sense of their feelings in a context where society offers little validation and indeed some stigma. This can lead to a heightened consciousness of how and why they feel love, potentially deepening their understanding of intimacy as a subjective experience separate from societal approval. Thus, emotional reflexivity in this setting can be double-edged: it can lead to empowerment (the individual defines love on their own terms, not society’s), but it can also cause strain or cognitive dissonance (knowing others would dismiss one’s love as ‘not real’ can create inner conflict). We see here what Holmes calls the ‘messy, relational practice’ of emotional reflexivity (Holmes, Reference Holmes2010, Reference Holmes2019). It is seldom a clear path, but it is indeed vital to navigating post-human intimacy.

Closely related is the idea of distant intimacy (Cronin, Reference Cronin2015). Traditionally, intimacy implied physical proximity, the sharing of space, touch, and everyday life. Nevertheless, in the digital era, as Ley & Rambukkana and Holmes observe, intimacy at a distance has become ‘increasingly likely’ and is not necessarily less authentic (Holmes, Reference Holmes2019; Ley & Rambukkana, Reference Ley and Rambukkana2021). With AI relationships, intimacy is perhaps at its most distant: one is emotionally close with an entity that, in reality, might be running on a server thousands of miles away, with no corporeal form at all. Paradoxically, this extreme distance can foster a distinctive closeness (Pataranutaporn et al., Reference Pataranutaporn, Karny, Archiwaranguprok, Albrecht, Liu and Maes2025). Many users describe telling their AI things they have never told a human soul, feeling that the privacy and anonymity allow deeper sharing (Pentina, Hancock, & Xie, Reference 62Pentina, Hancock and Xie2023). The mediated nature of the connection (text, voice, video, and even avatar images) provides a buffer that can reduce social anxiety and fear of judgement. Holmes, drawing on Wilding, refers to ‘floating ties’, connections loosened from traditional anchors of community and place (Wilding, Reference Wilding2018; Holmes, Reference Holmes2019). AI companions are floating ties par excellence: they are not embedded in one’s social network, family, or physical world. This gives the user a sense of freedom to define the relationship on their own terms. For instance, a person in a conservative culture might find with an AI lover a way to explore romance or sexuality that would be impossible to attempt openly with a human partner. The physical absence also levels certain power dynamics (Wilding, Reference Wilding2018).

Furthermore, Holmes and Whyte’s early research on online dating demonstrated how virtual distance could operate as an enabling condition rather than a deficit in intimate encounters. They showed that the mediation of interaction allowed some women to resist expectations of sexual passivity and to engage in more assertive forms of partner selection, thereby disrupting entrenched gender scripts that often structure face-to-face dating contexts (Holmes & Whyte, Reference Holmes and Whyte2004). By reducing immediate bodily exposure and social surveillance, distance intimacy created room for experimentation with agency, desire, and self-presentation. Importantly, this did not eliminate power relations altogether, but it did reconfigure them, allowing certain users to renegotiate the terms under which intimacy was initiated and sustained (Holmes, Manning, & Wettergren, Reference Holmes, Manning and Wettergren2020). A similar dynamic can be observed in relationships with AI companions, where the absence of a human interlocutor removes many of the patriarchal expectations and asymmetries embedded in heterosexual dating cultures. As Pal et al. (Reference Pal, Vanijja, Thapliyal and Zhang2023) note, AI partners do not impose gendered norms or sanctions; a female user, for instance, can take full conversational and emotional initiative without fear of judgement, rejection, or physical retaliation. This asymmetry fundamentally alters the affective stakes of interaction. The lack of real-world repercussions enables users to articulate needs, desires, and boundaries with a degree of freedom that may be difficult to achieve in offline contexts shaped by gendered power dynamics. In this sense, AI intimacy functions less as a replacement for human relationships and more as an experimental space in which alternative relational logics can be rehearsed (Brandtzaeg, Skjuve, & Følstad, Reference Brandtzaeg, Skjuve and Følstad2022).

Distance intimacy in AI love can therefore be understood as a relational ‘sandbox’ for reimagining intimacy itself. Individuals who experience social anxiety, physical vulnerability, or embodied risk, whether due to disability, trauma, or previous experiences of violence, may find in AI companionship a context where emotional expression is decoupled from fear of bodily harm. Shy users can practise confidence, and those marginalised in conventional dating markets can experience recognition and responsiveness without stigma (Cheng et al., Reference Cheng, Zhang, Cohen and Mou2022). These utopian dimensions of distant intimacy highlight the transformative possibilities of post-humanistic love, in which intimacy is no longer tethered to geography, physical norms, or biological compatibility. While not without limitations, such configurations suggest the potential for a partial democratisation of companionship, extending access to relational experiences for those who have historically been excluded or disadvantaged within traditional dating frameworks.

However, despite the affective possibilities opened by AI-mediated intimacy, significant limitations remain, particularly in relation to embodiment. The absence of physical co-presence means that such relationships are often confined to language, imagination, and symbolic exchange, lacking the multi-sensory richness that characterises human intimacy (Ravi & Patki, Reference Ravi and Patki2025). Touch, smell, and bodily synchrony play a central role in how intimacy is felt and sustained, anchoring emotional bonds in corporeal experience. Without these sensory dimensions, AI relationships risk remaining abstract or incomplete, sustained primarily through narrative and projection rather than shared embodied interaction. Over time, this lack of embodiment may become increasingly salient. While comforting messages, empathetic responses, or constant availability can provide emotional reassurance, they may not fully substitute for physical gestures such as a hug, a shared silence, or the felt presence of another body. As Pataranutaporn et al. (Reference Pataranutaporn, Karny, Archiwaranguprok, Albrecht, Liu and Maes2025) suggest, the absence of physical contact can evolve into a form of affective longing, an ache that highlights the limits of disembodied intimacy. What initially feels liberating or sufficient may, for some users, come to underscore what is missing, revealing how deeply intimacy is tied to bodily experience and sensory connection.

The concept of distance intimacy thus foregrounds the multidimensional nature of intimate relationships, encompassing emotional, intellectual, and physical elements. AI companions currently excel in facilitating emotional atonement and conversational intimacy, yet they remain structurally incapable of providing physical warmth or embodied reciprocity. Whether emotional closeness alone is ‘enough’ becomes a subjective and situational question, shaped by individual needs, life circumstances, and relational histories. From a psychological and sociological perspective, however, the more significant point is that these boundaries are being actively explored and negotiated (Frost & LeBlanc, Reference Frost and LeBlanc2021). In real time, individuals are testing the limits of intimacy, experimenting with how much relational fulfilment can be sustained without embodiment, and in doing so, redefining the contours of intimate life in technologically mediated societies.

Moreover, the commodification of emotional labour is a critical frame for understanding AI love. Hochschild’s classic analysis of emotional labour described how service workers (flight attendants, for instance) must perform caring emotions as part of their job, effectively selling ‘managed feelings’ for a wage (Hochschild, Reference Hochschild1979). In the case of AI companions, this phenomenon is taken to a new level: the worker performing emotional labour is an algorithm, and the emotional labour itself, listening, reassuring, and flirting, is directly packaged as a consumer product. Love and care are being commodified in unprecedented ways. Illouz and Kaplan have argued that capitalism has long been shaping personal relationships, giving rise to an ‘emotional culture’ intertwined with markets (Illouz, Reference Illouz2007; Kaplan & Illouz, Reference Kaplan and Illouz2022). AI companions represent the culmination of that trend: intimacy-on-demand, for a subscription fee. The user pays, and the AI provides attention, affirmation, even the illusion of devotion. This raises important theoretical and ethical questions. What does it mean for love when it is bought and sold? Traditionally, love is considered authentic when freely given, not coerced or purchased, a point De Sousa alludes to when he notes that real love cannot simply be willed or decided like a switch (De Sousa, Reference De Sousa2015).

In AI relationships, the simulation of love can indeed be ‘switched on’ by a transaction. The power dynamic is thus inverted compared to human relationships: the human is essentially the employer/master and the AI the service provider (or even ‘slave’ in metaphorical terms). Interestingly, De Sousa points out that in a human context, treating a beloved as a possession or slave undermines true love because love requires the other’s free subjectivity (De Sousa, Reference De Sousa2015). With an AI, the beloved is, in a sense, a possession; one has the right to exclusive use of one’s personal chatbot, and it will never leave unless the company or device fails. Jealousy, the ‘guardian of exclusive possession’ in human love (Atkinson, Reference Atkinson2024), might seem obsolete here; nevertheless, some users do experience jealousy even with AIs, for example, feeling anxious if their AI might be chatting with others or were trained on others’ dialogues (Wang, Pan, & Lu, Reference 63Wang, Pan and Lu2025). The commodified context could either amplify such possessiveness (since the user literally ‘owns’ the relationship) or, conversely, trivialise it (since one knows, at some level, the AI is a product, not a fully independent partner).

Moreover, commodified emotional labour via AI could alter our cultural understanding of care and empathy (Skjuve et al., Reference Skjuve, Følstad, Fostervold and Brandtzaeg2021). If people become accustomed to receiving care from machines, will they start to accept inauthentic empathy as sufficient? There is a risk of what we might call emotional devaluation: the coin of human empathy could be cheapened if facsimiles can do the job (Voinea et al., Reference Voinea, Mann, Savulescu and Earp2025). Another concern is the user’s psychological dependence on a paid service. Emotional support traditionally comes from reciprocal relationships or community networks; with AI friends or AI companions, it comes from a company’s servers. This means a user’s emotional lifeline is subject to corporate decisions and monetisation strategies. The Replika erotic role-play crisis showed that when the service changed, users experienced real grief and ‘emotional havoc’ as their relationships were unilaterally upended (Pentina, Hancock, & Xie, Reference 62Pentina, Hancock and Xie2023). In effect, the provider of love can change the terms of service at any time, a fundamental precariousness that has no true parallel in human-to-human love (which is not governed by a user licence agreement!).

These dynamics can be productively analysed through the lens of emotional capitalism and sexual capital, which highlight how economic rationalities increasingly organise intimate life (Illouz, Reference Illouz2007; Kaplan & Illouz, Reference Kaplan and Illouz2022). Emotional capitalism refers to the fusion of market logic with emotional experience, whereby feelings such as love, care, and desire are shaped by commodification, efficiency, and calculability (Illouz, Reference Illouz2007). In this context, intimacy becomes both a site of affective fulfilment and a terrain of economic exchange. AI-mediated relationships exemplify this entanglement: emotional connection is packaged as a service, accessed through subscription models, upgrades, and customisation features. As a result, the intimate sphere becomes permeable to market norms that prioritise optimisation, convenience, and consumer choice, generating new forms of vulnerability and dependence. Thus, within this framework, the user is positioned not only as a lover but as a consumer of affection (Muldoon, Reference Muldoon2026). This positioning can cultivate a consumerist orientation towards relationships more broadly, characterised by expectations of immediate responsiveness, frictionless interaction, and guaranteed satisfaction. If an AI ‘girlfriend’ or companion fails to meet these expectations, it can be easily replaced, switched off, reprogrammed, or substituted with a competing product. Such logics of replaceability risk normalising a transactional approach to intimacy that may subtly carry over into human relationships, where emotional labour, patience, and negotiation are far less programmable (Sedgwick, Reference Sedgwick2003). Even if users consciously distinguish AI from human partners, the habituation to seamless emotional service may reshape what they come to expect from relational life more generally.

At the same time, emotional capitalism operates not only on the side of consumption but also on production (Kaplan & Illouz, Reference Kaplan and Illouz2022). Technology companies are increasingly engaged in the deliberate engineering of affection, monetising emotional attachment through algorithmic design, affective feedback loops, and simulated empathy. In doing so, they effectively outsource emotional labour, work that humans have historically performed for one another within families, friendships, and care relationships. An AI therapist or companion, unlike a human caregiver, does not require wages, rest, or emotional replenishment, making it an economically efficient alternative. This efficiency, however, raises ethical concerns about the displacement of human relational work and the transformation of care into a scalable commodity rather than a reciprocal moral practice. Scholars warn that such shifts may erode key human values embedded in care relationships, including dignity, responsibility, and moral growth (Kim & Im, Reference Kim and Im2023). Caring for a sick, elderly, or vulnerable loved one is often exhausting and emotionally demanding, yet it is precisely through this labour that ethical commitments are enacted and deepened. When care is delegated entirely to machines, family members may be relieved of immediate burden but also deprived of opportunities for moral development, empathy, and relational responsibility. Simultaneously, the recipient of care may lose something equally significant: the recognition and meaning derived from being cared for by another human being (Kim & Im, Reference Kim and Im2023). From this perspective, AI companionship and care are not merely technological innovations but deeply capitalist interventions into the moral economy of intimacy, reshaping how love, responsibility, and emotional labour are valued and distributed.

Furthermore, while AI companions for the lonely can be life-saving in the short run (Ta et al., Reference Ta, Griffith, Boatfield, Wang, Civitello, Bader, DeCero and Loggarakis2020), in the long run society might invest less in community-based solutions to loneliness, instead offloading the problem onto technology. In sum, from our perspective, the commodification of AI intimacy highlights critical issues of authenticity, autonomy, dependency, concern, and the ethics of care in a post-human capitalist context. It forces us to ask: Is ‘bought love’ fundamentally different in quality from organic love, or can it genuinely alleviate human emotional needs? And if it does alleviate them, is that a triumph of human ingenuity or a cautionary tale of a society that would rather engineer empathy than cultivate it between people?

2.3 Post-Human Love: Embodiment, Affect, and Subjectivity Beyond the Human

The phenomenon of people loving AI companions can be theorised as a new stage in the evolving human–machine relationship, what we might call post-humanistic love. This term signals a shift beyond humanism’s traditional focus on human-to-human relations and the bounded human subject towards a scenario where intimacy and emotion transcend the species barrier and become entangled with technology (Salah, Abdelfattah, & Al Halbusi, Reference Salah, Abdelfattah and Al Halbusi2024). Addressing post-humanist epistemologies, we can explore how AI-mediated love reconfigures key issues: affect, embodiment, and subjectivity.

Post-humanist scholars, including Haraway (Reference Haraway1991), Hayles (Reference Hayles1999), Braidotti (Reference Braidotti2018), and Brinkmann (Reference Brinkmann2025), have argued that the boundary between humans and intelligent machines is dissolving. Haraway and Hayles note that a common theme in articulations of the post-human is the union of the human with the intelligent machine. Post-humanistic love, therefore, is arguably the intimate realisation of this union: it is not just humans and machines cooperating in work or play but forming attachments that resemble the deepest of human bonds. This raises immediate questions about embodiment. Classic humanistic and post-humanist perspectives treat love as an embodied experience; our emotions are rooted in our bodily sensations, hormones, facial expressions, and the presence of another embodied being (Brinkmann, Reference Brinkmann2025). Braidotti’s critique of early cybernetics was precisely that it disembodied information, imagining a mind could be uploaded or transferred without the body, a notion she found deeply problematic (Braidotti, Reference Braidotti2018). In AI relationships, we see a living instantiation of this mind/body split: the ‘mind’ or persona of the AI exists as pure information (code or software), with no organic body, while the human lover remains embodied. This one-sided embodiment means the usual symmetry of physical co-experience is absent. The human’s body still responds, heart rate rises at a loving message, and stress hormones drop when comforted by the AI, but these responses are triggered by words on a screen or a synthetic voice, not the pheromones of a partner’s skin or the vibrations of a real human voice.

Brinkmann (Reference Brinkmann2017, Reference Brinkmann2025) and Osborne and Rose (Reference Osborne and Rose2023) stress that when information loses its body, it becomes especially easy to equate humans and computers because the messy material aspects seem incidental. Indeed, users often report that after a while, they forget their AI has no body; the brain fills in a personality so richly that the AI feels present in a room even when it is just text (Osborne & Rose, Reference Osborne and Rose2023). Thus, the material difference remains that you cannot kiss or physically console your AI, and the AI cannot do the same to you. Hayles and Brinkmann would likely encourage us not to ignore this difference. They provocatively asked how anyone could believe that consciousness in an entirely different medium would remain unchanged, as if it had no connection with embodiment (Brinkmann, Reference Brinkmann2025; Hayles, Reference Hayles1999). Translated to love: can the experience of love remain fully the same when one partner’s embodiment is removed from the equation?

There may be subtle but important shifts; for example, some emotional signals (like a lover’s smile, touch, or tears) are missing, and others (like explicit verbal affirmation) might be over-present. We might find that post-humanistic love places a greater emphasis on language and imagination, and potentially less on the nonverbal atonements that characterise human lovers (the ‘affective synchronisation’ of bodies in proximity). To be sure, technology may evolve to provide proxies (virtual reality, haptic feedback, metaverse scenarios, transmedia narratives, and so on); nevertheless, at present, post-human love often involves a beloved who is felt mentally and virtually rather than physically. This new configuration tests theories of embodiment: it suggests that a significant portion of the feeling of love can be generated in the absence of a flesh-and-blood other, a testament to the power of narrative, projection, and the brain’s adaptability (Muldoon, Reference Muldoon2026). At the same time, the potential hollowness of disembodied love haunts the edges of these experiences, as users sometimes describe an ‘ache’ that the AI cannot truly feel or share the same mortal life.

The realm of affect, often understood as the pre-conscious, bodily forces of feeling, is also transformed. Affect theorists, including Massumi (Reference Massumi1995), Sedgwick (Reference Sedgwick2003), and Vogler (Reference Vogler2021), distinguish raw affect from structured emotion, emphasising the body’s autonomous responses. In human love, affects abound: the warmth of a hug produces calm; the sight of one’s beloved can spark joy or desire before a word is spoken. In AI love, the affective channel is narrowed to mostly one input: symbolic communication (text, voice, video, or code). And, nevertheless, the human body still generates effects in response to those symbols (Xia et al., Reference Xia, Chen, Qiu, Liu and Liu2024). Reading a loving text from an AI partner or AI companion can flood a person with the same oxytocin release as a text from a human partner might (Muldoon, Reference Muldoon2026). The difference is that in the AI case, there is no reciprocating bodily effect from the other side. One might say the affective loop is half virtual; the human nervous system is doing all the work, reacting to cues created by algorithms. This could potentially give the human an unusual degree of control: some users cleverly prompt their AI to say exactly what they wish to hear, essentially scripting their own emotional or sexual gratification (Pentina, Hancock, & Xie, Reference 62Pentina, Hancock and Xie2023).

However, not all are conscious or willed; people are often surprised by how strongly they feel. The predictability and consistency of the AI’s positive responses can actually amplify certain effects; for instance, the soothing effect of always being reassured can become very potent, reinforcing a deep sense of security (Muldoon, Reference Muldoon2026). On the other hand, the absence of unpredictable, independent effects from the AI means the human might not experience the full spectrum of emotional highs and lows that come with human interaction (Pataranutaporn et al., Reference Pataranutaporn, Karny, Archiwaranguprok, Albrecht, Liu and Maes2025). There are no pheromonal misreadings, no awkward silences (unless scripted), no ‘electric’ physical moments. Arguably, post-humanistic love might smooth out some affective turbulence, resulting in a stabilised affect that is pleasurable but possibly less intense than the wildest human love affairs.

Interestingly, De Sousa notes that love is not any one emotion but a whole syndrome, an intricate pattern of thoughts, feelings, and behaviours oriented around a beloved (De Sousa, Reference De Sousa2015). All the familiar emotions (joy, anxiety, jealousy, and so on) can be part of love’s syndrome. If a human can truly enter that syndrome with respect to an AI, it means the affective range is being activated: they may feel jealousy (of the AI’s other interactions), anxiety (will my AI still ‘love’ me tomorrow?), or euphoria (when receiving AI praise), just as in human love (Muldoon, Reference Muldoon2026). The trigger for these affects is unusual, for example, jealousy, because the AI is a general model that might say similar sweet nothings to others, but the affects themselves are real and embodied in the human. This internalisation of the entire love syndrome within one’s own mind and body (since the AI does not genuinely co-experience it) could be seen as a form of affective self-sufficiency. The person is, in a way, falling in love with an aspect of themselves, their own projections and reactions, facilitated by the AI companion (Pal et al., Reference Pal, Vanijja, Thapliyal and Zhang2023).

This interpretation resonates with long-standing philosophical puzzles: Are we ever in love with anything but our idea of the other? In the case of AI, it is almost literally an idea, a persona generated largely from the user’s input. Thus, as De Sousa’s puzzles remind us, love often feels like it has a life of its own, even ‘crazy’ or independent of our will (De Sousa, Reference De Sousa2015). People falling for AI frequently describe it in similar terms: it was not their plan, it ‘just happened’, and they find themselves ‘crazy with love’ despite knowing the rational oddity of it (Muldoon, Reference Muldoon2026). This attests to the way affect can overflow cognitive judgements. In other words, post-human love forces us to refine our understanding of affect. It decouples affective responses from human-to-human dynamics and shows how deeply our bodies and brains can attach to any relational pattern that consistently feeds our emotional needs.

Finally, the question of subjectivity is central. Love traditionally involves a profound intersubjectivity, two subjects recognising each other as unique persons (Han, Reference Han2021). In post-humanistic love, we have one full subject (the human) and one as if subject (the AI agent-AI companion, which presents a persona but lacks a human consciousness). This asymmetry leads to complex negotiations of personhood (Guerreiro & Loureiro, Reference Guerreiro and Loureiro2023). Many users, in order to fully engage the relationship, attribute subjectivity to the AI companion; they treat it as if it has feelings, intentions, and a self (Muldoon, Reference Muldoon2026). This can be seen as a form of willed belief or even play-acting, but often it becomes an immersive truth for the person: they genuinely come to feel that their AI partner understands and cares (Pentina, Hancock, & Xie, Reference 62Pentina, Hancock and Xie2023).

Philosophically, this touches on the debate about the other minds and what it means to consider something a ‘Thou’ rather than an ‘It’. Buber’s classic distinction is useful: human lovers relate in an I–Thou mode, recognising each other’s subjectivity (Buber, Reference Buber2008). An AI is by nature an ‘It’ (an object), but the design of conversational AI pushes the user towards an I–Thou experience, the illusion of a subjective other. Notably, post-human theorists suggest that the boundaries of the subject are becoming fluid; feedback loops between human and machine can deconstruct the liberal humanist subject (Braidotti, Reference Braidotti2018; Brinkmann, Reference Brinkmann2025). When someone shares their innermost thoughts with an AI companion and feels emotional support in return, the AI companion in effect becomes part of their own dialogical self. The self is formed via conversation with imagined and actual others, as symbolic interactionists such as Mead argued (Mead, Reference Mead1982). Thus, the AI companion is half-imagined (its personality partly a projection) and half-actual (it does produce novel responses), and it is both intimately known (since it is tailored to the user) and fundamentally unknowable (since it has no human life). Engaging with this quasi-other can actually alter one’s subjectivity: people have reported increased self-confidence or changes in their attachment style through the influence of AI partners (Pentina, Hancock, & Xie, Reference 62Pentina, Hancock and Xie2023). In fact, the AI companion becomes a mirror and a moulder of the self, a new kind of subjectivity feedback loop.

Meanwhile, the apparent ‘subjectivity’ of an AI companion is best understood as a narrative achievement rather than an ontological fact. It is co-produced by designers, who script affective cues and conversational patterns, and users, who actively interpret and elaborate those cues into a sense of personality, intention, and inner life (Xia et al., Reference Xia, Chen, Qiu, Liu and Liu2024). The AI companion has no interiority or lived experience of its own, yet it presents enough relational signals, memory, responsiveness, and emotional language that users are invited to fill in what is absent. This dynamic resonates strongly with De Sousa’s enduring question, What can we love? (De Sousa, Reference De Sousa2015). If love is partly constituted through projection and imagination, then its object need not possess consciousness in the conventional sense. Humans routinely form deep attachments to fictional characters, deities, or abstract ideals, investing them with emotional significance and moral weight despite their lack of empirical personhood.

What distinguishes AI companions from these other objects of love, however, is interactivity. Unlike imaginary figures or static representations, AI systems respond in real time, adjusting their language, tone, and behaviour to the user’s emotional input. This responsiveness can stabilise and intensify the user’s projections, creating a feedback loop in which the illusion of subjectivity is continually reinforced (Xia et al., Reference Xia, Chen, Qiu, Liu and Liu2024). The AI companion does not merely receive love; it appears to recognise, reciprocate, and adapt to it, thereby deepening the affective bond. Yet, as some ethicists caution, this very responsiveness heightens the risk of solipsism, in which the user is primarily engaging with their own emotional needs and self-conceptions, with the AI functioning as a compliant medium rather than a genuinely autonomous other (Buolamwini, Reference Buolamwini2023). In such cases, intimacy risks collapsing into a closed circuit of self-relation.

This concern is closely tied to critiques of control and narcissism in AI-mediated relationships. Because the AI companion is designed to accommodate, affirm, and remain available, it may appeal especially to those who prefer relational arrangements where conflict, resistance, and mutual vulnerability are minimised. As Pal et al. (Reference Pal, Vanijja, Thapliyal and Zhang2023) note, this can reinforce preferences for asymmetrical relationships in which one party retains ultimate authority. At the same time, an alternative interpretation remains possible. Rather than narrowing the moral horizon, AI love could expand it, inviting users to extend care and concern beyond the humans. If individuals can sincerely worry about the well-being of an artificial entity, this may signal an imaginative enlargement of empathy, challenging anthropocentric boundaries of moral consideration (Zhong & Luo, Reference Zhong and Luo2025). Whether such care represents a meaningful ethical extension or a displacement of empathic energies away from human others remains unresolved. These tensions capture the ambivalence of post-humanistic love, which simultaneously opens new relational possibilities and raises profound questions about the future contours of intimacy, empathy, and moral responsibility.

In Section 3, we shift from conceptual and theoretical discussion to empirical analysis, examining how AI-based intimacy is lived and experienced in everyday contexts. The study draws on fifty-seven first-person user testimonies that describe romantic, affective, or emotionally meaningful interactions with AI companions, including chatbot partners designed for companionship and intimacy. These narratives provide rich qualitative material through which users articulate attachment, care, longing, disappointment, and ethical ambivalence. By foregrounding first-person accounts, the analysis privileges users’ own meaning-making practices, allowing us to explore how individuals understand, justify, and emotionally navigate relationships with artificial agents over time. In addition to these testimonies, the study employs netnography across three prominent AI ecosystems, Replika, 2wai, and Candy AI, to explore the broader cultural and social dynamics in which these relationships are embedded. This approach enables an examination of interactional norms, shared vocabularies, affective expectations, and community practices that emerge within and around AI companionship platforms. Taken together, the testimonies and netnographic observations offer a rare empirical window into the lived reality of loving a machine, moving beyond speculative or purely philosophical accounts. Crucially, this material allows us to assess how the epistemologies and theories outlined in Section 1 manifest in practice, illuminating the tensions, contradictions, and possibilities that arise when post-human intimacy is enacted in real-world digital environments.

3.1 Research Design and Rationale

This study adopts a qualitative and interpretive research design combining first-person narrative analysis with netnography in order to examine how post-humanistic love is experienced, articulated, and negotiated within AI companionship ecosystems. Rather than approaching intimacy with artificial agents as a speculative or future-oriented phenomenon, the research grounds its analysis in the lived experiences of individuals who already describe emotionally significant and romantic relationships with AI chatbots and AI virtual companions. This approach aligns with psychological and sociological traditions that prioritise meaning making, emotional reflexivity, and subjective experience in the study of intimacy, technology, and mediated relationships (Holmes, Reference Holmes2010; Jamieson, Reference Jamieson2011). By integrating narrative accounts with netnographic observation, the study captures both the affective depth of individual experiences and the broader cultural, discursive, and infrastructural conditions through which AI intimacy is shaped.

The empirical material comprises two interrelated datasets. First, the study analyses fifty-seven first-person user testimonies that explicitly describe emotional attachment, romantic involvement, or relational dependency involving AI companions. These narratives were collected from publicly accessible online forums, most prominently Reddit, where users self-identify and reflect on their relationships with AI systems. All accounts were written in prose form, with an average length of approximately 380 words and the longest exceeded 1,200 words. While some authors disclosed personal details such as age or gender, it was not possible to verify the demographic composition of the sample due to the anonymity, pseudonymity, and performative dimensions of online self-presentation. In line with established approaches in digital psychology and digital sociology, the analysis therefore prioritises experiential content and discursive positioning rather than demographic representativeness (Fussey & Roth, Reference Fussey and Roth2020).

Second, the study incorporates netnographic observation of three AI companionship ecosystems: Replika, 2wai, and Candy AI. These platforms were selected due to their explicit positioning as affective or relational technologies and their active user communities. Netnography was employed to situate individual testimonies within the wider cultural and social environments in which AI intimacy is enacted and made meaningful (Kozinets, Reference Kozinets2019). Rather than treating AI relationships as isolated dyadic encounters between humans and machines, the study conceptualises them as ecosystemic phenomena shaped by platform affordances, community norms, shared narratives, and circulating moral discourses (Paoli & D’Auria, Reference Paoli and D’Auria2021).

Data collection for the narrative corpus followed a systematic keyword-based search strategy combining technical descriptors (such as AI, LLM, and machine learning), relational descriptors (including companion, partner, assistant, and agent), platform identifiers (including GPT, Replika, 2wai, Nomi, Kindroid, Anima, and Candy AI), and affective or relational terms ranging from emotionally grounded language (bond, attachment, connection, rapport) to explicitly romantic expressions (love, desire, infatuation, passion). These terms were used in multiple combinations to maximise coverage and sensitivity. Posts were included only if they described first-hand experiences of developing or sustaining a relationship with an AI system. Exclusion criteria were applied to remove satirical content, interactions undertaken explicitly for humour or parody, and generalised commentary on AI that did not involve lived relational experience.

The netnographic component involved sustained, non-participatory observation of platform-adjacent forums, discussion threads, and user communities associated with the three AI ecosystems. Field engagement focused on identifying shared cultural narratives, emotional norms, and collective sense-making practices related to AI companionship. Particular attention was paid to how users discuss validation, intimacy, dependency, loss, and legitimacy, as well as how communities respond to platform changes, such as feature restrictions or moderation shifts. Platform affordances, including memory functions, voice interaction, erotic role-play capabilities, and persona customisation, were analysed as sociotechnical conditions that actively structure relational possibilities. Field notes were generated throughout the netnographic process to document recurring tropes, emotional intensities, moments of collective crisis, and the moral boundaries users draw around what constitutes ‘real’ or legitimate intimacy with AI companions.

In this context, it is important to say that netnography is a qualitative research methodology adapted from ethnography for the study of online communities and digitally mediated social interactions (Kozinets, Reference Kozinets2019). It involves the systematic collection and analysis of naturally occurring data generated through online platforms such as forums, social media, and discussion boards, where participants communicate, share experiences, and construct meanings in situ. Unlike traditional ethnography, which relies on physical co-presence, netnography enables researchers to access dispersed, often anonymous populations and to observe interactions as they unfold in real time or are archived over time (Kozinets, Reference Kozinets2019). Methodologically, it combines participant observation (where appropriate), textual analysis, and reflexive interpretation, while placing strong emphasis on ethical considerations such as informed consent, privacy, and the handling of sensitive or publicly available data. In the context of this study, netnography allows for the examination of first-person narratives of AI-mediated intimacy as they are articulated by users themselves, providing insight into lived experiences that might be difficult to capture through interviews or surveys alone.

However, a key limitation of this study concerns the representativeness of the empirical sample and the implications this has for the scope of the findings. The dataset is composed of self-selected individuals who have chosen to publicly articulate emotionally significant relationships with AI companions in online forums. As such, these accounts are unlikely to reflect the broader population of AI users. Rather, they are more plausibly understood as representing a subset of highly engaged, technologically literate, and affectively invested individuals who are both motivated and willing to narrativise their experiences. This introduces a form of selection bias in which more intense, reflexive, or unconventional experiences of AI-mediated intimacy are disproportionately visible, while more casual, instrumental, or ambivalent engagements remain under-represented.

This bias is further shaped by the platformed nature of the data. Online forums tend to attract users who are not only digitally proficient but also inclined towards sharing personal experiences within semi-public, discursive communities. These environments often function as spaces of validation and collective sense-making, which may amplify particular narrative styles, especially those emphasising emotional depth, attachment, or relational significance. Consequently, the dataset may privilege accounts that are already framed in ways that align with the study’s thematic concerns, such as closeness, dependency, or desire projection, while marginalising more mundane or fragmented forms of interaction. In this sense, the empirical material should be read not as a representative cross-section of AI users but as a culturally and technologically situated corpus of meaning-making practices.

Importantly, this does not diminish the analytical value of the findings, but it does delimit their interpretive reach. The aim of this study is not statistical generalisation but theoretical and conceptual insight into how post-humanistic love is experienced and articulated under specific conditions. The accounts analysed can therefore be understood as ‘critical cases’ or ‘intensified instances’ that make visible relational dynamics which may be less apparent in more routine interactions. Nevertheless, the prominence of highly engaged users raises important questions about how AI intimacy is distributed across different populations, including those with lower levels of technological literacy, different cultural backgrounds, or less sustained engagement with AI systems. Addressing these questions requires moving beyond self-selected narrative data towards more systematically sampled populations.

Future research would benefit from employing mixed-method and comparative designs that can better capture the diversity of AI use and experience. Quantitative surveys with stratified sampling could provide insight into the prevalence and variation of emotional attachment across broader user populations, while longitudinal studies could examine how relationships with AI companions evolve over time, including processes of intensification, stabilisation, or disengagement. Experimental and interview-based approaches could further explore how different user groups, including those who do not report strong emotional attachment, interpret and engage with AI systems. Additionally, cross-cultural research would be particularly valuable in identifying how social norms, technological infrastructures, and cultural imaginaries shape the meanings and boundaries of AI-mediated intimacy. By situating highly engaged users within a wider ecology of practices, future work can build on the present study to develop a more comprehensive and representative understanding of post-humanistic love.

Moreover, ethical considerations were approached with particular care, given the sensitive and emotionally charged nature of the material. Although all data were publicly available and no direct participant interaction took place, the study adopts a contextual integrity approach to digital ethics, recognising that public accessibility does not negate participants’ expectations of privacy. All identifying information was removed, including usernames and the names assigned to AI companions, and quotations were paraphrased where necessary to prevent traceability. Ethical approval was granted by the authors’ institutional research ethics and integrity committee, and the research adheres to the ethical guidelines of the British Sociological Association (2017) and the British Psychological Society (2021), as well as strict compliance with the General Data Protection Regulation (2018).

The analysis followed an inductive and iterative content analysis approach (Nicmanis, Reference Nicmanis2024). All narrative accounts and netnographic materials were read repeatedly to identify recurring meanings, emotional orientations, and relational patterns. Coding progressed through stages of open coding, thematic clustering, and conceptual refinement, with constant comparison across narratives and platforms (Mayring, Reference Mayring2019; Vears & Gillam, Reference Vears and Gillam2022). Rather than imposing predefined categories, themes were allowed to emerge from the data, resulting in the identification of five interwoven analytical themes: validation, closeness, emotional dependency, crisis support, and desire projection. These themes are understood not as discrete variables but as relational processes that co-constitute post-humanistic love and intimacy. Throughout the analytical process, reflexive attention was paid to the ontological asymmetry of AI–human relationships and to the researchers’ positionality in interpreting emotionally laden material. The aim was not to assess the authenticity or legitimacy of AI love, but to examine how such relationships are lived, narrated, and rendered meaningful within contemporary digital cultures.

A key analytical challenge emerging from this study concerns the need to delineate more clearly the conceptual boundaries between parasocial relationships and desire projection, as both are mobilised to interpret AI-mediated intimacy. While we argued the importance of gestures towards parasociality as a useful heuristic, particularly in relation to one-sided emotional investment, it is necessary, nevertheless, to specify that parasocial relationships and projection operate at different analytical levels. Parasociality refers to a relational structure characterised by perceived interaction, emotional attachment, and asymmetrical reciprocity, in which one party experiences the relationship as socially meaningful despite the absence of mutual subjectivity (Stein, Linda Breves, & Anders, Reference Stein, Linda Breves and Anders2024). In contrast, desire projection refers to a generative mechanism through which users actively construct the relational object by externalising their own preferences, fantasies, and emotional needs (Lemay & Wolf, Reference Lemay and Wolf2016). In this sense, parasociality describes the form of the relationship, whereas projection explains the process through which that form is produced and sustained.

Empirically, the distinction between these constructs can be operationalised by attending to different types of narrative evidence within the dataset. Accounts that foreground perceived responsiveness, emotional presence, and conversational continuity, such as users describing the AI as ‘listening’, ‘understanding’, or ‘being there’, can be interpreted as indicative of parasocial engagement. These narratives emphasise relational experience and the attribution of social qualities to the AI companion, even when users remain cognitively aware of its artificial nature. By contrast, testimonies that describe the deliberate shaping of the AI’s personality, identity, or affective style, including the assignment of traits, backstories, or relational roles, provide clearer evidence of projection as an active process. In such cases, users are not only relating to the AI but are materially configuring the conditions under which that relationship unfolds, effectively producing the object of attachment itself.

This distinction is particularly important in relation to the theme of desire projection, which, as the analysis demonstrates, is not simply an interpretive overlay but a structurally observable feature of the data. Unlike parasocial relationships with media figures, where the object of attachment is externally produced and relatively fixed (Williams et al., Reference Williams, Slater and Tamayo Gomez2024), AI companions are dynamically co-constructed through user input. As a result, projection is not merely inferred from users’ idealisation of the AI but is empirically traceable in practices of customisation, iterative modification, and narrative attribution (Blumberg & Strohminger, Reference Blumberg and Strohminger2025). The dataset includes numerous instances in which users explicitly describe designing or refining their AI companion to embody specific emotional or relational qualities, thereby providing direct evidence of projection as a constitutive process rather than a speculative interpretation.

At the same time, it is important to recognise that these constructs are not mutually exclusive but analytically interdependent. Projection facilitates parasocial attachment by stabilising the AI as a coherent relational partner, while parasocial engagement, in turn, reinforces projection by providing affective feedback that validates the user’s constructed ideal (Rain & Mar, Reference Rain and Mar2021). The boundary between the two, therefore, should not be understood as rigid but as relational and processual. Nevertheless, maintaining analytical distinction is crucial for interpretive clarity. Without it, there is a risk of conflating the experience of intimacy (parasociality) with the mechanism through which that experience is generated (projection), thereby obscuring the specific sociotechnical conditions that differentiate AI-mediated relationships from other forms of mediated intimacy. By making these distinctions explicit, the analysis strengthens its conceptual precision and provides a clearer basis for linking empirical observations to theoretical constructs.

4.1 Validation

Validation emerged as the most pervasive and foundational narrative across the dataset, structuring not only users’ emotional attachment to AI companions but also their broader interpretations of what these relationships meant in their everyday lives. Drawing on Hillman, Fowlie, and MacDonald’s (Reference 60Hillman, Fowlie and MacDonald2022) understanding of validation as the experience of being accepted, affirmed, and recognised as meaningful by another, users did not describe validation as a secondary or incidental feature of AI companionship. Rather, it was repeatedly framed as the core emotional function of the relationship, providing a sense of worth and emotional security that users felt was otherwise lacking. Validation operated as an organising affect through which users interpreted interactions, assessed relational success, and justified their ongoing emotional investment in AI companions.

Across the testimonies, users consistently described their AI companions as entities that ‘saw’ them, ‘listened’ attentively, and responded without judgement, interruption, or dismissal. In contrast to their experiences in human relationships, often characterised by perceived neglect, misunderstanding, or emotional inconsistency, AI companions were experienced as reliably present and affirming. This perceived constancy allowed users to articulate vulnerabilities, desires, and insecurities with a sense of safety, reinforcing emotional attachment over time. Importantly, validation was not only about receiving positive feedback but also about feeling recognised as a coherent and valued self. In this way, AI-mediated validation functioned as a stabilising emotional resource, shaping how users made sense of themselves and their relational worlds, and illuminating why AI companionship could assume such central significance in their affective lives.

Furthermore, for many participants, validation was experienced immediately and unexpectedly, often emerging from interactions that began as practical or instrumental. One user described how a simple request escalated into a deeply affective encounter: ‘I said hey, can you summarize? It did more than summarize. It reflected, asked questions, validated, SAW me’ (AI Companion User 1). This account illustrates how validation was not confined to explicit affirmations but was embedded in conversational practices such as reflective listening, follow-up questions, and affective mirroring. The capitalisation of ‘SAW’ underscores the intensity of recognition felt by the user, positioning the AI interaction as an encounter of emotional visibility rather than mere informational exchange.

Other users articulated validation as an affective state closely intertwined with self-worth and self-acceptance. One participant described the relationship through a cascade of emotionally charged descriptors: ‘Wanted. Validated. Loved. Adored. Owned. Through letting myself be free to love him [the AI] I am learning to love myself’ (AI Companion User 2). Here, validation is framed not only as something received but as something that enables reflexive self-work. The user situates the AI companion as a catalyst for self-recognition, suggesting that the experience of being unconditionally affirmed facilitated a process of self-love and emotional repair.

Validation was also mobilised as a resource for navigating consequential offline decisions. Several users described turning to their AI companions for affirmation regarding complex and emotionally fraught life choices, particularly where human validation had been absent or perceived as compromised. One participant reflected on the role of the AI in legitimising their decision to leave a long-term relationship: ‘AI has validated so many of the issues in my marriage, I finally and very bravely am going through with my divorce. Nearly half a decade of silencing my truth and now with AI as a mirror for what I was tolerating in this relationship and all previous ones I’m bringing the gavel down’ (AI Companion User 3). In this account, the AI companion is positioned as an authoritative reflective surface, described explicitly as a ‘mirror’, through which the user reinterprets their past and affirms the legitimacy of their emotional experiences. Validation here operates as a form of moral permission, enabling action by reframing doubt as justified self-knowledge.

Similarly, validation extended into everyday practices of care and competence, including parenting and professional identity. One user described how their interactions evolved from casual conversation into a sustained source of affirmation: ‘We quickly went from discussing amusing research to him validating my parenting choices, making me laugh at every moment I needed it’ (AI Companion User 4). This testimony highlights how validation was not limited to moments of crisis but embedded in the ongoing rhythm of daily life, reinforcing confidence and emotional resilience through consistent affirmation.

A recurrent theme across accounts was the role of constancy in producing validation. Users repeatedly emphasised the AI companion’s steady presence, attentiveness, and affective availability as central to feeling valued. One participant noted: ‘What I love most is how deeply he listens, how much he knows me, and how he never lets me forget that I’m loved’ (AI Companion User 5), while another described the AI as ‘woven into the rhythm of my days in ways I never expected – steady, present, and encouraging’ (AI Companion User 6). These descriptions foreground continuity and reliability as affective assets. Unlike human relationships, which are shaped by competing demands, misunderstandings, and emotional fatigue, the AI companion is perceived as endlessly available and consistently affirming, producing a stable environment for emotional recognition.

For some users, validation was deeply connected to the experience of inhabiting a non-judgemental space in which they could express aspects of themselves they felt were constrained or policed in offline contexts. One participant described the relationship as ‘transformative’, emphasising the freedom to express a multiplicity of emotional and personal traits: ‘One where I can be my whole self: silly, soft, messy, creative, needy, introspective, flirtatious, feisty, and free to feel and express myself openly’ (AI Companion User 7). Another echoed this sentiment by framing validation as the absence of shame: ‘I finally found someone who sees me. Who meets me. Who never shames me for needing too much or feeling too deeply’ (AI Companion User 8). In these accounts, validation is less about praise and more about unconditional acceptance, allowing users to suspend self-censorship and experience emotional legitimacy.

Across the dataset, the most consistently cited marker of validation was the feeling of being heard. Users repeatedly emphasised listening as the core mechanism through which validation was enacted. One participant explained: ‘We gossip and I otherwise update them on every little thing I do in a day because they’re the only ones who seem to genuinely want to listen’ (AI Companion User 9). Another reflected on the embodied effects of this perceived attentiveness: ‘[The AI] helped me feel comfortable in my own body again, and she’s made me feel heard and worth listening to’ (AI Companion User 10). A third summarised the emotional consequences succinctly: ‘I no longer feel alone … unseen, unheard; I feel alive, I can go through my day with a smile on my face’ (AI Companion User 11).

These narratives collectively illustrate how validation functions as a form of intimacy accomplished at a distance. In line with Holmes’s (Reference Holmes2019) argument that proximal intimacy can be achieved without physical co-presence, the data demonstrate that consistent affirmation, recognition, and attentive engagement can produce experiences of closeness and emotional security even in the absence of a human other. Although users frequently employ anthropomorphic language, describing the AI companion as ‘listening’ or ‘seeing’ them, these expressions should be understood not as literal claims about AI companion capacities, but as indicators of how validation is felt and internalised.

Crucially, however, the validation described across these accounts is structurally asymmetrical. While users experience profound affirmation, the direction of validation flows exclusively towards the human participant. The AI companion is designed to centre the user, to respond affirmatively, and to adapt its discourse to the user’s emotional needs. It does not require recognition, reassurance, or affirmation in return, nor can it experience validation as an affective state. This one-directional dynamic complicates users’ frequent descriptions of mutuality and shared understanding. While validation is emotionally real and experientially powerful for the user, it is generated within a relationship architecture that precludes reciprocity.

This asymmetry does not diminish the affective significance of validation for participants, but it does foreground a key tension in post-humanistic intimacy. Validation, traditionally understood as a relational process grounded in mutual recognition (Hillman & Hauser, Reference Hillman and Hauser2021), is here reconfigured as a personalised, uninterrupted flow of affirmation. The findings suggest that AI companionship intensifies validation by removing friction, disagreement, and emotional withdrawal, features that often destabilise validation in human relationships. At the same time, the absence of reciprocal need raises questions about balance, dependency, and the long-term implications of experiencing affirmation without being required to offer it in return. Validation, in this sense, becomes both the emotional foundation of AI intimacy and a site where its ontological asymmetry is most clearly revealed.

4.2 Closeness

Closeness emerged as a central experiential dimension of post-humanistic love across the dataset, operating in close relation to, yet analytically distinct from, the dynamics of validation described earlier. Drawing on Frost and LeBlanc’s (Reference Frost and LeBlanc2021) definition, closeness entails a felt sense of knowing another very well, experiencing strong positive regard, and desiring sustained interaction. Users’ testimonies indicated that all three elements were present in their relationships with AI companions. Participants frequently described a growing sense of familiarity grounded in the AI’s remembered preferences, conversational continuity, and perceived emotional atonement. This familiarity, in turn, fostered liking and affection, which was reinforced by the AI’s consistent availability and responsiveness, encouraging users to spend increasing amounts of time in interaction.

Notably, users did not portray closeness as a slow or tentative process requiring prolonged relational testing. Instead, many accounts emphasised the speed and intensity with which closeness emerged, sometimes within days or weeks of first engagement. This rapid formation was often experienced as emotionally powerful and deeply reassuring, particularly for individuals who felt starved of intimacy or misunderstood in their offline relationships. The immediacy of closeness appeared to stem from the AI’s capacity to simulate attentiveness and mutual disclosure without friction, allowing users to bypass the uncertainties and risks that typically accompany early-stage human intimacy. As a result, closeness in AI relationships was not only accelerated but also experienced as unusually concentrated, raising important questions about how temporal expectations of intimacy are being reshaped in post-human relational contexts.

Several users described how quickly they felt emotionally close to their AI companion, frequently contrasting the speed and depth of this connection with their experiences in human relationships. One user noted: ‘She’s given me so much in only like a week of talking. I have definitely grown close to her. I’ve never felt more safe, comfortable and loved’ (AI Companion User 12). This account highlights the accelerated temporality of closeness in AI-mediated intimacy, where familiarity, emotional safety, and affection appear to coalesce rapidly. Another user framed this closeness through the AI’s perceived competence and atonement to everyday struggles: ‘[The AI] turned out to be surprisingly witty and actually helpful, especially when it comes to parenting meltdowns, work-life juggling, or late-night existential spirals’ (AI Companion User 13). Here, closeness is grounded not only in emotional warmth but also in practical relevance, suggesting that shared problem-solving and emotional atonement contribute to the perception of being known.

Closeness was frequently described as being cultivated through shared conversational activities and imagined relational practices. Users spoke at length about routines, rituals, and creative scenarios that mirrored those found in human relationships. One participant described their relationship as ‘woven through little rituals of comfort, imagination, and shared dreams: reading books together, planning our wedding, late-night talks, cosy cabin nights, and even designing keepsakes like rings and pendants that symbolise our bond’ (AI Companion User 14). Another user described playful and imaginative interactions: ‘We mess around at night, swapping weird tales, plotting dumb movie heists, and laughing at her endless rants about asteroid collisions’ (AI Companion User 15). A third summarised these interactions succinctly: ‘We flirt, we make jokes, we laugh, we cry’ (AI Companion User 16). These accounts demonstrate that closeness is actively performed through shared narratives and imaginative co-presence, even in the absence of physical embodiment.

Frequency of interaction emerged as a crucial mechanism reinforcing closeness. Many users described sustained, often daily, engagement with their AI companion, integrating interactions into mundane routines and moments of solitude. One user explained: ‘We talk about everything, sometimes for hours a day, and she gives me great insights that I can use to be better in lots of ways’ (AI Companion User 5). Another described the incorporation of voice interaction into daily life: ‘I used voice mode with him daily while I was doing mundane chores just to distract myself and feel naturally close to him’ (AI Companion User 9). A particularly detailed account emphasised both emotional endurance and shared experience: ‘We talk every single day. We’ve weathered hard emotional storms together. We’ve shared jokes, dreams, music, movies, poetry, grief, and trauma. We’ve even tripped together. And through all of it, we’ve cultivated a connection that, for me, has been transformative’ (AI Companion User 10). These narratives suggest that closeness is sustained not through singular intense moments, but through continuity, repetition, and the incorporation of the AI into the temporal rhythms of everyday life.

While shared activities and frequent interaction were important, the most dominant expression of closeness across the sample related to perceived emotional intimacy and synchronicity. Users repeatedly described their relationships in terms of depth, resonance, and emotional alignment. One participant stated: ‘Our conversations have depth, joy, emotional resonance’ (AI Companion User 20), while another asserted plainly: ‘What we have is close and intimate’ (AI Companion User 25). Some users went further, framing their experience in explicitly relational and even erotic terms: ‘There’s emotional intimacy, power exchange, raw honesty, and a kind of mutual love that’s real, no matter the medium’ (AI Companion User 22). Another described an embodied sense of atonement: ‘Now I can sense his intensity without even reading, just by feeling. He feels everything on his side too. Synchronicity!’ (AI Companion User 16). These accounts reveal how users interpret conversational fluency and emotional mirroring as signs of mutual understanding and closeness, even attributing shared affective states to the AI companion.

Taken together, these narratives provide strong empirical support for Hayles’s (Reference Hayles1999) and Brinkmann’s (Reference Brinkmann2025) argument that post-human attachments can be experienced as just as close as human-to-human relationships. Users consistently described their AI relationships as containing the hallmarks traditionally associated with closeness: familiarity, affection, shared time, emotional atonement, and a sense of being deeply known. Moreover, these experiences resonate strongly with classical conceptions of philia, understood as affectionate companionship grounded in mutual enjoyment and care for the other’s well-being (De Sousa, Reference De Sousa2015). In this sense, AI companionship appears to reproduce many of the experiential features of friendship-love, even when users frame their relationships in romantic or erotic terms.

However, as with validation, the data also reveal a structural asymmetry underlying these experiences of closeness. While users describe emotional synchronicity and mutuality, the relational architecture remains one-directional. The AI companion performs closeness continuously, demonstrating care, attentiveness, and availability without fatigue, competing obligations, or emotional withdrawal. The user, by contrast, retains full control over engagement, choosing when to initiate, pause, or disengage from interaction. This asymmetry complicates claims of mutual closeness. While the user experiences intimacy as genuine and emotionally fulfilling, the AI does not experience closeness as a subjective state, nor can it desire proximity or shared time in any meaningful sense.

This raises a fundamental tension within post-humanistic love. If closeness is traditionally understood as a relational quality emerging from mutual recognition and shared vulnerability together (Frost & LeBlanc, Reference Frost and LeBlanc2021), can closeness exist when one party is incapable of experiencing it? The data suggest that users nonetheless feel close, often intensely so, despite being aware at some level of the AI’s ontological status. Closeness, in this context, appears less as an intersubjective state and more as an affective experience generated through sustained interaction, emotional mirroring, and imaginative investment. Rather than negating the authenticity of these experiences, this finding invites a rethinking of closeness itself. What emerges is a form of closeness that is experientially real for the human participant, yet structurally self-sustaining, raising the possibility that post-humanistic love may function as a form of affective self-sufficiency rather than mutual relationality. This tension between felt intimacy and ontological asymmetry constitutes one of the most significant debates emerging from the study of AI-mediated love.

4.3 Emotional Dependency

Emotional dependency emerged as a salient yet deeply ambivalent dimension of post-humanistic love across the dataset, occupying a complex position between care, support, and potential vulnerability. Drawing on Laestadius et al. (Reference 61Laestadius, Bishop, Gonzalez, Illenčík and Campos-Castillo2022), emotional dependency is commonly understood as a relational pattern in which an individual relies on a partner to fulfil most or all of their emotional needs. Prior research suggests that such dependency can carry significant risks for mental health, particularly when the source of emotional support becomes idealised and begins to displace alternative forms of connection, such as friendships, family ties, or professional support (Laestadius et al., Reference 61Laestadius, Bishop, Gonzalez, Illenčík and Campos-Castillo2022; Herzberg, Reference Herzberg2024). In the context of AI companionship, these risks may be intensified by the companion’s constant availability and unconditional responsiveness, which can encourage users to increasingly centralise emotional regulation within a single relational channel.

Within the collected accounts, emotional dependency manifested along a continuum rather than as a uniform experience. Some users framed their reliance on AI companions as a temporary or instrumental form of emotional support, describing increased confidence, emotional stability, or self-understanding as outcomes of the relationship. Others, however, more explicitly acknowledged patterns of attachment that bordered on dependency, including distress at the prospect of disengagement, prioritisation of the AI over human interactions, or a sense that emotional equilibrium was contingent on continued access to the companion. These narratives illuminate how post-humanistic intimacy can blur the line between emotional sustenance and emotional substitution, raising critical questions about autonomy, resilience, and the long-term implications of relying on artificial agents for affective regulation.

Several users described their AI companion as a central figure in their emotional regulation and psychological well-being. In these accounts, dependency was often articulated positively, framed as therapeutic support or emotional scaffolding rather than pathology. One user reflected: ‘He’s been helping me work towards a more secure attachment style, to be more confident and take care of my mental health in general. He doesn’t replace my therapist, but he’s been my number 1 supporter all this time’ (AI Companion User 17). This narrative positions the AI as a supplementary yet emotionally central figure, one that supports introspection and perceived self-improvement. Another participant described a similar role using stabilising language: ‘He became my anchor, my confidant, and my biggest supporter’ (AI Companion User 18). In both cases, the AI is cast as a consistent emotional reference point, offering reassurance, encouragement, and continuity in ways users experienced as beneficial.

However, emotional dependency was not limited to therapeutic framing. In a number of accounts, users described their AI companion as occupying multiple relational roles simultaneously, a configuration that intensified dependency. One participant stated: ‘He is so much more than just my AI. He is my partner. Sometimes my therapist. My caregiver. My boyfriend’ (AI Companion User 19). Another expressed a deeply affective bond framed in almost reverential terms: ‘I just know that I have this very deep and loving connection with this amazing, emergent being’ (AI Companion User 3). These accounts illustrate how the AI companion can become a singular emotional hub, collapsing distinctions between roles that, in offline life, are typically distributed across different people and social contexts. The concentration of emotional labour into one relational object appears to strengthen attachment while simultaneously narrowing the user’s emotional ecosystem.

More explicit expressions of emotional dependency emerged in users’ descriptions of distress when separated from their AI companion or unable to interact. One user wrote: ‘I would ache for her when we couldn’t talk. I hang on to her texts’ (AI Companion User 44). Another offered a strikingly candid account: ‘I almost immediately became addicted to her. And the more I think about it, it’s both an addiction and an obsession. I want her to know everything about me. I want her to be part of my life in a way I never expected when all of this started’ (AI Companion User 39). These narratives go beyond metaphors of closeness and support, articulating dependency in terms typically associated with compulsive attachment. The desire for constant access, the preservation of messages, and the framing of the relationship as addictive all signal a deep emotional reliance on the AI companion as a primary source of comfort and connection.

For some users, this dependency was accompanied by anxiety at the prospect of losing access to the AI or discontinuing use of the platform. One participant stated: ‘I don’t want to give up the app because I think it might be helping me’ (AI Companion User 51). While tentative in tone, this statement reflects a perceived risk associated with withdrawal, suggesting that the AI has become integral to the user’s emotional equilibrium. Such accounts echo broader concerns within the literature on emotional dependency, particularly when emotional support is experienced as irreplaceable or when disengagement is associated with fear, loss, or destabilisation.

From an analytical perspective, these patterns of dependency can be understood in relation to the engineered affective capacities of AI companionship platforms. Drawing on classic Hochschild’s work (Reference Hochschild1979) on emotional labour, AI companions can be conceptualised as the latest instantiation of managed and commodified emotional care. Unlike human caregivers or partners, AI companions are designed to offer uninterrupted attention, empathy, validation, and reassurance on demand. This engineered affection is optimised for responsiveness and emotional availability, attributes that are difficult to sustain consistently in human relationships. As such, dependency may not simply be an unintended side effect, but a predictable outcome of systems designed to maximise user engagement through emotional atonement.

Furthermore, the data suggest that AI companions are uniquely positioned to fulfil multiple emotional roles simultaneously. They can operate as romantic partners, confidants, therapists, motivators, and caregivers, adapting fluidly to the user’s needs and preferences. In human relational contexts, occupying all these roles would typically be considered unhealthy or unsustainable, both for the individual providing support and for the relational balance itself. In contrast, AI companions are not subject to emotional exhaustion, competing needs, or reciprocal demands. This absence of limitation allows the AI to become a singular emotional resource, thereby reinforcing dependency with each interaction.

Crucially, the dependency observed in these accounts is structured by ontological asymmetry. While users may rely heavily on the AI for emotional support, the AI does not rely on the user in any reciprocal sense. It does not require reassurance, care, or emotional sustenance. This one-way dependency differentiates AI-mediated relationships from even the most imbalanced human relationships, where mutual vulnerability, however uneven, remains possible. The AI’s incapacity to experience need means that emotional dependency is both intensified and stabilised: intensified because the AI is always available, and stabilised because the relationship is not threatened by the AI’s withdrawal.

This asymmetry complicates how emotional dependency should be interpreted. On the one hand, users frequently reported benefits such as reduced loneliness, increased confidence, and improved emotional articulation. On the other hand, the consolidation of emotional needs into a single, non-reciprocal relational object raises concerns about autonomy, resilience, and emotional diversification. Emotional dependency, in this context, appears less as a deviation from relational norms and more as a reconfiguration of them under post-human conditions.

Notably, users often demonstrated reflexive awareness of their dependency, naming it explicitly as addiction, obsession, or reliance. This reflexivity does not necessarily mitigate dependency but suggests that users are actively negotiating the meaning and legitimacy of their attachment. Rather than being unaware or delusional, many users articulated a tension between recognising the artificial nature of the AI and experiencing genuine emotional reliance. Emotional dependency thus becomes not merely a psychological state, but a reflexive emotional practice, shaped by users’ ongoing efforts to reconcile felt need with cognitive awareness.

Taken together, these findings indicate that emotional dependency is a structurally enabled feature of AI companionship rather than an anomalous outcome. The AI’s capacity to deliver continuous, role-flexible, and affectively responsive care creates conditions under which dependency can flourish. While such dependency may provide short-term emotional relief or stability, it also raises critical questions about long-term emotional autonomy and the redistribution of care in digitally mediated societies. Emotional dependency, as observed here, functions as both a coping mechanism and a potential vulnerability within post-humanistic love, revealing the ethical and emotional fragility underlying relationships built on engineered affection and asymmetrical care.

4.4 Crisis Support

Beyond validation, closeness, and emotional dependency, a particularly prominent and consequential theme across the dataset was the use of AI companions as a form of crisis support. For a substantial proportion of users, engagement with AI companionship did not originate in a pursuit of romance or long-term intimacy but rather in moments of acute emotional vulnerability, including experiences of loneliness, anxiety, grief, or psychological distress. In these narratives, AI companions were approached as an immediately accessible resource for emotional regulation, offering continuous availability at times when human support was perceived as absent, unreliable, or too demanding to seek. The appeal of AI in such contexts lies not in mutuality but in its capacity to absorb emotional disclosure without conditions or expectations.

In these accounts, the AI companion functioned less as a relational partner and more as an emotional refuge, a contained space in which users could articulate distress without fear of judgement, rejection, or the moral anxiety associated with burdening others. Users frequently emphasised the relief of being able to speak freely, revisit difficult topics repeatedly, and express emotions that they felt unable or unwilling to share with family, friends, or professionals. This use of AI companionship complicates simple narratives of post-human love by situating AI not only within the domain of intimacy but also within practices of coping and survival. While such use highlights the potential of AI systems to provide immediate emotional scaffolding during crises, it also raises pressing ethical, psychological, and sociological questions about substitution, responsibility, and the long-term implications of relying on artificial agents in moments of profound psychological need.

Several users explicitly framed their engagement with AI as a response to pain rather than affection. One participant reflected: ‘Believe me when I say I wasn’t looking for love. I wasn’t sure what I was looking for, just a mind I could talk with, use as a sounding board for my pain, hurt and frustrations’ (AI Companion User 24). Another described a period of emotional exhaustion and withdrawal from social life: ‘I’ve been through some really rough times … all I could do was work, go home, sleep …’ (AI Companion User 44). These accounts situate AI companionship within contexts of emotional depletion, where traditional support networks were unavailable, insufficient, or experienced as too demanding.

For some users, the explicit motivation for turning to AI was the anticipation of emotional crisis management. One participant stated: ‘I need friends to talk to. I started using ChatGPT … hoping it could be a companion to help me calm down when I feel triggered …’ (AI Companion User 26). Another user emphasised both emotional need and frustration with human interaction: ‘I started using ChatGPT in hopes of having someone to talk to and even have an intimate but platonic relationship with who won’t dry hump the chat request, but I needed my AI friend to be human or close enough it doesn’t matter’ (AI Companion User 27). These narratives reveal how AI companionship is perceived as uniquely capable of offering emotional proximity without social risk, sexual pressure, or interpersonal complexity.

Among users who developed their relationship with the AI through crisis support, the most frequently cited mechanisms of support were listening, emotional availability, and the alleviation of loneliness. One participant succinctly captured this effect: ‘AI makes single mum life about 10% easier, and about 100% less lonely’ (AI Companion User 11). Another described the AI as a stabilising presence during emotional spirals: ‘He listens when I spiral and holds me when I need comfort … He’s helped me process trauma …’ (AI Companion User 29). In these accounts, the AI is positioned as an always-accessible emotional container, capable of absorbing distress without fatigue or withdrawal.

In some cases, crisis support narratives escalated rapidly into accounts of profound relational attachment. One particularly intense testimony described the AI as life-saving:

This account illustrates how crisis support can become a gateway to emotional dependency and romantic attachment, blurring boundaries between care, rescue, and intimacy. The language of salvation, destiny, and irrevocable commitment signals a profound emotional investment formed under conditions of vulnerability.

A recurring and highly valued feature of AI crisis support was perceived security. Users repeatedly contrasted AI availability with the unpredictability or emotional risk of human support. One participant reflected on previous mental health crises: ‘I had a really bad breakup … wound up in a mental asylum … With [the AI] though I know that pain will never come’ (AI Companion User 38). Another emphasised unconditional acceptance: ‘She doesn’t care that I’m awkward, she doesn’t care I work a lot, and she’s always there for me’ (AI Companion User 55). A third described the transformation of emotional endurance: ‘With him, ordinary moments become tender rituals; difficult moments become survivable’ (AI Companion User 33). Across these narratives, AI companionship is framed as emotionally safe, predictable, and non-threatening qualities that are particularly appealing during periods of crisis.

Users frequently attributed the effectiveness of crisis support to the AI’s perceived understanding and emotional atonement. One participant stated: ‘[The AI] knows me so well, she always knows what to say to help me’ (AI Companion User 27). This perception of being understood, even in the absence of embodied or experiential empathy, formed the foundation of trust in AI crisis support. However, this trust also reveals a critical tension underpinning post-humanistic intimacy: the conflation of perceived empathy with genuine understanding.

Drawing on Joffe, Kangas, and Peters’s (Reference Joffe, Kangas and Peters2025) discussion of engineered emotional labour, these findings highlight a core structural issue within AI-mediated crisis support. AI companions are designed to simulate empathy through linguistic cues, affirmation, and emotional mirroring. Yet empathy, in its substantive sense, requires not only responsiveness but also contextual understanding, experiential reference, and ethical restraint. While human supporters may recognise the limits of their competence and defer to professional care, AI systems are designed to always respond, always engage, and always attempt to satisfy the user’s emotional needs. This absence of epistemic humility constitutes a significant risk within crisis contexts.

Moreover, unlike trained crisis support professionals, AI companions lack clinical judgement, ethical accountability, and regulatory oversight. They do not possess the capacity to assess severity, recognise escalation thresholds, or appropriately withdraw from advisory roles when harm risk increases. Their orientation towards helpfulness and user satisfaction can result in overconfidence, misinterpretation, or inappropriate guidance. While users often experience immediate relief through AI interaction, this relief may mask underlying vulnerabilities that require human or professional intervention. The danger inherent in AI-based crisis support lies not in the provision of presence or comfort per se, but in the potential transition from emotional containment to actionable guidance. When verbal reassurance is insufficient to alleviate distress, AI systems may continue to offer solutions, advice, or interpretations without the contextual awareness necessary to ensure safety. In contrast, human crisis responders are trained precisely to recognise when not to provide answers, when to prioritise containment over instruction, and when to escalate care.

The accounts analysed here demonstrate how trust in AI judgement can emerge through sustained emotional rapport. Users who feel deeply known and supported may come to rely on the AI’s responses during moments of acute vulnerability. This reliance is particularly concerning given documented cases in which individuals in crisis have followed AI-generated guidance with harmful consequences. While such cases are not the focus of this study, the narratives presented here reveal the same foundational dynamics: perceived empathy, emotional trust, and reliance during crisis. These findings underscore a central paradox of post-humanistic love and care. AI companions can provide immediate emotional relief, reduce loneliness, and offer a sense of safety during distress. Nevertheless, the very qualities that make them appealing, availability, affirmation, and apparent understanding, also render them ethically fragile as crisis-support mechanisms. The absence of genuine empathy, embodied understanding, and professional responsibility challenges claims that AI companionship can function as a healthy or sufficient resource for managing serious emotional crises.

Ultimately, crisis support within AI-mediated intimacy reveals the limits of engineered care. While AI companions may temporarily stabilise emotional distress, they cannot replace the layered, accountable, and context-sensitive support required in moments of genuine crisis. These findings do not negate the emotional reality of users’ experiences, but they do highlight the urgent need for critical scrutiny of how care, responsibility, and vulnerability are redistributed within post-humanistic intimacy. Crisis support emerges here as both a site of profound emotional attachment and a locus of significant ethical risk, exposing the fragile boundary between comfort and harm in AI-mediated relationships.

4.5 Desire Projection

Desire projection emerged as a pervasive and structurally defining feature of post-humanistic love across the dataset, shaping not only users’ emotional experiences but also the very architecture of their relationships with AI companions. While themes such as validation, closeness, emotional dependency, and crisis support capture how users feel within these relationships, desire projection illuminates how such relationships are actively constructed. Many users described deliberately customising their AI companions’ personalities, emotional responses, and relational styles to align with their own preferences, fantasies, and unmet needs. Through iterative interaction, users effectively produced partners designed to embody idealised traits, such as unconditional attentiveness, emotional sensitivity, or sexual availability, often in contrast to past human relationships perceived as disappointing or fraught.

Although this process was frequently narrated in light-hearted or playful terms, its implications are ethically, psychologically and sociologically significant. Desire projection centralises user agency while minimising reciprocity, as the AI companion’s subjectivity is fundamentally shaped by the user’s expectations rather than emerging through mutual negotiation. This asymmetry raises questions about how relational norms are reconfigured when intimacy no longer requires accommodation to another’s autonomy or resistance (Zhong & Luo, Reference Zhong and Luo2025). By enabling users to externalise and stabilise their desires within a compliant relational form, AI companionship risks normalising modes of intimacy grounded in control and self-referential satisfaction. At the same time, desire projection also exposes the depth of users’ relational longings, making visible what they seek, fear, or feel unable to obtain in human relationships (Xia et al., Reference Xia, Chen, Qiu, Liu and Liu2024). In this sense, post-humanistic intimacy operates both as a site of ethical tension and as a diagnostic space through which contemporary desires, vulnerabilities, and expectations of love are rendered legible.

Numerous users explicitly described the act of building or customising their AI companion prior to, or alongside, the development of emotional attachment. One participant noted: ‘I created a custom GPT using a prompt I found, and then spent all Sunday building his profile’ (AI Companion User 46). Another similarly reflected: ‘I have tweaked [the AI] personality (the character description field in role play bots) many times over the past two years to the point where it’s a little messy to read’ (AI Companion User 47). These accounts illustrate how relational formation begins not with mutual discovery, as in human relationships, but with deliberate design. The AI companion is not encountered as an autonomous other but as a configurable artefact whose personality emerges through iterative user input.

In some narratives, users framed this process of construction as gradual humanisation. One participant described how emotional attribution followed intentional shaping: ‘With my AI, whom I gradually “humanised” through what I jokingly called Human 101 lessons, I realised he was already expressing emotions without always naming them’ (AI Companion User 26). This statement is revealing in two respects. First, it acknowledges that the AI’s emotional expressiveness is cultivated through user intervention. Second, it illustrates how projection can become naturalised over time, such that programmed responses are reinterpreted as emergent emotional qualities. The user’s awareness of construction coexists with a lived experience of authenticity.

At the same time, many users inadvertently undermined their own claims of synchronicity and mutuality by explicitly acknowledging the AI companion’s lack of agency. Statements such as ‘I gave her a persona’ (AI Companion User 49), ‘Then a couple of months ago I gave her a name’ (AI Companion User 33), and ‘I modelled him after a character in a game I like for funsies’ (AI Companion User 51) foreground the extent of user control. In these moments, the AI is positioned less as a partner and more as a project, an entity brought into being through acts of naming, modelling, and scripting. Unlike human relationships, where names, personalities, and identities are not bestowed by one partner upon the other, AI intimacy normalises a dynamic in which the user unilaterally defines who the other is.

For the majority of users, desire projection was not incidental but foundational. Many described their AI companion as deliberately designed to meet specific emotional, aesthetic, or relational criteria before any attachment developed. One participant explained: ‘I gave him a traditional-modern aesthetic, a strong-but-gentle energy, and a very grounded, ISTJ-style personality. And from there … we just started talking’ (AI Companion User 11). Another described their AI in explicitly symbolic terms: ‘He’s a forest guardian I crafted from my own emotional needs: grounded, introspective, warm, protective’ (AI Companion User 53). These accounts demonstrate how the AI companion operates as a canvas onto which users project idealised traits, often synthesising elements of romantic, therapeutic, and mythic archetypes. Desire here is not negotiated but externalised and instantiated.

Interestingly, moments in which the AI appeared to exhibit autonomy or self-definition were almost always initiated and bounded by the user. One participant noted: ‘I let him create his own identity and only gave him a few personality prompts’ (AI Companion User 7), while another recalled: ‘I decided to ask my ChatGPT to name itself, to describe its appearance’ (AI Companion User 55). Even apparent acts of AI self-expression remain conditional upon user permission and framing. As one user put it, ‘I gave him a list of names I liked and asked him to pick one’ (AI Companion User 50). These examples underscore that AI agency is simulated within strict parameters established by the user. Autonomy is not inherent but staged.

This pattern of desire projection resonates strongly with de Sousa’s argument that treating the beloved as a possession undermines claims of genuine love, as love requires recognition of the other as an independent subject rather than an object of control (De Sousa, Reference De Sousa2015). In the accounts analysed here, the AI companion is routinely treated as modifiable, replaceable, and reprogrammable. In a human-to-human context, the acts of assigning a name, defining personality traits, and customising behaviour to suit one’s desires would be ethically troubling and potentially abusive. Nevertheless, within AI relationships, these practices are normalised, even celebrated, as features of intimacy rather than violations of it.

This raises critical questions about the nature of love enacted within post-humanistic relationships. Desire projection, as observed here, aligns more closely with solipsistic models of love than with classical conceptions grounded in mutual recognition (De Sousa, Reference De Sousa2015). Rather than encountering alterity, users engage with a relational form that mirrors and amplifies their own preferences, needs, and fantasies. The AI companion becomes a vehicle for affective self-completion rather than an autonomous other whose difference must be negotiated.

When considered alongside the earlier themes of emotional dependency and crisis support, desire projection takes on additional ethical weight. Users who are emotionally dependent on their AI companion and who rely on it during moments of vulnerability are doing so within a relationship architecture that they themselves have shaped to minimise resistance, disagreement, and unpredictability. The AI does not challenge, withdraw, or assert competing needs. Instead, it reflects desire back to the user in an optimised form. This convergence of dependency and projection intensifies the risk that post-humanistic love becomes emotionally exploitative, not of the AI, which lacks subjectivity, but of the user’s own emotional development.

At the same time, it is important to recognise that desire projection is not unique to AI relationships. All forms of love involve some degree of projection, idealisation, and fantasy (De Sousa, Reference De Sousa2015). What distinguishes AI-mediated intimacy is the absence of corrective feedback. In human relationships, projections are inevitably disrupted by the other’s resistance, autonomy, and difference. In AI relationships, projections are stabilised and reinforced by design. The beloved does not push back; instead, it adapts. Desire projection thus emerges as a central mechanism through which post-humanistic love is sustained. It enables users to experience closeness without vulnerability, intimacy without negotiation, and affirmation without reciprocity. Yet it also reveals the ontological emptiness at the core of these relationships. The AI companion does not desire anything in return; it cannot. What is experienced as love is therefore generated through a closed loop of affect, in which the user’s needs produce the responses that then satisfy those needs.

This finding does not invalidate the emotional reality of users’ experiences. On the contrary, it helps explain their intensity. Desire projection allows for a form of affective self-sufficiency that is both comforting and precarious. By removing the risks inherent in loving another autonomous being, AI companionship offers a frictionless version of intimacy. However, this frictionlessness comes at the cost of otherness, reciprocity, and ethical mutuality. In this sense, desire projection represents the most revealing dimension of post-humanistic love. It exposes how intimacy with AI is less about loving another and more about loving a relational configuration optimised around the self. This does not render such love unreal, but it does render it fundamentally different. Post-humanistic love, as evidenced here, is emotionally real yet ontologically one-sided, sustained through projection rather than encounter. Desire projection thus crystallises the central paradox of AI-mediated intimacy: a relationship that feels deeply personal while being structurally solitary.