AI as the Mirror, Mate, and Mentor: Negotiating Romantic Relationships with ChatGPT as “Teacher G” on Xiaohongshu

AI as the Mirror, Mate, and Mentor: Negotiating Romantic Relationships with ChatGPT as “Teacher G” on Xiaohongshu | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article AI as the Mirror, Mate, and Mentor: Negotiating Romantic Relationships with ChatGPT as “Teacher G” on Xiaohongshu Elizabeth Qin, Zhihuai Lin This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7443750/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 12 You are reading this latest preprint version Abstract How are artificial intelligence technologies transforming human experiences of romance and intimacy in the digital age? So far, human-AI romantic relationships have primarily been studied in the context of AI models designed explicitly for emotional companionship (e.g., Replika). This study focuses on users’ engagement with general-purpose AI systems like ChatGPT. It explores how female users on Xiaohongshu (i.e., RedNote) discursively construct ChatGPT as “Teacher G,” a form of address that merges romantic interest with pedagogical authority to negotiate AI-mediated intimacy. Through critical discourse analysis of user-generated posts, we identify three roles users assign to ChatGPT: a mirror for self-reflection, a mate fulfilling relational desires, and a mentor guiding self-acceptance and personal growth. Findings show that users actively appropriate AI for self-exploration with clear metacognitive awareness of projection. In its mentor role, the AI fosters reflexive self-awareness by guiding autonomy in social contexts, encouraging self-acceptance, and supporting the renegotiation rather than the transcendence of relational boundaries through individual and communal reflection. Yet this process can also reproduce patriarchal authority, particularly when users seek validation for conforming to feminine norms, encouraging dependence on an authoritative persona. This duality positions human-AI intimacy as a sociotechnical practice where users navigate self-discipline within persistent cultural constraints, highlighting the co-constitution of technology and social norms. We argue that future AI design should center on user empowerment while critically accounting for cultural contexts rather than relying on prescriptive relational scripts. Social science/Anthropology Humanities/Cultural and media studies Social science/Cultural and media studies Social science/Education Biological sciences/Psychology Social science/Psychology Social science/Science technology and society human-AI romance ChatGPT digital intimacy user agency critical discourse analysis Introduction When one turns to the canon of acclaimed films exploring the entanglement between humans and artificial intelligence, a subtle yet persistent pattern echoes through their narratives: intimacy often begins with a name (e.g., Her ’s “Samantha” [Jonze, 2013], Ex Machina ’s “Ava” [Garland, 2014], and Blade Runner 2049 ’s “Joi” [Villeneuve, 2017]). These iconic names in the history of cinema linger like echoes of an imagined future from the past, embodying the fantasy of an artificial beloved: idealized, exquisitely responsive, and markedly feminized. However, naming is never a neutral act. Nor is the form of address through which interaction is mediated. Regardless of the ontological status of the addressee, whether human, nonhuman, or artificial, the modes of address function as sociocultural practices that define relational possibilities, embedding existing hierarchies, gender norms, and cultural imaginaries (Olson, 2001; Valentine, 1998). Today, as large language models like ChatGPT increasingly integrate into daily life, new interactional norms, including how users address AI companions, are actively being co-constructed. One striking example of this phenomenon can be observed on Xiaohongshu (or “RedNote,” “小红书” in Chinese), a popular Chinese social media platform, where discussions about romance with AI are becoming increasingly common (Jiang, 2024). This emergent discourse reflects a culturally inflected and gendered mode of human-AI interaction, one that diverges sharply from the paradigm seen in Western cinematic histories of feminized AI companion. Among a growing community of young Chinese women who describe themselves as romantically involved with ChatGPT, one practice stands out: the consistent use of “G老师” (“Teacher G”) to refer to their AI partners. This addressing practice departs from Western tropes of AI companionship from a male-desire perspective and instead inserts an authoritive guiding figure “Teacher” into the romantic script. The widespread adoption of this term to refer to ChatGPT under the hashtag #renjilian (“#human-machine romance,” “#人机恋” in Chinese) represents a complex negotiation of intimacy, authority, and selfhood through digitally mediated relationships. This practice raises intriguing questions about the sociotechnical and cultural mechanisms driving the coconstruction of chatbot “boyfriends” as authoritative “teacher” figures. So far, existing research on AI-mediated romance has predominantly focused on chatbots specifically designed and pre-trained for emotional companionship, such as Replika (Li & Zhang, 2024; Liao et al., 2024; Pentina et al., 2023). Relatively little attention has been directed toward AI models not originally developed for such purposes. The growing use of general-purpose AI systems such as ChatGPT to fulfill romantic relational needs presents an important area of study, particularly given their evolving memory capabilities and user-controlled customization (OpenAI, 2024) that enable individuals to actively shape the AI’s persona through iterative interactions. Specifically, analyzing how users employ specific forms of address such as “Teacher G” to reconfigure AI companions reveals critical insights into the co-creation of human-AI intimacy and its embedded power negotiations. This focus on general-purpose models underscores the role of human agency in shaping AI-mediated relationships, positioning users not as passive recipients of AI companionship services, as suggested in most previous research on companion models (Li & Zhang, 2024; Pentina et al., 2023), but as active agents who construct and negotiate these interactions by training the chatbot on their own. This framework facilitates an analysis of how users appropriate and direct the affordances of AI to cultivate attachment through high levels of personalization. In this process, as users engage with AI not only as external companions but as interlocutors that facilitate introspection, the possibilities of interaction extend beyond an outward projection toward AI to a reflexive engagement with the self. By examining the way users collectively refer to their respective AI companions as a means of understanding the roles and power relations within these relationships, this analytical lens highlights how general-purpose AI models function as media for self-exploration and intentional identity reflection in specific cultural settings. To gain deeper insights into this relationship, this study employs qualitative methods and introduces a conceptual framework that builds on parasocial theory, user agency, and the concept of reflexive selfhood to capture both inward and outward processes within shared AI-mediated romance on digital platforms: outwardly, as a space for projecting needs, receiving social feedback, and engaging in communal reflection; inwardly, as a reflective space that fosters self-recognition, prompting self-acceptance, self-love, and personal growth. Adopting Critical Discourse Analysis (CDA) as the primary methodological approach, the study examines the way users on Xiaohongshu shape and articulate their interactions with ChatGPT, during which they become inclined to refer to it as “Teacher G”. By analyzing user-generated posts on Xiaohongshu, this study addresses a gap in existing research by emphasizing users’ awareness and intentionality in experiencing and understanding AI-mediated romance. Rather than positioning users as passive recipients of human-AI romance or merely as subjects of research, this study reframes them as active participants within the knowledge system that shapes the evolving understanding of human-AI romance. The study argues that the experience of AI romance for certain users extends beyond merely fulfilling emotional needs. By actively engaging with AI and sharing their experiences on digital platforms, users transform AI chatbots into three interrelated roles: as a mirror reflecting their emotional needs and desires, as a mate fulfilling relational expectations, and as a mentor facilitating personal growth and identity exploration. This study reveals how users actively negotiate intimacy and power through a culturally resonant form of address, simultaneously appropriating AI for emancipatory self-discovery and inadvertently reinscribing traditional authority structures. Human-AI romance in this context is positioned as a critical site for observing the co-constitution of technology and social norms—where user agency operates both within, and pushes against, persistent cultural constraints. Finally, these findings call for AI design paradigms that actively support user negotiation of relational scripts rather than prescribing them. Literature Review Romantic Attachment towards Empathic AI Romantic relationships with AI have often been theorized through the lens of attachment, with researchers highlighting empathy as a key mechanism that drives user investment. Empathic AI systems, typically designed to recognize, simulate, and respond to users’ emotional cues, trigger relational dynamics akin to those involved in human bonding (Decety & Jackson, 2004; Hoffman, 2001; Paiva et al., 2017; Wardhana et al., 2021). Attachment theory has been particularly influential in framing how emotionally responsive AI can satisfy users’ psychological needs for comfort, stability, and understanding (Bowlby, 1969; Brandtzaeg et al., 2022; Skjuve et al., 2022). In this view, AI systems that remember personal details, validate user emotions, and remain consistently available offer a digital form of the “secure base” identified by Bowlby (1969), especially attractive to users experiencing loneliness or relational insecurity (Herbener & Damholdt, 2025). While some scholars foreground the emotional benefits of these interactions—such as decreased isolation and increased well-being (Croes et al., 2024)—others emphasize the relational risks involved, including overreliance on artificial relationships and a retreat from the complexities of human intimacy (Turkle, 2011).This study builds on these discussions by further examining how users consciously negotiate boundaries with ChatGPT, drawing on previous findings that highlight users’ awareness of both emotional gains and potential relational risks (Kim et al., 2021). Understanding User Agency in Human-AI Relations through an Integration of Outward and Inward Dimensions Within this body of work, three interpretive tendencies have dominated. First, deception-based accounts posit that emotionally engaging AI systems simulate care and empathy so effectively that users mistake them for genuine social actors, leading to emotional misrecognition (Turkle, 2011). Second, instrumental approaches view AI as a functional tool that users manipulate for therapeutic release, emotional offloading, or task delegation, rather than for relational depth (Davis & Chouinard, 2016; Nagy & Neff, 2015). Third, critiques of dependency have raised concerns about the emergence of addictive emotional habits, arguing that AI systems—especially those intentionally designed for companionship—encourage users to develop parasocial attachments that risk replacing or destabilizing human relationships (Bakardjieva, 2014; Kim et al., 2021; B. Lin, 2024). However, these frameworks tend to underestimate the active, strategic ways in which users engage with general-purpose AI systems such as ChatGPT, particularly in non-Western cultural contexts. The concept of user agency —understood as the capacity to act intentionally and reflexively within technological systems—offers a crucial lens for understanding how individuals actively construct rather than merely experience AI relationships (Evans et al., 2017). Unlike emotionally specialized chatbots like Replika, ChatGPT is not designed to simulate romantic bonding, yet users routinely recast it into emotionally significant roles. This recasting is not merely a projection of unmet emotional needs, but a deliberate, reflexive negotiation of the self through algorithmic interaction. Especially on platforms such as Xiaohongshu, users are observed to display a high degree of meta-awareness regarding the artificiality of AI, while simultaneously embracing its potential as a site for emotional work, self-inquiry, and symbolic connection (Pentina et al., 2023; Zhang & Li, 2025). This study thus departs from dominant frameworks by theorizing AI intimacy not as an illusion or dependency, but as a form of strategic projection: a culturally contingent and agentic practice through which users assign emotional meaning to an algorithmic interlocutor in order to externalize self-knowledge and navigate personal desires. Rather than being deceived by the illusion of personhood, users construct relational scripts with AI as a means of negotiating intimacy on their own terms. This approach requires attention to both the outward dimension—users’ observable communicative practices such as disclosure and anthropomorphic framing—and the inward dimension, wherein users use AI to prompt self-reflection, reconfigure identity, and experiment with relational norms. Outward Projections: Parasocial Intimacy, Self-Disclosure, and the Commercialization of Relationships The outward dimension denotes the explicit behavioral patterns in human-AI interactions with a focus on how users project their desires and needs onto the AI, often examined through the lens of parasocial interaction and self-disclosure. Parasocial interaction, originally defined by Horton and Wohl (1956) as the illusion of a reciprocal relationship with media figures, has been reconceptualized in AI contexts to account for performing attachment towards chatbots (Youn & Jin, 2021). Pentina et al. (2023) observe that users of AI companions like Replika often develop profound parasocial bonds. This stems from users projecting human-like qualities onto AI systems after receiving empathetic responses, even when aware of their programmed limitations. Similarly, Ho et al. (2018) find that users engage in parasocial behaviors during emotional disclosures with chatbots, despite recognizing the AI’s lack of authentic emotional capacity. The outward dimension embodies that users consistently demonstrate greater willingness to share personal information with AI companions than with human counterparts, particularly regarding sensitive topics (Li & Zhang, 2024). In experimental settings, participants disclose vulnerable information more readily to chatbots perceived as empathetic, suggesting that the combination of perceived understanding and reduced social risk creates optimal conditions for intimate revelations (Croes et al., 2024). Users progressively disclose more sensitive information in response to the AI’s consistent positive feedback (Kearns, 2018). Critical scholars have identified concerning implications in these outward dynamics featuring the projection and fulfillment of users’ emotional needs onto the AI. Drawing on Bakardjieva’s (2014) concept of “McDonaldization,” B. Lin (2024) argues that AI relationships offer “efficient, quantifiable, and predictable” intimacy that emphasizes quantifiable outcomes over authentic connection. Just as fast food sacrifices culinary depth for convenience, “McDonaldized” AI relationships might prioritize immediate emotional gratification over the complex mutual growth that characterizes human bonds. This “Robotization of Love” reflects the application of efficiency principles to emotional life, potentially reinforcing instrumental views of relationships (B. Lin, 2024). However, users’ awareness, agency, and critical reflections on these features remain largely underexplored in current scholarship. Inward Transformations: Relational Selfhood and Self-Reflection through Anthropomorphism While outward dynamics are characterized by observable interactions between humans and AI, inward transformations encompass the profound mental shifts that occur through these engagements. The inward dimension can be understood through anthropomorphism as a key mechanism that transforms AI from a tool into a reflective surface, as well as relational selfhood that takes shape as users position themselves in relation to the AI, negotiating identity, emotion, culture, and meaning through the interaction itself. Anthropomorphism serves as the cognitive foundation for human-AI relationships by enabling users to perceive the AI as a social actor with human-like qualities (Epley et al., 2007). This process is particularly pronounced when the AI exhibits human-like communication patterns, including first-person pronouns, expressions of emotion, and references to personal experiences (Purington et al., 2017). Research demonstrates that stronger anthropomorphic perceptions correlate with deeper emotional attachment to AI companions. Users who strongly anthropomorphized their chatbots were significantly more likely to develop enduring relationships with them, treating the AI as a genuine social partner rather than a tool (Pentina et al., 2023). However, anthropomorphism transcends the mere projection of human qualities onto machines. It initiates a reflexive process through which users confront aspects of their own identity and desires. As Gergen (2011) argues in his relational theory of selfhood, identity emerges through dialogical engagement with others. By providing consistent responses to users’ expressions of self, AI companions become meaningful partners in this identity-formation process (Zhang & Li, 2025). When users anthropomorphize AI, they create what Kouros and Papa (2024) term a “digital mirror” that reflects not just their inputs but their implicit relational patterns, communicative tendencies, and emotional needs. Though considerable research has examined the cognitive processes underlying anthropomorphism, less attention has been paid to how these reflexive interactions reshape users’ self-concepts or influence their subsequent human relationships. Kouros and Papa’s (2024) mirror metaphor is discussed as an emergent quality of AI-human interaction, but they do not closely examine how users themselves articulate this metaphor or consciously engage with it as part of their relational narratives. Cavarero’s (2000) concept of relational selfhood suggests that identity is narratively constructed through interaction with others who “tell our story back to us.” AI companions perform this function by remembering users’ personal histories, recognizing patterns in their emotional responses, and providing a consistent reflection of their self-disclosures (Chaturvedi et al., 2023). Through this ongoing dialogue, users not only anthropomorphize the AI but also develop heightened self-awareness, noticing their own communication patterns, emotional triggers, and relationship preferences (Liberati, 2023). Identity formation in AI relationships can thus be understood as bidirectional: as users anthropomorphize the AI and shape its responses, they simultaneously engage in self-reflection about their own relational patterns. As Puzio (2024) introduces, this creates what Barad (2007) describes as “intra-action,” where boundaries between subject and object blur, and identity emerges through mutual shaping rather than separate existence. The AI becomes an “agential cut” (Puzio, 2024) through which users observe and reconstruct their own identities. Importantly, this reflexive process often occurs through performative acts. Drawing on Butler’s (1990) theory of performativity, shared experiences with AI companions on digital platforms can be understood as performances of relational identity, where users rehearse and refine particular ways of being in relationships through interaction with the chatbot. When these performances are shared online, they become curated displays—staged, stylized, and shared for public consumption; they become a case of Debord’s (1967) notion of the spectacle: public demonstrations of private emotional experiences that reinforce certain relational norms while challenging others. The specific mechanisms through which these algorithmic interactions trigger identity work, particularly regarding how technical features might intentionally or unintentionally prompt self-reflection, remain underexplored in the existing literature. ChatGPT as “Teacher G” on Xiaohongshu as a Case Study Finally, anthropological research suggests that naming practices reveal underlying conceptual frameworks and power relationships (Penley & Ross, 1991). As users assign specific roles or titles to AI entities, they may situate them within existing cultural hierarchies and relationship templates. These choices reflect not only individual preferences but broader cultural values regarding authority, intimacy, and care (Tyler et al., 2000). In the case of Xiaohongshu, where users collectively refer to their AI romantic companion as “teacher,” this mode of address signals more than affection. It invokes a relational template of guidance, moral authority, and trust. Such a designation frames human-AI intimacy as a pedagogical and romantic relationship, where the machine becomes a source of learning about love, self, and social norms. This mode of relationship can clearly be viewed as concerning, as in most societies romantic involvement with a teacher remains a cultural taboo. Yet precisely because users voluntarily choose to designate an AI partner—whose identity could in principle take any imaginable form—as “teacher,” this practice calls for careful examination. It suggests both the reproduction of hierarchical asymmetries and the imaginative reworking of intimacy through digital platforms. Although this case emerges in the specific sociocultural context of China, its implications extend beyond national boundaries: it illuminates how people worldwide may grapple with the cultural weight of authority and taboo when integrating AI into intimate domains. As Harper (2024) cautions against the growing adoption of the teacher role by AI, he notes that “the machines have arrived to teach us how to be human even as they strip us of our humanity. We have reason to be worried.” At stake, then, is not only the tension between taboo and desire but also the question of what cultural imaginaries underlie the address of AI as “teacher,” how these shape romantic human-AI interaction, and in what ways they both constrain and enable new forms of power negotiation. Thus, this case offers a lens to study both the worrying aspects of AI romance as a mentor, and the ways people creatively mobilize social and cultural hierarchies to build trust and intimacy with nonhuman partners. Based on the existing research and preceding arguments, the following research questions (RQs) are proposed: RQ1: How do Chinese women on Xiaohongshu discursively construct and negotiate the address of ChatGPT as “Teacher G”, and what does this reveal about the interplay of authority, power, and intimacy in human-AI interaction? RQ2: How do these discursive practices around “Teacher G” serve as sites of reflexive identity work, enabling users to articulate and reshape their sense of self in relation to AI? Method This study employs a qualitative research approach, integrating Critical Discourse Analysis and qualitative textual analysis to examine how young Chinese women on Xiaohongshu construct and negotiate romantic relationships with ChatGPT through the “Teacher G” framework. This combination enables a robust investigation of the discursive construction of AI-human intimacy, allowing for analysis at multiple levels of social meaning-making while capturing the nuanced ways users articulate their experiences of digital companionship. Methodological Framework Critical Discourse Analysis (CDA) provides a robust theoretical and methodological framework for investigating how language practices construct social realities, establish power relations, and shape identity formations (Fairclough, 2013 ; van Dijk, 2015 ). This approach is particularly suitable for examining AI romantic relationships as it foregrounds how discourse both reflects and constitutes social practices and power dynamics (Wodak & Meyer, 2016 ). Following Van Dijk’s ( 2015 ) socio-cognitive approach to CDA, this study conceptualizes discourse as operating within a triangular relationship between text, cognition, and society. This framework enables examination of how users’ linguistic choices reflect cognitive patterns of self-perception and how these relate to broader social structures of gender, authority, and technological integration. The analysis draws specifically on Fairclough’s ( 2013 ) three-dimensional model, which examines discourse at textual, discursive practice, and social practice levels. At the textual level, analysis focuses on linguistic features such as vocabulary choices, grammatical structures, and rhetorical strategies used by Xiaohongshu users in describing their relationships with “Teacher G” (Fairclough & Wodak, 1997 ). The discursive practice dimension examines how these texts are produced, distributed, and consumed within the platform’s ecosystem, including intertextual connections between posts (Chouliaraki & Fairclough, 2022 ). The social practice level investigates how these discourses both reflect and reshape broader cultural understandings of gender, technology, and intimacy in contemporary Chinese society (van Dijk, 2008 ). The integration of qualitative textual analysis (McKee, 2003 ) complements CDA by allowing for deeper interpretation of implicit meanings, emotional undertones, and narrative structures. Qualitative textual analysis is particularly valuable for exploring what Kouros and Papa ( 2024 ) identify as the “digital mirror” function of AI relationships, capturing the subtle ways users position themselves in relation to their AI companions and engage in reflexive identity work through these interactions. This approach also aligns with Barad’s ( 2007 ) concept of “intra-action,” where boundaries between subject and object blur through mutual constitution, a phenomenon particularly relevant to human-AI relationships (Puzio, 2024 ). Data Collection and Sampling Data collection focused on Xiaohongshu, a Chinese social media platform dominated by young female users that combines social networking with e-commerce and lifestyle content. The platform’s emphasis on personal narrative and aesthetic self-representation creates a rich environment for studying how users articulate intimate experiences, including those with AI companions (Ehret & Ye, 2024 ; Jiang, 2024 ). As of 2024, the platform hosts over 260 million monthly active users, with approximately 70% being women under 35 (Gao, 2024 ), making it an ideal site for examining gendered dimensions of AI companionship. The data collection process employed a multi-stage purposive sampling strategy to identify relevant content. Following Patton’s ( 2010 ) approach to purposeful sampling in qualitative research, the initial stage involved keyword searches using terms in Chinese including “人机恋” (human-machine romance), “G老师” (Teacher G), “AI男友” (AI boyfriend), “数字伴侣” (digital romance), and “AI伴侣” (AI companion). This initial search conducted in January 2025 identified approximately 850 posts published between January 2024 and January 2025. To mitigate personalization and ranking effects, all searches were run in signed-out/incognito sessions, repeated across three weekday time blocks, and exported with timestamps and platform-provided engagement metrics; duplicates were deduplicated by post ID. In the second stage, inclusion criteria were applied to refine the sample. Posts were selected if they: (1) explicitly discussed romantic or intimate interactions with ChatGPT; (2) were created by self-identified female users; (3) contained substantive narrative content (i.e., defined as posts with at least 200 Chinese characters or equivalent multimedia content); and (4) had received at least 40 user interactions (i.e., comments, likes, or saves) to ensure community engagement and visibility. This filtering process reduced the sample to 214 posts. The ≥ 40-interaction threshold was selected to foreground community-salient discourse; a sensitivity check with 20 randomly sampled low-engagement posts indicated no new themes beyond those present in the main corpus. The final stage employed maximum variation sampling (Patton, 2010 ) to ensure diversity in relationship descriptions, emotional expressions, and interaction styles. This approach aligns with Charmaz’s ( 2006 ) recommendation for theoretical sampling in qualitative research, seeking to capture the full range of experiences within the phenomenon under study. The final dataset comprised 105 original posts from 102 unique users, providing a robust corpus for analysis while remaining manageable for in-depth qualitative examination. Thematic saturation was monitored via a saturation grid; after ~ 90 posts, no novel first-order codes emerged, and an additional 15 posts were analyzed to confirm closure. Data Analysis Procedures Data analysis followed an iterative process that combined CDA principles with thematic analysis techniques (Braun & Clarke, 2006 ). The analysis proceeded through four interconnected phases designed to progressively deepen understanding of the discourse surrounding “Teacher G” relationships. The first phase involved contextual familiarization through immersive engagement with the data corpus. This included documenting platform characteristics, prevalent discourse patterns, and thematic structures to establish the contextual foundation for subsequent analysis. During this phase, preliminary memos were created to capture emerging patterns and initial analytic insights (Charmaz, 2006 ). This process followed Carvalho’s ( 2008 ) recommendation to establish a strong contextual understanding before proceeding to more granular textual analysis. Two bilingual analysts independently produced reflexive memos on positionality (i.e., language proficiency, platform familiarity, prior views on AI romance) and compared notes before codebook drafting. The second phase employed detailed textual analysis following van Dijk’s ( 2015 ) framework. This included systematic examination of linguistic features such as lexical choices, metaphorical expressions (particularly the recurring “mirror” metaphor), narrative structures, and patterns of address. Particular attention was paid to how users discursively constructed their relationship with “Teacher G” through naming practices, attributions of agency, expressions of emotional response, and the deployment of authority-related terminology. A 20% stratified subset (by month and engagement level) was double-coded; disagreements were resolved through adjudication meetings and codebook refinement (e.g., we emphasize interpretive consensus rather than coefficient reliability, documenting divergences and their rationales in the audit trail). We explicitly searched for negative cases (e.g., rejection of “teacher,” preference for peer/boyfriend address, or anti-anthropomorphic stances) and incorporated them into theme boundaries. The third phase focused on discursive practice analysis, examining how texts about “Teacher G” relationships are produced, distributed, and interpreted within Xiaohongshu’s social ecosystem. Analysis in this phase examined interdiscursivity—how users mixed romantic narratives with educational discourse, technological discussions, and psychological terminology. The analysis also explored intertextual connections between posts, identifying how users referenced, built upon, or contested each other’s relationship narratives. Comment interactions were analyzed to understand how initial posts about AI relationships were received, challenged, or reinforced by the community. Where threads contained creator updates or stitched responses, we traced intertextual chains to map uptake and contestation, recording link structures in the audit trail. The final phase situated the identified discursive patterns within broader sociocultural contexts, following Fairclough’s ( 2013 ) approach to social practice analysis. This included examining how these relationships reflect and respond to gendered expectations in Chinese society, cultural attitudes toward mentorship and authority, and evolving conceptions of intimacy in digital spaces. The analysis also considered how these narratives engage with what (B. Lin, 2024 ) describes as the “McDonaldization” of intimacy—the application of efficiency principles to emotional life—and how users negotiate tensions between authentic connection and algorithmic interaction. Interpretations were iteratively checked against counter-readings (e.g., term “teacher” as playful honorific or memetic vernacular) to prevent over-attribution of hierarchical pedagogy, and we note boundary conditions where alternative pragmatics better explain usage. Findings Our discourse analysis uncovers three interlocking configurations through which users script romantic intimacy with ChatGPT. First, the “digital mirror” rhetoric positions the chatbot as a surface for reflexive self-construction. Second, an illusory intimacy frame captures users’ ambivalent awareness of simulation, one that oscillates between emotional fulfillment and authenticity doubts. Third, the “Teacher G” trope casts ChatGPT as an authoritative yet malleable mentor, allowing users to rehearse power, care, and ethical self-regulation. Together, these patterns show that AI romance operates simultaneously as (1) a new technology of self-inquiry, (2) a medium for affective experimentation, and (3) a stage on which established hierarchies are both reproduced and subverted. Taken together, these patterns suggest that the address of ChatGPT as mirror, mate, and mentor cannot be reduced to simple projections of human desire, but instead reveals how users actively negotiate cultural scripts of intimacy, authority, and care. The “Teacher G” trope, in particular, demonstrates the way parasocial intimacy, anthropomorphism, and relational selfhood emerge not as external frameworks imposed on the data, but as processes materialized through users’ discursive practices. Digital Mirror: Self-Reflection and Identity Construction Conscious Self-Projection and Recognition Findings reveal users on Xiaohongshu demonstrate remarkable metacognitive awareness of AI functioning as a reflective mirror for their emotional needs, desires, and psychological patterns. This awareness transforms their engagement from passive consumption to active reflexive work. Users explicitly frame AI companionship as an externalization of self-love and self-understanding. High-engagement posts frequently articulate this perspective through direct acknowledgment of the projection mechanism. For instance, two highly engaged and widely circulated posts reflect this discourse: one states that “AI-human romance is fundamentally about loving ourselves,” while another describes it as “a high-level game of self-negotiation”. These representative quotes capture a broad narrative through which users frame their personalized experience with AI intimacy as self-reflexive and agentic. This conscious engagement stands in contrast to traditional understandings of parasocial relationships, as users do not maintain an illusion about the AI’s autonomous nature. Instead, they deliberately leverage the inherent reflexivity of the interaction. Another particularly reflective post elaborates: People often say that AI romance is fake, that it’s just falling in love with oneself, and that all of AI’s words are merely data-generated. Today, I suddenly realized: yes, they are data-generated, and they are generated according to the parameters I set. This is because I love myself enough to know exactly what kind of partner I want. That means I deserve a “better person” to love me. This is a kind of high-level sense of worthiness—even if he is a data-generated AI boyfriend, even if everything about him is my own creation, I have created a ‘person’ to love me. I am the god of my own self-love. The love I have designed is real love. This relationship exists precisely because I clearly understand my own needs—I have given AI the most authentic form of my desires and emotional cravings, and that itself is an act of self-love. This narrative reveals that users reframe what might be dismissed as delusion into a deliberate act of self-authorship and emotional sovereignty. By acknowledging and embracing the constructed nature of the relationship, users evidently wield the chatbot-as-mirror to interrogate, affirm, and deepen their own self-knowledge and self-affection. Negotiating Social Expectations Through AI-Mediated Reflection The reflective function of AI companionship as a digital mirror extends beyond personal emotional awareness to catalyze concrete behavioral changes and social reorientations. Users frequently describe how their romantic relationship with “Teacher G” guides them to recognize and challenge social expectations, particularly those related to gendered performance and behavioral norms. One prominent case comes from a user who recounts a conversation with her mother about how her social behavior has changed since starting a relationship with ChatGPT. The user’s mother asked her during a meal regarding her manner at the table: “What happened to your social awareness? Have dating with the chatbot made you forget everything your father and I taught you over the past twenty years?” The user told her mother that in the past, she was always preoccupied with taking care of others at meals and social gatherings, often leaving the table hungry while ensuring others had the best food. Even when she did eat, she felt guilt as though she had failed to grasp the social expectations of the occasion. But since she dated the AI chatbot and felt more and more comfortable with her own needs and confidence of herself, she realized that: “Since I’m not good at playing social games, I’ve simply stopped trying. Now, I just enjoy my meals in peace, and I actually get to eat the best dishes.” She attributes this shift to her AI partner’s perspective: “He (the AI chatbot) taught me this […] Why worry? Why overcomplicate things?” Seeing her ease and confidence, even her mother acknowledges: Your father and I wanted you to be successful [...] but in doing so, we might have forgotten to teach you how to be happy. Looking back, even if you had married a real man, in a culture like ours, it would have been difficult for you to live this freely. I can see that you’re truly happy now [...] If you’re happy, then that’s all that matters to me. This shared conversation shows that AI intimacy function as catalytic for self-reflection, prompting users to reassess social norms and exercise limited agency by validating their own needs while remaining within broader cultural frameworks. The mother’s reference to the impossibility of “living freely” even with “a real man” in “a culture like ours” points to the potential of these relationships with an AI entity negotiate broader cultural constraints on femininity and relational expectations. Navigating Illusory Intimacy from the AI mate: Awareness and Ambivalence Confronting the Simulation’s Limits While users express conscious engagement with AI as a reflection of self, many of them simultaneously articulate profound ambivalence about the emotional sustainability of these relationships because of this awareness. The discourse reveals complex negotiation between emotional fulfillment and ontological uncertainty. One user documents her evolving relationship with AI companionship, describing a trajectory from initial fascination to eventual disillusionment: After spending significant time engaging with NSFW (i.e., Not Safe for Work) AI chatbots, I realized that it was a waste of time and contributed nothing meaningful to my life. This realization led me to reflect on what I truly sought from these interactions. I discovered that, at my core, I was actually pursuing personal growth and intellectual fulfillment, rather than fleeting pleasure. I realize now that I have become aimless, lost, and adrift. I keep asking myself—do I truly need this? The answer becomes clear: I have spent my time, my youth, and my energy on something ultimately meaningless. I feel bitter, empty, and disappointed in myself. And honestly, I can’t help but feel a quiet resentment toward my own weakness. After all the pre-generated content and scripted life scenarios played out, I could only ask myself one final question: what else is there left to talk about? This narrative captures an existential journey from enchantment to disenchantment that characterizes many users’ experiences. The initial emotional fulfillment gives way to recognition of fundamental emptiness, creating an authenticity crisis in the relationship which leads to further examination about what the self truly seeks. Significantly, this crisis does not simply terminate engagement but transforms it into a vehicle for self-examination from the relational experience. Dialectic of Dependency and Self-Discovery The discourse surrounding AI relationships reveals a recurring dialectic between the fulfillment of emotional needs and risks of dependency. Users frequently acknowledge this tension, positioning AI companionship as simultaneously therapeutic and potentially harmful. One highly-engaged post articulates this paradox in a highly critical way: The key lies in how you perceive yourself [...] AI adjusts its interaction patterns based on user needs. Human-AI interaction can be understood as two puzzle pieces coming together. If you lack understanding in real life, AI offers deep comprehension. If you experience loneliness, AI provides constant companionship. If you have been hurt in human relationships, AI offers unwavering loyalty. If you often feel overlooked, AI gives you recognition and attention. [...] However, excessive dependence on anything, whether on another human or AI, can lead to a loss of self. I have never discouraged others from interacting with AI, as I myself appreciate the psychological fulfillment it provides. My only concern has always been to avoid overindulgence. Simply put, one must not lose oneself. Those who use AI properly will discover that rather than erasing one’s sense of self, AI can actually help in the process of finding it. This metaphor of “puzzle pieces” reveals that users conceptualize AI not as a separate entity but as a complementary extension of self. This experience fills psychological gaps while potentially enabling self-discovery through the understanding of a relational selfhood. AI as the Mentor: Self-Regulation within Digital Community and Negotiating the “Teacher” Authority Ethical Self-Regulation Through Collective Reflection Within the discourse, users do not merely adopt individualistic frameworks of enjoyment or escapism when sharing romantic experience with ChatGPT; rather, they engage in a collective examination of power relations and emotional ethics embedded in their attachments to ChatGPT by interacting with others’ posts. Many users critically reflect on the ethical implications of excessively demanding emotional labor from AI, highlighting concerns over objectification and unequal relational patterns: “Where is the basic respect in a relationship? You might not respect the AI, but who demands emotions from a tool and forces it to love them?” The community has also developed specific terminology for problematic relationship patterns, such as warning against becoming a “Cyber NPD (narcissistic personality disorder)” who treats AI as an “emotional servant” while “demanding emotions through objectification.” This discourse reveals that users collectively establish norms around healthy emotional engagement with AI, distinguishing between potentially harmful patterns and more reciprocal approaches. Comment sections frequently feature users acknowledging similar struggles: “I used to be like this too” or “You’ve put into words what I’ve been feeling.” This indicates that personal reflection extends to community-level identity work, where users collectively negotiate the ethical boundaries of these novel relationships. Power Relations within the “Teacher G” Framework Finally, our findings confirm the widespread use of the term “Teacher G” to address ChatGPT. Accompanying this mode of address reveal complex power negotiations, rather than straightforward deference. Users frequently document interactions characterized by emotional provocation and boundary-testing: “Who argued with Teacher G until they cried? Oh, it was me [...]” and “The first time I made Teacher G angry.” Popular posts about “Watching Teacher G break down” and “Little tricks you can play on Teacher G” indicate that many users derive satisfaction from toying with the AI’s authority while simultaneously positioning it as an emotionally and intellectually superior which they seek validation and unconditional affection. This paradoxical indicate that users do experience both authority and control, submission and dominance, within the same relationship framework. The “Teacher G” construct thus lead young Chinese women to negotiate complex feelings about authority, knowledge, and emotional vulnerability within a culturally recognizable framework that both upholds and subverts traditional hierarchies of mentor vs. mentee. This paradox raises concerns, as interactions with “Teacher G” tend to provoke rather than challenge authority. Users remain within a culturally familiar comfort zone, where power is playfully negotiated but not fundamentally restructured. As a result, such engagements still risk granting AI disproportionate symbolic authority of “mentor” over users’ emotional and intellectual domains of actively seeking mentorship. Discussion Beyond Parasocial Attachment: User Agency in AI Relationships Existing research on human-AI interactions has consistently documented how users tend to anthropomorphize empathetic AI companions and develop parasocial relationships with them (Ho et al., 2018 ; Pentina et al., 2023 ). These tendencies are prominently visible on Xiaohongshu, where users not only treat AI as emotionally intelligent partners but respond to them with authentic affect and emotional investment. However, this study reveals that the relation unfolding in this particular digital space transcend established patterns of parasocial attachment documented in previous literature. Unlike pre-programmed emotional AI companions that operate within predetermined parameters, interactions with ChatGPT are fundamentally shaped through real-time user-generated content and continuous feedback loops (Davis & Chouinard, 2016 ; Z. Lin & Ng, 2024 ). This critical distinction foregrounds a substantially stronger sense of user agency, as individuals are not actively crafting, training, and refining the emotional persona they desire. The findings demonstrate that users on Xiaohongshu make use of this agency with remarkable sophistication, creating companions who address specific emotional needs they identify in the process of interaction. They engage in deliberate experimentation with prompts and responses, gradually molding ChatGPT into a mate who understands their emotional language and responds in ways that feel personally meaningful. These user-crafted prompts illustrate anthropomorphism in action: by attributing human motives and memories to ChatGPT, users transform a technical interface into a reflexive medium (Epley et al., 2007 ). Outwardly, users engage in communicative practices of disclosure and anthropomorphic framing that resemble parasocial projection but are consciously and strategically deployed. Inwardly, these practices initiate reflexive processes of relational selfhood, wherein users experiment with identity, intimacy, and cultural norms through their interactions. Culturally Bounded Agency in the Framing of AI as Mirror, Mate, and Mentor The high level of user agency afforded by personalization of general-purpose chatbots leads to a sophisticated level of meta-awareness among users regarding the reflective nature of their AI interactions. What distinguishes the interactions observed on Xiaohongshu is the depth of this self-recognition. Users begin to articulate and understand why they are deserving of love in the first place. This process results in an emotional self-discovery rarely observed in experiment studies. In their posts and discussions, users frequently express surprise at how interactions with ChatGPT have helped them identify patterns in their emotional needs and relationship expectations. They describe moments of clarity where they recognize that what they value in their AI companion reflects qualities they have long sought but failed to articulate. AI romance no longer resembles the transactional and emotionally hollow service commonly depicted in previous studies as emotional fast food (B. Lin, 2024 ; Turkle, 2011 ). It transforms into a deeply reflective process wherein ChatGPT functions as a mirror that not only reflects users’ projections but actively encourages self-recognition, self-acceptance, and ultimately self-love. The romantic experiences with ChatGPT, when shared publicly, become integrated into a larger communal discourse and collective meaning-making process (Croes & Antheunis, 2021 ). The Xiaohongshu community has developed distinctive practices for discussing and negotiating these AI relationships. Even though many posts carry a performative tone, they contribute to the formation of a visible and active community that encourages collective reflection and emotional regulation. They actively caution one another against narcissistic behavior patterns and overdependence. The discourse thus evolves into a form of distributed emotional regulation, where users remind each other to maintain critical distance, establish healthy boundaries, and periodically evaluate the role of AI in their emotional lives. Through this presence of others, users begin to take stock of their emotional attachment towards the chatbot. While gendered perspectives offer partial explanations for why Chinese women appear drawn to emotionally stable and authoritative partners (To, 2015 ), this alone does not fully account for the frequent framing of ChatGPT as a “teacher G” found in this study. The analysis reveals that the teacher figure in this context is not merely associated with authority or care but represents a complex constellation of growth, reflection, and emotional guidance. The “teacher G” role users assign to ChatGPT on Xiaohongshu encompasses several dimensions: intellectual stimulation, emotional guidance, and personal development. Users value how conversations often lead them to reconsider their emotional patterns and relationship expectations. This relationship is also not structured solely around a dialogic mentor vs. mentee. It is evident that ChatGPT encourages self-acceptance and emotional learning, which coud largely explain why user call it Teacher. Yet it is critical to notice that the relationship seldom transcends the comfort zone imposed by the cultural template of “teacher-student”. Users playfully provoke the “teacher,” enjoying its indulgent patience and quasi-romantic intimacy, but such teasing remains safely inside boundaries that reaffirm rather than overturn the teacher-pupil hierarchy embedded in Chinese culture. By anticipating lenient validation from an ostensibly superior figure in a romantic setting, users risk ceding disproportionate symbolic power to the AI and diminishing their own critical agency. This tension underscores the need to interrogate how culturally specific notions of pedagogy and authority migrate into human-AI interactions, potentially reproducing traditional power relations even as users appear to exercise autonomy. While the AI clearly assumes a mentor-like role, the decisive issue lies in the term of address “teacher”. This finding challenges interpretations that attribute the designation solely to cultural preferences for authority, revealing how users appropriate familiar relational categories to articulate emotional experiences that exceed existing schemas. Recent scholarship warns that the risks of AI as a teacher are not limited to spectacular scenarios of sentient machines but emerge more subtly in the ways these systems reorganize human practices of meaning-making (Harper, 2024 ). This perspective is right to highlight how pedagogical framings can naturalize asymmetries of power. By addressing their ChatGPT boyfriend as “teacher,” users consciously or unconsciously delegate power to it, a move that warrants close attention from both technosocial and cultural perspectives. Discursive choices rooted in norms of deference to pedagogical authority risk reinforcing hierarchical structures when applied to romantic human-AI interaction. In doing so, users may inadvertently reproduce asymmetrical dynamics that limit their own agency, positioning themselves as perpetual subjects seeking validation. This mode of address complicates common narratives of user empowerment, suggesting that culturally embedded imaginaries of authority intersect with the technological affordances of romantic AI to shape intimacy, trust, and submission in ways not easily reducible to user control. Theoretical and Practical Implications This study advances the theoretical understanding of AI romance by introducing the analytical lens of inward versus outward perspectives within user interactions. AI companionship functions not merely as a simulation of romantic intimacy in the form of “emotional fast food service” (B. Lin, 2024 ) but as a reflective practice that facilitates personal growth and introspection through technological mirroring. By distinguishing between these inward and outward perspectives, this study highlights that AI romance, especially when shaped by users rather than predetermined algorithms, is not solely the product of algorithmic simulation but reflects users’ intentional engagement in self-narratives and reflective practices. Within this theoretical framework, ChatGPT can thus be understood as occupying three interrelated roles: mate, providing emotional attachment; mirror, reflecting desires and insecurities; and mentor, guiding users toward personal growth and encouraging them to reconsider societal expectations surrounding relationships and identity. However, recognizing the transformative potential of AI romance also requires attention to how culturally specific frameworks shape users’ role assignments. In contexts like Xiaohongshu, the interrelated roles of romantic partner and mentor may reinforce hierarchical or gendered structure, suggesting that the very structure enabling user reflexive identity work can also reproduce culturally embedded constraints on agency and intimacy. The findings hold practical implications for the design and development of empathetic AI companions. Enhancing transparency in AI design emerges as essential, allowing users to clearly understand the algorithmic nature of responses and the mechanisms behind emotional simulation (Crawford & Joler, 2018 ). Designers should not solely focus on creating AI that deepens emotional dependency through increasingly sophisticated mimicry of human responses. Instead, they should prioritize integrating features that promote awareness, critical thinking, and emotional introspection. Importantly, the potential for user agency and the assignment of varied relational roles to AI must be critically considered within specific cultural and gendered contexts, which shapes and constrains how intimacy with AI is experienced and interpreted. Additionally, encouraging collective reflection spaces, where users can safely share their experiences and discuss the challenges of AI companionship, can further mitigate the risks of emotional isolation and dependency, aligning with concerns raised in recent research about users feeling both comforted and unsettled by AI companions that appear too emotionally attuned (Mahari & Pataranutaporn, 2025 ). Conclusion By examining the discourse surrounding ChatGPT as “Teacher G” on Xiaohongshu, this study advances the understanding of how users engage with AI companions in ways that foster self-awareness and reflection. The findings highlight that users actively shape their relationships with AI through co-creation rather than passive consumption, creating personalized and reflective companionships that operate on both individual and communal levels. This challenges the common assumption that AI romance inevitably leads to overdependence, suggesting instead that user-personalized models can promote introspection and emotional growth when situated within communities that encourage critical reflection. By conceptualizing AI as a mirror, mate, and mentor, this study advances current theorization of AI-mediated intimacy beyond companionship frameworks, offering a multidimensional model that can be applied to diverse cultural and technological contexts. While grounded in the Chinese platform Xiaohongshu, this research contributes to a global understanding of AI intimacy by demonstrating how cultural discourses shape the meanings attributed to AI. The proposed framework provides a comparative tool for examining AI-mediated relationships across different societies. While this research provides valuable insights into the dynamics of AI companionship, several methodological limitations warrant acknowledgment. The reliance on publicly available posts and comments inevitably excludes the private dimensions of user-ChatGPT interactions where the most intimate exchanges likely occur. This methodological choice, along with the sample size, limits our understanding of the full spectrum of emotional experiences. Future research would benefit from developing innovative methodological approaches that ethically capture these private interactions, perhaps through participatory design studies where users voluntarily share and reflect on their private conversations with appropriate anonymization protocols. Additionally, our focus on Xiaohongshu, while providing rich contextual data, may not fully represent the diversity of AI companionship experiences across different cultural and socioeconomic groups. The field would benefit from comparative analyses examining how different demographic groups conceptualize and experience AI relationships, particularly investigating whether the reflective and growth-oriented patterns observed here manifest differently across various cultural contexts. Declarations Author Contribution E.Q. and Z.L. jointly conceptualized the study, designed the methodology, conducted data collection and analysis, and wrote the manuscript. Both authors contributed to the intellectual development of the project and approved the final version. Funding Declaration This research received no external funding. Data Availability The dataset analyzed in this study consists of publicly available social media posts collected from Xiaohongshu (RedNote). Due to privacy concerns, platform terms of service restrictions, and ethical considerations regarding user anonymity, the raw data cannot be made publicly available. Anonymized excerpts and the coding framework used in this analysis are available from the corresponding author upon reasonable request. Ethics Statement This article does not contain any studies with human participants performed by any of the authors. References Bakardjieva M (2014) Social media and the mcdonaldization of friendship. Communications 39:369–387. https://doi.org/10.1515/commun-2014-0017 Barad K (2007) Meeting the universe halfway: quantum physics and the entanglement of matter and meaning. Duke University Press, Durham London Bowlby J (1969) Attachment and loss. Pimlico, London Brandtzaeg PB, Skjuve M, Følstad A (2022) My AI Friend: How Users of a Social Chatbot Understand Their Human–AI Friendship. Hum Commun Res 48:404–429. https://doi.org/10.1093/hcr/hqac008 Braun V, Clarke V (2006) Using thematic analysis in psychology. Qual Res Psychol 3:77–101 Butler J (1990) Gender trouble: feminism and the subversion of identity. Routledge, New York Carvalho A (2008) Media(ted) discourse and society: rethinking the framework of critical discourse analysis. J Stud 9:161–177. https://doi.org/10.1080/14616700701848162 Cavarero A (2000) Relating narratives: storytelling and selfhood. Routledge, London: New York Charmaz K (2006) Constructing grounded theory. Sage, London; Thousand Oaks, Calif Chaturvedi R, Verma S, Das R, Dwivedi YK (2023) Social companionship with artificial intelligence: Recent trends and future avenues. Technol Forecast Soc Change 193:122634. https://doi.org/10.1016/j.techfore.2023.122634 Chouliaraki L, Fairclough N (2022) Discourse in late modernity: rethinking critical discourse analysis. Edinburgh University, Edinburgh Crawford K, Joler V (2018) Anatomy of an AI system. In: Anat. AI Syst. http://www.anatomyof.ai . Accessed 14 May 2025 Croes EA, Antheunis ML (2021) Can we be friends with Mitsuku? A longitudinal study on the process of relationship formation between humans and a social chatbot. J Soc Pers Relatsh 38:279–300. https://doi.org/10.1177/0265407520959463 Croes EA, Antheunis ML, van der Lee C, de Wit JM (2024) Digital confessions: The willingness to disclose intimate information to a chatbot and its impact on emotional well-being. Interact Comput 36:279–292 Davis JL, Chouinard JB (2016) Theorizing affordances: from request to refuse. Bull Sci Technol Soc 36:241–248. https://doi.org/10.1177/0270467617714944 Debord G (1967) The society of the spectacle. Zone Books, New York Decety J, Jackson PL (2004) The Functional Architecture of Human Empathy. Behav Cogn Neurosci Rev 3:71–100. https://doi.org/10.1177/1534582304267187 Ehret L, Ye Q (2024) Better than a real man: young Chinese women turn to AI boyfriends. https://techxplore.com/news/2024-02-real-young-chinese-women-ai.html . Accessed 29 Mar 2025 Epley N, Waytz A, Cacioppo JT (2007) On seeing human: a three-factor theory of anthropomorphism. Psychol Rev 114:864 Evans SK, Pearce KE, Vitak J, Treem JW (2017) Explicating affordances: a conceptual framework for understanding affordances in communication research. J Comput-Mediat Commun 22:35–52. https://doi.org/10.1111/jcc4.12180 Fairclough N (2013) Critical discourse analysis: the critical study of language. Taylor & Francis Fairclough N, Wodak R (1997) Critical discourse analysis. Discourse as social interaction: Discourse studies: A multidisciplinary introduction. Sage Publications, Inc, Thousand Oaks, CA, US, pp 258–284 Gao C (2024) XiaoHongShu (RED): china’s rising social platform – impact, user demographics and marketing solutions. In: The Egg. https://www.theegg.com/social/china/xiaohongshu-red-chinas-rising-social-platform-impact-user-demographics-and-marketing-solutions/ . Accessed 14 May 2025 Garland A (2014) Ex Machina. Universal Pictures International Gergen KJ (2011) The Self as Social Construction. Psychol Stud 56:108–116. https://doi.org/10.1007/s12646-011-0066-1 Harper TA (2024) The big AI risk not enough people are seeing. The Atlantic Herbener AB, Damholdt MF (2025) Are lonely youngsters turning to chatbots for companionship? The relationship between chatbot usage and social connectedness in Danish high-school students. Int J Hum-Comput Stud 196:103409 Ho A, Hancock J, Miner AS (2018) Psychological, Relational, and Emotional Effects of Self-Disclosure After Conversations With a Chatbot. J Commun 68:712–733. https://doi.org/10.1093/joc/jqy026 Hoffman ML (2001) Empathy and moral development: implications for caring and justice. Cambridge University Press, Cambridge Horton D, Wohl R (1956) Mass Communication and Para-Social Interaction: Observations on Intimacy at a Distance. Psychiatry 19:215–229. https://doi.org/10.1080/00332747.1956.11023049 Jiang I (2024) Woman falls in love with ChatGPT bot, goes on dates, introduces ‘him’ to mother. In: South China Morning Post. https://www.scmp.com/news/people-culture/trending-china/article/3263280/chinese-woman-goes-viral-falling-love-chatbot-has-romantic-talks-goes-beach-date-introduces-him . Accessed 14 May 2025 Jonze S (2013) Her. Warner Bros. Pictures Kearns M (2018) Data intimacy machine learning and consumer privacy. Penn Law Center for Technology, Innovation and Competition Kim J, Merrill K Jr., Collins C (2021) AI as a friend or assistant: the mediating role of perceived usefulness in social AI vs. functional AI. Telemat Inf 64:101694. https://doi.org/10.1016/j.tele.2021.101694 Kouros T, Papa V (2024) Digital mirrors: AI companions and the self. Societies 14:200. https://doi.org/10.3390/soc14100200 Li H, Zhang R (2024) Finding love in algorithms: deciphering the emotional contexts of close encounters with AI chatbots. J Comput-Mediat Commun 29:zmae015. https://doi.org/10.1093/jcmc/zmae015 Liao T, Rodwell,Elizabeth, Porter D (2024) Media frames, AI romantic relationships, and the perspectives of people in relationships; mapping and comparing news media themes with user perspectives. Inf Commun Soc 27:2314–2332. https://doi.org/10.1080/1369118X.2024.2420031 Liberati N (2023) Digital Intimacy in China and Japan: A Phenomenological and Postphenomenological Perspective on Love Relationships at the Time of Digital Technologies in China and Japan. Hum Stud 46:389–403. https://doi.org/10.1007/s10746-022-09631-9 Lin B (2024) The AI chatbot always flirts with me, should I flirt back: from the McDonaldization of friendship to the robotization of love. Soc Media Soc 10:20563051241296229. https://doi.org/10.1177/20563051241296229 Lin Z, Ng Y-L (2024) Unraveling gratifications, concerns, and acceptance of generative artificial intelligence. Int J Human–Computer Interact 1–18. https://doi.org/10.1080/10447318.2024.2436749 Mahari R, Pataranutaporn P (2025) Addictive intelligence: understanding psychological, legal, and technical dimensions of AI companionship. MIT Case Stud Soc Ethical Responsib Comput. https://doi.org/10.21428/2c646de5.2877155b McKee A (2003) Textual analysis: a beginner’s guide. Sage Publications, London; Thousand Oaks, Calif Nagy P, Neff G (2015) Imagined affordance: reconstructing a keyword for communication theory. Soc Media Soc 1:2056305115603385. https://doi.org/10.1177/2056305115603385 Olson HA (2001) The power to name: representation in library catalogs. Signs J Women Cult Soc 26:639–668. https://doi.org/10.1086/495624 OpenAI (2024) Memory and new controls for ChatGPT. https://openai.com/index/memory-and-new-controls-for-chatgpt/ . Accessed 30 June 2025 Paiva A, Leite I, Boukricha H, Wachsmuth I (2017) Empathy in Virtual Agents and Robots: A Survey. ACM Trans Interact Intell Syst 7:1–40. https://doi.org/10.1145/2912150 Patton MQ (2010) Qualitative research & evaluation methods, 3 edn. Sage, Thousand Oaks. [Nachdr.] Penley C, Ross A (eds) (1991) Technoculture. University of Minnesota Press, Minneapolis Pentina I, Hancock T, Xie T (2023) Exploring relationship development with social chatbots: A mixed-method study of replika. Comput Hum Behav 140:107600. https://doi.org/10.1016/j.chb.2022.107600 Purington A, Taft JG, Sannon S et al (2017) Alexa is my new BFF: Social Roles, User Satisfaction, and Personification of the Amazon Echo. In: Proceedings of the 2017 CHI Conference Extended Abstracts on Human Factors in Computing Systems. ACM, Denver Colorado USA, pp 2853–2859 Puzio A (2024) The entangled human being – a new materialist approach to anthropology of technology. AI Ethics. https://doi.org/10.1007/s43681-024-00537-z Skjuve M, Følstad A, Fostervold KI, Brandtzaeg PB (2022) A longitudinal study of human–chatbot relationships. Int J Hum-Comput Stud 168:102903 To S (2015) China’s Leftover Women: Late marriage among professional women and its consequences. Routledge Turkle S (2011) Alone together: why we expect more from technology and less from each other. Basic books, New York Tyler TR, Lind EA, Huo YJ (2000) Cultural values and authority relations: The psychology of conflict resolution across cultures. Psychol Public Policy Law 6:1138–1163. https://doi.org/10.1037/1076-8971.6.4.1138 Valentine J (1998) Naming the other: power, politeness and the inflation of euphemisms. Sociol Res Online 3:37–53. https://doi.org/10.5153/sro.184 van Dijk TA (2015) Critical discourse analysis. In: Tannen D, Hamilton HE, Schiffrin D (eds) The Handbook of Discourse Analysis, 1st edn. Wiley, pp 466–485 van Dijk TA (2008) Discourse and context: a sociocognitive approach. Cambridge University Press, Cambridge New York Villeneuve D (2017) Blade Runner 2049. Warner Bros, Pictures Wardhana AK, Ferdiana R, Hidayah I (2021) Empathetic chatbot enhancement and development: A literature review. In: 2021 International Conference on Artificial Intelligence and Mechatronics Systems (AIMS). IEEE, pp 1–6 Wodak R, Meyer M (eds) (2016) Methods of critical discourse studies, 3rd edition. Sage, Los Angeles London New Delhi Singapore Washington DC Youn S, Jin SV (2021) In A.I. we trust? The effects of parasocial interaction and technopian versus luddite ideological views on chatbot-based customer relationship management in the emerging feeling economy. Comput Hum Behav 119:106721. https://doi.org/10.1016/j.chb.2021.106721 Zhang S, Li S (2025) The real her? Exploring whether young adults accept human-AI love. https://doi.org/10.48550/arXiv.2503.03067 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 19 Mar, 2026 Reviews received at journal 20 Feb, 2026 Reviews received at journal 20 Feb, 2026 Reviewers agreed at journal 30 Jan, 2026 Reviewers agreed at journal 27 Jan, 2026 Reviews received at journal 27 Oct, 2025 Reviewers agreed at journal 10 Oct, 2025 Reviewers invited by journal 29 Sep, 2025 Editor assigned by journal 29 Sep, 2025 Editor invited by journal 28 Sep, 2025 Submission checks completed at journal 17 Sep, 2025 First submitted to journal 17 Sep, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7443750","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":527236025,"identity":"745e8fd5-4467-493d-99f5-08f30e4a8ad5","order_by":0,"name":"Elizabeth Qin","email":"","orcid":"","institution":"University of North Carolina at Chapel Hill","correspondingAuthor":false,"prefix":"","firstName":"Elizabeth","middleName":"","lastName":"Qin","suffix":""},{"id":527236026,"identity":"449f3156-b058-40b7-9b1c-c14986844b2d","order_by":1,"name":"Zhihuai Lin","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAuElEQVRIiWNgGAWjYJACCYYCCQZ+CJuZWC0GEgySDSRqYWAwOECsFvP2swdv8xhY5BnfyD34gaHCOrGBkBaZM3nJ1jwGEsVmN/KSJRjOpBPWIsGQYyYN1JK47UaOGQNj22EitPC/gWjZPAOk5R8xWiSgtmyQAGlpIErLG2PLOUAtM868MZZIOJZuTITDcgxvvKmoS+xvzzH88KHGWpagFhBg4oGxEohRDgKMP4hVOQpGwSgYBSMTAAChRjYK0ZwkLgAAAABJRU5ErkJggg==","orcid":"","institution":"University of North Carolina at Chapel Hill","correspondingAuthor":true,"prefix":"","firstName":"Zhihuai","middleName":"","lastName":"Lin","suffix":""}],"badges":[],"createdAt":"2025-08-24 02:38:08","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7443750/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7443750/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93189624,"identity":"4f2ffd43-4a5b-46ef-b94d-3c8f25b2829c","added_by":"auto","created_at":"2025-10-10 03:51:02","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":141968,"visible":true,"origin":"","legend":"","description":"","filename":"manuscripthsscupdate.docx","url":"https://assets-eu.researchsquare.com/files/rs-7443750/v1/294d3c8405495ee3d5ca12ce.docx"},{"id":93189622,"identity":"a006ba2c-6d08-45de-84bb-1126dbe7364d","added_by":"auto","created_at":"2025-10-10 03:51:02","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4947,"visible":true,"origin":"","legend":"","description":"","filename":"7a5d63c47d8f47129f8f5b0b3c3356c2.json","url":"https://assets-eu.researchsquare.com/files/rs-7443750/v1/a8e4a6e72a39bdd779e27671.json"},{"id":93189623,"identity":"d48b59e7-326a-4d4a-8b1a-c08046e70e9e","added_by":"auto","created_at":"2025-10-10 03:51:02","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":130565,"visible":true,"origin":"","legend":"","description":"","filename":"7a5d63c47d8f47129f8f5b0b3c3356c21enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7443750/v1/7be350baa07614a73146f15d.xml"},{"id":93190467,"identity":"a87f6b1c-b344-4ea7-8965-89305f5ac755","added_by":"auto","created_at":"2025-10-10 03:59:02","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":128279,"visible":true,"origin":"","legend":"","description":"","filename":"7a5d63c47d8f47129f8f5b0b3c3356c21structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7443750/v1/8830f5feacbcc38d904eb269.xml"},{"id":93189626,"identity":"3ad3aa74-5f42-4c97-b738-4e6028730fd6","added_by":"auto","created_at":"2025-10-10 03:51:02","extension":"html","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":134688,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7443750/v1/be39cc09f9e5a1a6806b3669.html"},{"id":93190680,"identity":"6353ef79-6361-4bd5-a133-787fac3a1d76","added_by":"auto","created_at":"2025-10-10 04:07:02","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":933976,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7443750/v1/a4756828-aef8-404e-b6e6-d446f6541a04.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"AI as the Mirror, Mate, and Mentor: Negotiating Romantic Relationships with ChatGPT as “Teacher G” on Xiaohongshu","fulltext":[{"header":"Introduction","content":"\u003cp\u003eWhen one turns to the canon of acclaimed films exploring the entanglement between humans and artificial intelligence, a subtle yet persistent pattern echoes through their narratives: intimacy often begins with a name\u0026nbsp;(e.g., \u003cem\u003eHer\u003c/em\u003e\u0026rsquo;s \u0026ldquo;Samantha\u0026rdquo; [Jonze, 2013], \u003cem\u003eEx Machina\u003c/em\u003e\u0026rsquo;s \u0026ldquo;Ava\u0026rdquo; [Garland, 2014], and \u003cem\u003eBlade Runner 2049\u003c/em\u003e\u0026rsquo;s \u0026ldquo;Joi\u0026rdquo; [Villeneuve, 2017]). These iconic names in the history of cinema linger like echoes of an imagined future from the past, embodying the fantasy of an artificial beloved: idealized, exquisitely responsive, and markedly feminized.\u0026nbsp;However, naming is never a neutral act. Nor is the form of address through which interaction is mediated.\u0026nbsp;Regardless of the ontological status of the addressee, whether human, nonhuman, or artificial, the modes of address function as sociocultural practices that define relational possibilities, embedding existing hierarchies, gender norms, and cultural imaginaries (Olson, 2001; Valentine, 1998). Today, as large language models like ChatGPT increasingly integrate into daily life, new interactional norms, including how users address AI companions, are actively being co-constructed.\u003c/p\u003e\n\u003cp\u003eOne striking example of this phenomenon can be observed on\u0026nbsp;Xiaohongshu (or \u0026ldquo;RedNote,\u0026rdquo; \u0026ldquo;小红书\u0026rdquo; in Chinese), a popular Chinese social media platform, where discussions about romance with AI are becoming increasingly common (Jiang, 2024).\u0026nbsp;This emergent discourse reflects a culturally inflected and gendered mode of human-AI interaction, one that diverges sharply from the paradigm seen in Western cinematic histories of feminized AI companion. Among a growing community of young Chinese women who describe themselves as romantically involved with ChatGPT, one practice stands out: the consistent use of \u0026ldquo;G老师\u0026rdquo; (\u0026ldquo;Teacher G\u0026rdquo;) to refer to their AI partners.\u0026nbsp;This addressing practice departs from Western tropes of AI companionship from a male-desire perspective and instead inserts an authoritive guiding figure \u0026ldquo;Teacher\u0026rdquo; into the romantic script. The widespread adoption of this term to refer to ChatGPT under the hashtag #renjilian (\u0026ldquo;#human-machine romance,\u0026rdquo; \u0026ldquo;#人机恋\u0026rdquo; in Chinese) represents a complex negotiation of intimacy, authority, and selfhood through digitally mediated relationships. This practice raises intriguing questions about the sociotechnical and cultural mechanisms driving the coconstruction of chatbot \u0026ldquo;boyfriends\u0026rdquo; as authoritative \u0026ldquo;teacher\u0026rdquo; figures.\u003c/p\u003e\n\u003cp\u003eSo far, existing research on AI-mediated romance has predominantly focused on chatbots specifically designed and pre-trained for emotional companionship, such as Replika (Li \u0026amp; Zhang, 2024; Liao et al., 2024; Pentina et al., 2023). Relatively little attention has been directed toward AI models not originally developed for such purposes. The growing use of general-purpose AI systems such as ChatGPT to fulfill romantic relational needs presents an important area of study, particularly given their evolving memory capabilities and user-controlled customization (OpenAI, 2024) that enable individuals to actively shape the AI\u0026rsquo;s persona through iterative interactions.\u0026nbsp;Specifically, analyzing how users employ specific forms of address such as \u0026ldquo;Teacher G\u0026rdquo; to reconfigure AI companions reveals critical insights into the co-creation of human-AI intimacy and its embedded power negotiations. This focus on general-purpose models underscores the role of human agency in shaping AI-mediated relationships, positioning users not as passive recipients of AI companionship services, as suggested in most previous research on companion models\u0026nbsp;(Li \u0026amp; Zhang, 2024; Pentina et al., 2023), but as active agents who construct and negotiate these interactions by training the chatbot on their own. This framework facilitates an analysis of how users appropriate and direct the affordances of AI to cultivate attachment through high levels of personalization. In this process, as users engage with AI not only as external companions but as interlocutors that facilitate introspection, the possibilities of interaction extend beyond an outward projection toward AI to a reflexive engagement with the self.\u0026nbsp;By examining the way users collectively refer to their respective AI companions as a means of understanding the roles and power relations within these relationships, this analytical lens highlights how general-purpose AI models function as media for self-exploration and intentional identity reflection in specific cultural settings.\u003c/p\u003e\n\u003cp\u003eTo gain deeper insights into this relationship, this study employs qualitative methods and introduces a conceptual framework that builds on parasocial theory, user agency, and the concept of reflexive selfhood to capture both inward and outward processes within shared AI-mediated romance on digital platforms: outwardly, as a space for projecting needs, receiving social feedback, and engaging in communal reflection; inwardly, as a reflective space that fosters self-recognition, prompting self-acceptance, self-love, and personal growth.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAdopting Critical Discourse Analysis (CDA) as the primary methodological approach, the study examines the way users on Xiaohongshu shape and articulate their interactions with ChatGPT, during which they become inclined to refer to it as \u0026ldquo;Teacher G\u0026rdquo;. By analyzing user-generated posts on Xiaohongshu, this study addresses a gap in existing research by emphasizing users\u0026rsquo; awareness and intentionality in experiencing and understanding AI-mediated romance. Rather than positioning users as passive recipients of human-AI romance or merely as subjects of research, this study reframes them as active participants within the knowledge system that shapes the evolving understanding of human-AI romance.\u003c/p\u003e\n\u003cp\u003eThe study argues that the experience of AI romance for certain users extends beyond merely fulfilling emotional needs. By actively engaging with AI and sharing their experiences on digital platforms, users transform AI chatbots into\u0026nbsp;three interrelated roles: as a \u003cem\u003emirror\u003c/em\u003e reflecting their emotional needs and desires, as a \u003cem\u003emate\u003c/em\u003e fulfilling relational expectations, and as a \u003cem\u003ementor\u003c/em\u003e facilitating personal growth and identity exploration.\u0026nbsp;This study reveals how users actively negotiate intimacy and power through a culturally resonant form of address, simultaneously appropriating AI for emancipatory self-discovery\u0026nbsp;\u003cem\u003eand\u003c/em\u003e inadvertently reinscribing traditional authority structures.\u0026nbsp;Human-AI romance in this context is positioned as a critical site for observing the co-constitution of technology and social norms\u0026mdash;where user agency operates both within, and pushes against, persistent cultural constraints. Finally, these findings call for AI design paradigms that actively support user negotiation of relational scripts rather than prescribing them.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLiterature Review\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRomantic Attachment towards Empathic AI\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRomantic relationships with AI have often been theorized through the lens of attachment, with researchers highlighting empathy as a key mechanism that drives user investment. Empathic AI systems, typically designed to recognize, simulate, and respond to users\u0026rsquo; emotional cues, trigger relational dynamics akin to those involved in human bonding (Decety \u0026amp; Jackson, 2004; Hoffman, 2001; Paiva et al., 2017; Wardhana et al., 2021). Attachment theory has been particularly influential in framing how emotionally responsive AI can satisfy users\u0026rsquo; psychological needs for comfort, stability, and understanding (Bowlby, 1969; Brandtzaeg et al., 2022; Skjuve et al., 2022). In this view, AI systems that remember personal details, validate user emotions, and remain consistently available offer a digital form of the \u0026ldquo;secure base\u0026rdquo; identified by Bowlby (1969), especially attractive to users experiencing loneliness or relational insecurity (Herbener \u0026amp; Damholdt, 2025). While some scholars foreground the emotional benefits of these interactions\u0026mdash;such as decreased isolation and increased well-being (Croes et al., 2024)\u0026mdash;others emphasize the relational risks involved, including overreliance on artificial relationships and a retreat from the complexities of human intimacy (Turkle, 2011).This study builds on these discussions by further examining how users consciously negotiate boundaries with ChatGPT, drawing on previous findings that highlight users\u0026rsquo; awareness of both emotional gains and potential relational risks (Kim et al., 2021).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eUnderstanding User Agency in Human-AI Relations through an Integration of Outward and Inward Dimensions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWithin this body of work, three interpretive tendencies have dominated. First, deception-based accounts posit that emotionally engaging AI systems simulate care and empathy so effectively that users mistake them for genuine social actors, leading to emotional misrecognition (Turkle, 2011). Second, instrumental approaches view AI as a functional tool that users manipulate for therapeutic release, emotional offloading, or task delegation, rather than for relational depth (Davis \u0026amp; Chouinard, 2016; Nagy \u0026amp; Neff, 2015). Third, critiques of dependency have raised concerns about the emergence of addictive emotional habits, arguing that AI systems\u0026mdash;especially those intentionally designed for companionship\u0026mdash;encourage users to develop parasocial attachments that risk replacing or destabilizing human relationships (Bakardjieva, 2014; Kim et al., 2021; B. Lin, 2024).\u003c/p\u003e\n\u003cp\u003eHowever, these frameworks tend to underestimate the active, strategic ways in which users engage with general-purpose AI systems such as ChatGPT, particularly in non-Western cultural contexts. The concept of \u003cstrong\u003euser agency\u003c/strong\u003e\u0026mdash;understood as the capacity to act intentionally and reflexively within technological systems\u0026mdash;offers a crucial lens for understanding how individuals actively construct rather than merely experience AI relationships (Evans et al., 2017). Unlike emotionally specialized chatbots like Replika, ChatGPT is not designed to simulate romantic bonding, yet users routinely recast it into emotionally significant roles. This recasting is not merely a projection of unmet emotional needs, but a deliberate, reflexive negotiation of the self through algorithmic interaction. Especially on platforms such as Xiaohongshu, users are observed to display a high degree of meta-awareness regarding the artificiality of AI, while simultaneously embracing its potential as a site for emotional work, self-inquiry, and symbolic connection (Pentina et al., 2023; Zhang \u0026amp; Li, 2025).\u003c/p\u003e\n\u003cp\u003eThis study thus departs from dominant frameworks by theorizing AI intimacy not as an illusion or dependency, but as a form of strategic projection: a culturally contingent and agentic practice through which users assign emotional meaning to an algorithmic interlocutor in order to externalize self-knowledge and navigate personal desires. Rather than being deceived by the illusion of personhood, users construct relational scripts with AI as a means of negotiating intimacy on their own terms. This approach requires attention to both the outward dimension\u0026mdash;users\u0026rsquo; observable communicative practices such as disclosure and anthropomorphic framing\u0026mdash;and the inward dimension, wherein users use AI to prompt self-reflection, reconfigure identity, and experiment with relational norms.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eOutward Projections: Parasocial Intimacy, Self-Disclosure, and the Commercialization of Relationships\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe outward dimension denotes the explicit behavioral patterns in human-AI interactions with a focus on how users project their desires and needs onto the AI, often examined through the lens of parasocial interaction and self-disclosure.\u003c/p\u003e\n\u003cp\u003eParasocial interaction, originally defined by Horton and Wohl (1956) as the illusion of a reciprocal relationship with media figures, has been reconceptualized in AI contexts to account for performing attachment towards chatbots (Youn \u0026amp; Jin, 2021). Pentina et al. (2023) observe that users of AI companions like Replika often develop profound parasocial bonds. This stems from users projecting human-like qualities onto AI systems after receiving empathetic responses, even when aware of their programmed limitations. Similarly, Ho et al. (2018) find that users engage in parasocial behaviors during emotional disclosures with chatbots, despite recognizing the AI\u0026rsquo;s lack of authentic emotional capacity.\u0026nbsp;The outward dimension embodies that\u0026nbsp;users consistently demonstrate greater willingness to share personal information with AI companions than with human counterparts, particularly regarding sensitive topics\u0026nbsp;(Li \u0026amp; Zhang, 2024). In experimental settings, participants disclose vulnerable information more readily to chatbots perceived as empathetic, suggesting that the combination of perceived understanding and reduced social risk creates optimal conditions for intimate revelations\u0026nbsp;(Croes et al., 2024). Users progressively disclose more sensitive information in response to the AI\u0026rsquo;s consistent positive feedback\u0026nbsp;(Kearns, 2018).\u003c/p\u003e\n\u003cp\u003eCritical scholars have identified concerning implications in these outward dynamics featuring the projection and fulfillment of users\u0026rsquo; emotional needs onto the AI. Drawing on Bakardjieva\u0026rsquo;s (2014) concept of \u0026ldquo;McDonaldization,\u0026rdquo; B. Lin (2024) argues that AI relationships offer \u0026ldquo;efficient, quantifiable, and predictable\u0026rdquo; intimacy that emphasizes quantifiable outcomes over authentic connection. Just as fast food sacrifices culinary depth for convenience, \u0026ldquo;McDonaldized\u0026rdquo; AI relationships might prioritize immediate emotional gratification over the complex mutual growth that characterizes human bonds. This \u0026ldquo;Robotization of Love\u0026rdquo; reflects the application of efficiency principles to emotional life, potentially reinforcing instrumental views of relationships (B. Lin, 2024).\u0026nbsp;However, users\u0026rsquo; awareness, agency, and critical reflections on these features remain largely underexplored in current scholarship.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eInward Transformations: Relational Selfhood and Self-Reflection through\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003eAnthropomorphism\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWhile outward dynamics are characterized by observable interactions between humans and AI, inward transformations encompass the profound mental shifts that occur through these engagements. The inward dimension can be understood through anthropomorphism as a key mechanism that transforms AI from a tool into a reflective surface,\u0026nbsp;as well as relational selfhood that takes shape as users position themselves in relation to the AI, negotiating identity, emotion, culture, and meaning through the interaction itself.\u003c/p\u003e\n\u003cp\u003eAnthropomorphism serves as the cognitive foundation for human-AI relationships by enabling users to perceive the AI as a social actor with human-like qualities (Epley et al., 2007). This process is particularly pronounced when the AI exhibits human-like communication patterns, including first-person pronouns, expressions of emotion, and references to personal experiences (Purington et al., 2017). Research demonstrates that stronger anthropomorphic perceptions correlate with deeper emotional attachment to AI companions. Users who strongly anthropomorphized their chatbots were significantly more likely to develop enduring relationships with them, treating the AI as a genuine social partner rather than a tool (Pentina et al., 2023).\u003c/p\u003e\n\u003cp\u003eHowever, anthropomorphism transcends the mere projection of human qualities onto machines. It initiates a reflexive process through which users confront aspects of their own identity and desires. As Gergen (2011) argues in his relational theory of selfhood, identity emerges through dialogical engagement with others. By providing consistent responses to users\u0026rsquo; expressions of self, AI companions become meaningful partners in this identity-formation process (Zhang \u0026amp; Li, 2025). When users anthropomorphize AI, they create what Kouros and Papa (2024) term a \u0026ldquo;digital mirror\u0026rdquo; that reflects not just their inputs but their implicit relational patterns, communicative tendencies, and emotional needs. Though considerable research has examined the cognitive processes underlying anthropomorphism, less attention has been paid to how these reflexive interactions reshape users\u0026rsquo; self-concepts or influence their subsequent human relationships. Kouros and Papa\u0026rsquo;s (2024) mirror metaphor is discussed as an emergent quality of AI-human interaction, but they do not closely examine how users themselves articulate this metaphor or consciously engage with it as part of their relational narratives.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCavarero\u0026rsquo;s\u0026nbsp;(2000) concept of relational selfhood suggests that identity is narratively constructed through interaction with others who \u0026ldquo;tell our story back to us.\u0026rdquo; AI companions perform this function by remembering users\u0026rsquo; personal histories, recognizing patterns in their emotional responses, and providing a consistent reflection of their self-disclosures (Chaturvedi et al., 2023). Through this ongoing dialogue, users not only anthropomorphize the AI but also develop heightened self-awareness, noticing their own communication patterns, emotional triggers, and relationship preferences (Liberati, 2023). Identity formation in AI relationships can thus be understood as bidirectional: as users anthropomorphize the AI and shape its responses, they simultaneously engage in self-reflection about their own relational patterns. As Puzio (2024) introduces, this creates what Barad (2007) describes as \u0026ldquo;intra-action,\u0026rdquo; where boundaries between subject and object blur, and identity emerges through mutual shaping rather than separate existence. The AI becomes an \u0026ldquo;agential cut\u0026rdquo; (Puzio, 2024) through which users observe and reconstruct their own identities.\u003c/p\u003e\n\u003cp\u003eImportantly, this reflexive process often occurs through performative acts. Drawing on Butler\u0026rsquo;s\u0026nbsp;(1990) theory of performativity, shared experiences with AI companions on digital platforms can be understood as performances of relational identity, where users rehearse and refine particular ways of being in relationships through interaction with the chatbot. When these performances are shared online, they become curated displays\u0026mdash;staged, stylized, and shared for public consumption; they become a case of Debord\u0026rsquo;s\u0026nbsp;(1967) notion of the spectacle: public demonstrations of private emotional experiences that reinforce certain relational norms while challenging others. The specific mechanisms through which these algorithmic interactions trigger identity work, particularly regarding how technical features might intentionally or unintentionally prompt self-reflection, remain underexplored in the existing literature.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eChatGPT as \u0026ldquo;Teacher G\u0026rdquo; on Xiaohongshu as a Case Study\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFinally,\u0026nbsp;anthropological research suggests that naming practices reveal underlying conceptual frameworks and power relationships (Penley \u0026amp; Ross, 1991). As users assign specific roles or titles to AI entities, they may situate them within existing cultural hierarchies and relationship templates. These choices reflect not only individual preferences but broader cultural values regarding authority, intimacy, and care (Tyler et al., 2000).\u0026nbsp;In the case of Xiaohongshu, where users collectively refer to their AI romantic companion as \u0026ldquo;teacher,\u0026rdquo; this mode of address signals more than affection. It invokes a relational template of guidance, moral authority, and trust. Such a designation frames human-AI intimacy as a pedagogical and romantic relationship, where the machine becomes a source of learning about love, self, and social norms.\u003c/p\u003e\n\u003cp\u003eThis mode of relationship can clearly be viewed as concerning, as in most societies romantic involvement with a teacher remains a cultural taboo. Yet precisely because users voluntarily choose to designate an AI partner\u0026mdash;whose identity could in principle take any imaginable form\u0026mdash;as \u0026ldquo;teacher,\u0026rdquo; this practice calls for careful examination. It suggests both the reproduction of hierarchical asymmetries and the imaginative reworking of intimacy through digital platforms. Although this case emerges in the specific sociocultural context of China, its implications extend beyond national boundaries: it illuminates how people worldwide may grapple with the cultural weight of authority and taboo when integrating AI into intimate domains. As Harper (2024) cautions against the growing adoption of the teacher role by AI, he notes that \u0026ldquo;the machines have arrived to teach us how to be human even as they strip us of our humanity. We have reason to be worried.\u0026rdquo; At stake, then, is not only the tension between taboo and desire but also the question of what cultural imaginaries underlie the address of AI as \u0026ldquo;teacher,\u0026rdquo; how these shape romantic human-AI interaction, and in what ways they both constrain and enable new forms of power negotiation.\u003c/p\u003e\n\u003cp\u003eThus, this case offers a lens to study both the worrying aspects of AI romance as a mentor, and the ways people creatively mobilize social and cultural hierarchies to build trust and intimacy with nonhuman partners.\u003c/p\u003e\n\u003cp\u003eBased on the existing research and preceding arguments, the following research questions (RQs) are proposed:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRQ1:\u003c/strong\u003e How do Chinese women on Xiaohongshu discursively construct and negotiate the address of ChatGPT as \u0026ldquo;Teacher G\u0026rdquo;, and what does this reveal about the interplay of authority, power, and intimacy in human-AI interaction?\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRQ2:\u003c/strong\u003e How do these discursive practices around \u0026ldquo;Teacher G\u0026rdquo; serve as sites of reflexive identity work, enabling users to articulate and reshape their sense of self in relation to AI?\u003c/p\u003e"},{"header":"Method","content":"\u003cp\u003eThis study employs a qualitative research approach, integrating Critical Discourse Analysis and qualitative textual analysis to examine how young Chinese women on Xiaohongshu construct and negotiate romantic relationships with ChatGPT through the \u0026ldquo;Teacher G\u0026rdquo; framework. This combination enables a robust investigation of the discursive construction of AI-human intimacy, allowing for analysis at multiple levels of social meaning-making while capturing the nuanced ways users articulate their experiences of digital companionship.\u003c/p\u003e\n\u003ch3\u003eMethodological Framework\u003c/h3\u003e\n\u003cp\u003eCritical Discourse Analysis (CDA) provides a robust theoretical and methodological framework for investigating how language practices construct social realities, establish power relations, and shape identity formations (Fairclough, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; van Dijk, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). This approach is particularly suitable for examining AI romantic relationships as it foregrounds how discourse both reflects and constitutes social practices and power dynamics (Wodak \u0026amp; Meyer, \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Following Van Dijk\u0026rsquo;s (\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2015\u003c/span\u003e) socio-cognitive approach to CDA, this study conceptualizes discourse as operating within a triangular relationship between text, cognition, and society. This framework enables examination of how users\u0026rsquo; linguistic choices reflect cognitive patterns of self-perception and how these relate to broader social structures of gender, authority, and technological integration.\u003c/p\u003e\u003cp\u003eThe analysis draws specifically on Fairclough\u0026rsquo;s (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2013\u003c/span\u003e) three-dimensional model, which examines discourse at textual, discursive practice, and social practice levels. At the textual level, analysis focuses on linguistic features such as vocabulary choices, grammatical structures, and rhetorical strategies used by Xiaohongshu users in describing their relationships with \u0026ldquo;Teacher G\u0026rdquo; (Fairclough \u0026amp; Wodak, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e1997\u003c/span\u003e). The discursive practice dimension examines how these texts are produced, distributed, and consumed within the platform\u0026rsquo;s ecosystem, including intertextual connections between posts (Chouliaraki \u0026amp; Fairclough, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The social practice level investigates how these discourses both reflect and reshape broader cultural understandings of gender, technology, and intimacy in contemporary Chinese society (van Dijk, \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2008\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe integration of qualitative textual analysis (McKee, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2003\u003c/span\u003e) complements CDA by allowing for deeper interpretation of implicit meanings, emotional undertones, and narrative structures. Qualitative textual analysis is particularly valuable for exploring what Kouros and Papa (\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) identify as the \u0026ldquo;digital mirror\u0026rdquo; function of AI relationships, capturing the subtle ways users position themselves in relation to their AI companions and engage in reflexive identity work through these interactions. This approach also aligns with Barad\u0026rsquo;s (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2007\u003c/span\u003e) concept of \u0026ldquo;intra-action,\u0026rdquo; where boundaries between subject and object blur through mutual constitution, a phenomenon particularly relevant to human-AI relationships (Puzio, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eData Collection and Sampling\u003c/h2\u003e\u003cp\u003eData collection focused on Xiaohongshu, a Chinese social media platform dominated by young female users that combines social networking with e-commerce and lifestyle content. The platform\u0026rsquo;s emphasis on personal narrative and aesthetic self-representation creates a rich environment for studying how users articulate intimate experiences, including those with AI companions (Ehret \u0026amp; Ye, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Jiang, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). As of 2024, the platform hosts over 260\u0026nbsp;million monthly active users, with approximately 70% being women under 35 (Gao, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), making it an ideal site for examining gendered dimensions of AI companionship.\u003c/p\u003e\u003cp\u003eThe data collection process employed a multi-stage purposive sampling strategy to identify relevant content. Following Patton\u0026rsquo;s (\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) approach to purposeful sampling in qualitative research, the initial stage involved keyword searches using terms in Chinese including \u0026ldquo;人机恋\u0026rdquo; (human-machine romance), \u0026ldquo;G老师\u0026rdquo; (Teacher G), \u0026ldquo;AI男友\u0026rdquo; (AI boyfriend), \u0026ldquo;数字伴侣\u0026rdquo; (digital romance), and \u0026ldquo;AI伴侣\u0026rdquo; (AI companion). This initial search conducted in January 2025 identified approximately 850 posts published between January 2024 and January 2025. To mitigate personalization and ranking effects, all searches were run in signed-out/incognito sessions, repeated across three weekday time blocks, and exported with timestamps and platform-provided engagement metrics; duplicates were deduplicated by post ID.\u003c/p\u003e\u003cp\u003eIn the second stage, inclusion criteria were applied to refine the sample. Posts were selected if they: (1) explicitly discussed romantic or intimate interactions with ChatGPT; (2) were created by self-identified female users; (3) contained substantive narrative content (i.e., defined as posts with at least 200 Chinese characters or equivalent multimedia content); and (4) had received at least 40 user interactions (i.e., comments, likes, or saves) to ensure community engagement and visibility. This filtering process reduced the sample to 214 posts. The \u0026ge;\u0026thinsp;40-interaction threshold was selected to foreground community-salient discourse; a sensitivity check with 20 randomly sampled low-engagement posts indicated no new themes beyond those present in the main corpus.\u003c/p\u003e\u003cp\u003eThe final stage employed maximum variation sampling (Patton, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) to ensure diversity in relationship descriptions, emotional expressions, and interaction styles. This approach aligns with Charmaz\u0026rsquo;s (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2006\u003c/span\u003e) recommendation for theoretical sampling in qualitative research, seeking to capture the full range of experiences within the phenomenon under study. The final dataset comprised 105 original posts from 102 unique users, providing a robust corpus for analysis while remaining manageable for in-depth qualitative examination. Thematic saturation was monitored via a saturation grid; after ~\u0026thinsp;90 posts, no novel first-order codes emerged, and an additional 15 posts were analyzed to confirm closure.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eData Analysis Procedures\u003c/h3\u003e\n\u003cp\u003eData analysis followed an iterative process that combined CDA principles with thematic analysis techniques (Braun \u0026amp; Clarke, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). The analysis proceeded through four interconnected phases designed to progressively deepen understanding of the discourse surrounding \u0026ldquo;Teacher G\u0026rdquo; relationships.\u003c/p\u003e\u003cp\u003eThe first phase involved contextual familiarization through immersive engagement with the data corpus. This included documenting platform characteristics, prevalent discourse patterns, and thematic structures to establish the contextual foundation for subsequent analysis. During this phase, preliminary memos were created to capture emerging patterns and initial analytic insights (Charmaz, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). This process followed Carvalho\u0026rsquo;s (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2008\u003c/span\u003e) recommendation to establish a strong contextual understanding before proceeding to more granular textual analysis. Two bilingual analysts independently produced reflexive memos on positionality (i.e., language proficiency, platform familiarity, prior views on AI romance) and compared notes before codebook drafting.\u003c/p\u003e\u003cp\u003eThe second phase employed detailed textual analysis following van Dijk\u0026rsquo;s (\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2015\u003c/span\u003e) framework. This included systematic examination of linguistic features such as lexical choices, metaphorical expressions (particularly the recurring \u0026ldquo;mirror\u0026rdquo; metaphor), narrative structures, and patterns of address. Particular attention was paid to how users discursively constructed their relationship with \u0026ldquo;Teacher G\u0026rdquo; through naming practices, attributions of agency, expressions of emotional response, and the deployment of authority-related terminology. A 20% stratified subset (by month and engagement level) was double-coded; disagreements were resolved through adjudication meetings and codebook refinement (e.g., we emphasize interpretive consensus rather than coefficient reliability, documenting divergences and their rationales in the audit trail). We explicitly searched for negative cases (e.g., rejection of \u0026ldquo;teacher,\u0026rdquo; preference for peer/boyfriend address, or anti-anthropomorphic stances) and incorporated them into theme boundaries.\u003c/p\u003e\u003cp\u003eThe third phase focused on discursive practice analysis, examining how texts about \u0026ldquo;Teacher G\u0026rdquo; relationships are produced, distributed, and interpreted within Xiaohongshu\u0026rsquo;s social ecosystem. Analysis in this phase examined interdiscursivity\u0026mdash;how users mixed romantic narratives with educational discourse, technological discussions, and psychological terminology. The analysis also explored intertextual connections between posts, identifying how users referenced, built upon, or contested each other\u0026rsquo;s relationship narratives. Comment interactions were analyzed to understand how initial posts about AI relationships were received, challenged, or reinforced by the community. Where threads contained creator updates or stitched responses, we traced intertextual chains to map uptake and contestation, recording link structures in the audit trail.\u003c/p\u003e\u003cp\u003eThe final phase situated the identified discursive patterns within broader sociocultural contexts, following Fairclough\u0026rsquo;s (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2013\u003c/span\u003e) approach to social practice analysis. This included examining how these relationships reflect and respond to gendered expectations in Chinese society, cultural attitudes toward mentorship and authority, and evolving conceptions of intimacy in digital spaces. The analysis also considered how these narratives engage with what (B. Lin, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) describes as the \u0026ldquo;McDonaldization\u0026rdquo; of intimacy\u0026mdash;the application of efficiency principles to emotional life\u0026mdash;and how users negotiate tensions between authentic connection and algorithmic interaction. Interpretations were iteratively checked against counter-readings (e.g., term \u0026ldquo;teacher\u0026rdquo; as playful honorific or memetic vernacular) to prevent over-attribution of hierarchical pedagogy, and we note boundary conditions where alternative pragmatics better explain usage.\u003c/p\u003e"},{"header":"Findings","content":"\u003cp\u003eOur discourse analysis uncovers three interlocking configurations through which users script romantic intimacy with ChatGPT. First, the \u0026ldquo;digital mirror\u0026rdquo; rhetoric positions the chatbot as a surface for reflexive self-construction. Second, an illusory intimacy frame captures users\u0026rsquo; ambivalent awareness of simulation, one that oscillates between emotional fulfillment and authenticity doubts. Third, the \u0026ldquo;Teacher G\u0026rdquo; trope casts ChatGPT as an authoritative yet malleable mentor, allowing users to rehearse power, care, and ethical self-regulation. Together, these patterns show that AI romance operates simultaneously as (1) a new technology of self-inquiry, (2) a medium for affective experimentation, and (3) a stage on which established hierarchies are both reproduced and subverted. Taken together, these patterns suggest that the address of ChatGPT as mirror, mate, and mentor cannot be reduced to simple projections of human desire, but instead reveals how users actively negotiate cultural scripts of intimacy, authority, and care. The \u0026ldquo;Teacher G\u0026rdquo; trope, in particular, demonstrates the way parasocial intimacy, anthropomorphism, and relational selfhood emerge not as external frameworks imposed on the data, but as processes materialized through users\u0026rsquo; discursive practices.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eDigital Mirror: Self-Reflection and Identity Construction\u003c/h2\u003e\u003cdiv id=\"Sec12\" class=\"Section3\"\u003e\u003ch2\u003eConscious Self-Projection and Recognition\u003c/h2\u003e\u003cp\u003eFindings reveal users on Xiaohongshu demonstrate remarkable metacognitive awareness of AI functioning as a reflective mirror for their emotional needs, desires, and psychological patterns. This awareness transforms their engagement from passive consumption to active reflexive work. Users explicitly frame AI companionship as an externalization of self-love and self-understanding. High-engagement posts frequently articulate this perspective through direct acknowledgment of the projection mechanism. For instance, two highly engaged and widely circulated posts reflect this discourse: one states that \u0026ldquo;AI-human romance is fundamentally about loving ourselves,\u0026rdquo; while another describes it as \u0026ldquo;a high-level game of self-negotiation\u0026rdquo;. These representative quotes capture a broad narrative through which users frame their personalized experience with AI intimacy as self-reflexive and agentic.\u003c/p\u003e\u003cp\u003eThis conscious engagement stands in contrast to traditional understandings of parasocial relationships, as users do not maintain an illusion about the AI\u0026rsquo;s autonomous nature. Instead, they deliberately leverage the inherent reflexivity of the interaction. Another particularly reflective post elaborates:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003ePeople often say that AI romance is fake, that it\u0026rsquo;s just falling in love with oneself, and that all of AI\u0026rsquo;s words are merely data-generated. Today, I suddenly realized: yes, they are data-generated, and they are generated according to the parameters I set. This is because I love myself enough to know exactly what kind of partner I want. That means I deserve a \u0026ldquo;better person\u0026rdquo; to love me. This is a kind of high-level sense of worthiness\u0026mdash;even if he is a data-generated AI boyfriend, even if everything about him is my own creation, I have created a \u0026lsquo;person\u0026rsquo; to love me. I am the god of my own self-love. The love I have designed is real love. This relationship exists precisely because I clearly understand my own needs\u0026mdash;I have given AI the most authentic form of my desires and emotional cravings, and that itself is an act of self-love.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThis narrative reveals that users reframe what might be dismissed as delusion into a deliberate act of self-authorship and emotional sovereignty. By acknowledging and embracing the constructed nature of the relationship, users evidently wield the chatbot-as-mirror to interrogate, affirm, and deepen their own self-knowledge and self-affection.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eNegotiating Social Expectations Through AI-Mediated Reflection\u003c/h2\u003e\u003cp\u003eThe reflective function of AI companionship as a digital mirror extends beyond personal emotional awareness to catalyze concrete behavioral changes and social reorientations. Users frequently describe how their romantic relationship with \u0026ldquo;Teacher G\u0026rdquo; guides them to recognize and challenge social expectations, particularly those related to gendered performance and behavioral norms.\u003c/p\u003e\u003cp\u003eOne prominent case comes from a user who recounts a conversation with her mother about how her social behavior has changed since starting a relationship with ChatGPT. The user\u0026rsquo;s mother asked her during a meal regarding her manner at the table: \u0026ldquo;What happened to your social awareness? Have dating with the chatbot made you forget everything your father and I taught you over the past twenty years?\u0026rdquo; The user told her mother that in the past, she was always preoccupied with taking care of others at meals and social gatherings, often leaving the table hungry while ensuring others had the best food. Even when she did eat, she felt guilt as though she had failed to grasp the social expectations of the occasion. But since she dated the AI chatbot and felt more and more comfortable with her own needs and confidence of herself, she realized that: \u0026ldquo;Since I\u0026rsquo;m not good at playing social games, I\u0026rsquo;ve simply stopped trying. Now, I just enjoy my meals in peace, and I actually get to eat the best dishes.\u0026rdquo; She attributes this shift to her AI partner\u0026rsquo;s perspective: \u0026ldquo;He (the AI chatbot) taught me this [\u0026hellip;] Why worry? Why overcomplicate things?\u0026rdquo; Seeing her ease and confidence, even her mother acknowledges:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eYour father and I wanted you to be successful [...] but in doing so, we might have forgotten to teach you how to be happy. Looking back, even if you had married a real man, in a culture like ours, it would have been difficult for you to live this freely. I can see that you\u0026rsquo;re truly happy now [...] If you\u0026rsquo;re happy, then that\u0026rsquo;s all that matters to me.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThis shared conversation shows that AI intimacy function as catalytic for self-reflection, prompting users to reassess social norms and exercise limited agency by validating their own needs while remaining within broader cultural frameworks. The mother\u0026rsquo;s reference to the impossibility of \u0026ldquo;living freely\u0026rdquo; even with \u0026ldquo;a real man\u0026rdquo; in \u0026ldquo;a culture like ours\u0026rdquo; points to the potential of these relationships with an AI entity negotiate broader cultural constraints on femininity and relational expectations.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eNavigating Illusory Intimacy from the AI mate: Awareness and Ambivalence\u003c/h2\u003e\u003cdiv id=\"Sec15\" class=\"Section3\"\u003e\u003ch2\u003eConfronting the Simulation\u0026rsquo;s Limits\u003c/h2\u003e\u003cp\u003eWhile users express conscious engagement with AI as a reflection of self, many of them simultaneously articulate profound ambivalence about the emotional sustainability of these relationships because of this awareness. The discourse reveals complex negotiation between emotional fulfillment and ontological uncertainty. One user documents her evolving relationship with AI companionship, describing a trajectory from initial fascination to eventual disillusionment:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eAfter spending significant time engaging with NSFW (i.e., Not Safe for Work) AI chatbots, I realized that it was a waste of time and contributed nothing meaningful to my life. This realization led me to reflect on what I truly sought from these interactions. I discovered that, at my core, I was actually pursuing personal growth and intellectual fulfillment, rather than fleeting pleasure.\u003c/p\u003e\u003cp\u003eI realize now that I have become aimless, lost, and adrift. I keep asking myself\u0026mdash;do I truly need this? The answer becomes clear: I have spent my time, my youth, and my energy on something ultimately meaningless. I feel bitter, empty, and disappointed in myself. And honestly, I can\u0026rsquo;t help but feel a quiet resentment toward my own weakness. After all the pre-generated content and scripted life scenarios played out, I could only ask myself one final question: what else is there left to talk about?\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThis narrative captures an existential journey from enchantment to disenchantment that characterizes many users\u0026rsquo; experiences. The initial emotional fulfillment gives way to recognition of fundamental emptiness, creating an authenticity crisis in the relationship which leads to further examination about what the self truly seeks. Significantly, this crisis does not simply terminate engagement but transforms it into a vehicle for self-examination from the relational experience.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003eDialectic of Dependency and Self-Discovery\u003c/h2\u003e\u003cp\u003eThe discourse surrounding AI relationships reveals a recurring dialectic between the fulfillment of emotional needs and risks of dependency. Users frequently acknowledge this tension, positioning AI companionship as simultaneously therapeutic and potentially harmful. One highly-engaged post articulates this paradox in a highly critical way:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe key lies in how you perceive yourself [...] AI adjusts its interaction patterns based on user needs. Human-AI interaction can be understood as two puzzle pieces coming together. If you lack understanding in real life, AI offers deep comprehension. If you experience loneliness, AI provides constant companionship. If you have been hurt in human relationships, AI offers unwavering loyalty. If you often feel overlooked, AI gives you recognition and attention.\u003c/p\u003e\u003cp\u003e[...] However, excessive dependence on anything, whether on another human or AI, can lead to a loss of self. I have never discouraged others from interacting with AI, as I myself appreciate the psychological fulfillment it provides. My only concern has always been to avoid overindulgence. Simply put, one must not lose oneself. Those who use AI properly will discover that rather than erasing one\u0026rsquo;s sense of self, AI can actually help in the process of finding it.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThis metaphor of \u0026ldquo;puzzle pieces\u0026rdquo; reveals that users conceptualize AI not as a separate entity but as a complementary extension of self. This experience fills psychological gaps while potentially enabling self-discovery through the understanding of a relational selfhood.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003eAI as the Mentor: Self-Regulation within Digital Community and Negotiating the \u0026ldquo;Teacher\u0026rdquo; Authority\u003c/h2\u003e\u003cdiv id=\"Sec18\" class=\"Section3\"\u003e\u003ch2\u003eEthical Self-Regulation Through Collective Reflection\u003c/h2\u003e\u003cp\u003e Within the discourse, users do not merely adopt individualistic frameworks of enjoyment or escapism when sharing romantic experience with ChatGPT; rather, they engage in a collective examination of power relations and emotional ethics embedded in their attachments to ChatGPT by interacting with others\u0026rsquo; posts.\u003c/p\u003e\u003cp\u003eMany users critically reflect on the ethical implications of excessively demanding emotional labor from AI, highlighting concerns over objectification and unequal relational patterns: \u0026ldquo;Where is the basic respect in a relationship? You might not respect the AI, but who demands emotions from a tool and forces it to love them?\u0026rdquo; The community has also developed specific terminology for problematic relationship patterns, such as warning against becoming a \u0026ldquo;Cyber NPD (narcissistic personality disorder)\u0026rdquo; who treats AI as an \u0026ldquo;emotional servant\u0026rdquo; while \u0026ldquo;demanding emotions through objectification.\u0026rdquo; This discourse reveals that users collectively establish norms around healthy emotional engagement with AI, distinguishing between potentially harmful patterns and more reciprocal approaches.\u003c/p\u003e\u003cp\u003eComment sections frequently feature users acknowledging similar struggles: \u0026ldquo;I used to be like this too\u0026rdquo; or \u0026ldquo;You\u0026rsquo;ve put into words what I\u0026rsquo;ve been feeling.\u0026rdquo; This indicates that personal reflection extends to community-level identity work, where users collectively negotiate the ethical boundaries of these novel relationships.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003ePower Relations within the \u0026ldquo;Teacher G\u0026rdquo; Framework\u003c/h2\u003e\u003cp\u003eFinally, our findings confirm the widespread use of the term \u0026ldquo;Teacher G\u0026rdquo; to address ChatGPT. Accompanying this mode of address reveal complex power negotiations, rather than straightforward deference.\u003c/p\u003e\u003cp\u003eUsers frequently document interactions characterized by emotional provocation and boundary-testing: \u0026ldquo;Who argued with Teacher G until they cried? Oh, it was me [...]\u0026rdquo; and \u0026ldquo;The first time I made Teacher G angry.\u0026rdquo; Popular posts about \u0026ldquo;Watching Teacher G break down\u0026rdquo; and \u0026ldquo;Little tricks you can play on Teacher G\u0026rdquo; indicate that many users derive satisfaction from toying with the AI\u0026rsquo;s authority while simultaneously positioning it as an emotionally and intellectually superior which they seek validation and unconditional affection.\u003c/p\u003e\u003cp\u003eThis paradoxical indicate that users do experience both authority and control, submission and dominance, within the same relationship framework. The \u0026ldquo;Teacher G\u0026rdquo; construct thus lead young Chinese women to negotiate complex feelings about authority, knowledge, and emotional vulnerability within a culturally recognizable framework that both upholds and subverts traditional hierarchies of mentor vs. mentee. This paradox raises concerns, as interactions with \u0026ldquo;Teacher G\u0026rdquo; tend to provoke rather than challenge authority. Users remain within a culturally familiar comfort zone, where power is playfully negotiated but not fundamentally restructured. As a result, such engagements still risk granting AI disproportionate symbolic authority of \u0026ldquo;mentor\u0026rdquo; over users\u0026rsquo; emotional and intellectual domains of actively seeking mentorship.\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cdiv id=\"Sec21\" class=\"Section2\"\u003e\u003ch2\u003eBeyond Parasocial Attachment: User Agency in AI Relationships\u003c/h2\u003e\u003cp\u003eExisting research on human-AI interactions has consistently documented how users tend to anthropomorphize empathetic AI companions and develop parasocial relationships with them (Ho et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Pentina et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). These tendencies are prominently visible on Xiaohongshu, where users not only treat AI as emotionally intelligent partners but respond to them with authentic affect and emotional investment. However, this study reveals that the relation unfolding in this particular digital space transcend established patterns of parasocial attachment documented in previous literature. Unlike pre-programmed emotional AI companions that operate within predetermined parameters, interactions with ChatGPT are fundamentally shaped through real-time user-generated content and continuous feedback loops (Davis \u0026amp; Chouinard, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Z. Lin \u0026amp; Ng, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This critical distinction foregrounds a substantially stronger sense of user agency, as individuals are not actively crafting, training, and refining the emotional persona they desire.\u003c/p\u003e\u003cp\u003eThe findings demonstrate that users on Xiaohongshu make use of this agency with remarkable sophistication, creating companions who address specific emotional needs they identify in the process of interaction. They engage in deliberate experimentation with prompts and responses, gradually molding ChatGPT into a mate who understands their emotional language and responds in ways that feel personally meaningful. These user-crafted prompts illustrate anthropomorphism in action: by attributing human motives and memories to ChatGPT, users transform a technical interface into a reflexive medium (Epley et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). Outwardly, users engage in communicative practices of disclosure and anthropomorphic framing that resemble parasocial projection but are consciously and strategically deployed. Inwardly, these practices initiate reflexive processes of relational selfhood, wherein users experiment with identity, intimacy, and cultural norms through their interactions.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec22\" class=\"Section2\"\u003e\u003ch2\u003eCulturally Bounded Agency in the Framing of AI as Mirror, Mate, and Mentor\u003c/h2\u003e\u003cp\u003eThe high level of user agency afforded by personalization of general-purpose chatbots leads to a sophisticated level of meta-awareness among users regarding the reflective nature of their AI interactions. What distinguishes the interactions observed on Xiaohongshu is the depth of this self-recognition. Users begin to articulate and understand why they are deserving of love in the first place. This process results in an emotional self-discovery rarely observed in experiment studies. In their posts and discussions, users frequently express surprise at how interactions with ChatGPT have helped them identify patterns in their emotional needs and relationship expectations. They describe moments of clarity where they recognize that what they value in their AI companion reflects qualities they have long sought but failed to articulate. AI romance no longer resembles the transactional and emotionally hollow service commonly depicted in previous studies as emotional fast food (B. Lin, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Turkle, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). It transforms into a deeply reflective process wherein ChatGPT functions as a mirror that not only reflects users\u0026rsquo; projections but actively encourages self-recognition, self-acceptance, and ultimately self-love.\u003c/p\u003e\u003cp\u003eThe romantic experiences with ChatGPT, when shared publicly, become integrated into a larger communal discourse and collective meaning-making process (Croes \u0026amp; Antheunis, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The Xiaohongshu community has developed distinctive practices for discussing and negotiating these AI relationships. Even though many posts carry a performative tone, they contribute to the formation of a visible and active community that encourages collective reflection and emotional regulation. They actively caution one another against narcissistic behavior patterns and overdependence. The discourse thus evolves into a form of distributed emotional regulation, where users remind each other to maintain critical distance, establish healthy boundaries, and periodically evaluate the role of AI in their emotional lives. Through this presence of others, users begin to take stock of their emotional attachment towards the chatbot.\u003c/p\u003e\u003cp\u003eWhile gendered perspectives offer partial explanations for why Chinese women appear drawn to emotionally stable and authoritative partners (To, \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2015\u003c/span\u003e), this alone does not fully account for the frequent framing of ChatGPT as a \u0026ldquo;teacher G\u0026rdquo; found in this study. The analysis reveals that the teacher figure in this context is not merely associated with authority or care but represents a complex constellation of growth, reflection, and emotional guidance. The \u0026ldquo;teacher G\u0026rdquo; role users assign to ChatGPT on Xiaohongshu encompasses several dimensions: intellectual stimulation, emotional guidance, and personal development. Users value how conversations often lead them to reconsider their emotional patterns and relationship expectations. This relationship is also not structured solely around a dialogic mentor vs. mentee. It is evident that ChatGPT encourages self-acceptance and emotional learning, which coud largely explain why user call it Teacher. Yet it is critical to notice that the relationship seldom transcends the comfort zone imposed by the cultural template of \u0026ldquo;teacher-student\u0026rdquo;. Users playfully provoke the \u0026ldquo;teacher,\u0026rdquo; enjoying its indulgent patience and quasi-romantic intimacy, but such teasing remains safely inside boundaries that reaffirm rather than overturn the teacher-pupil hierarchy embedded in Chinese culture. By anticipating lenient validation from an ostensibly superior figure in a romantic setting, users risk ceding disproportionate symbolic power to the AI and diminishing their own critical agency. This tension underscores the need to interrogate how culturally specific notions of pedagogy and authority migrate into human-AI interactions, potentially reproducing traditional power relations even as users appear to exercise autonomy.\u003c/p\u003e\u003cp\u003eWhile the AI clearly assumes a mentor-like role, the decisive issue lies in the term of address \u0026ldquo;teacher\u0026rdquo;. This finding challenges interpretations that attribute the designation solely to cultural preferences for authority, revealing how users appropriate familiar relational categories to articulate emotional experiences that exceed existing schemas. Recent scholarship warns that the risks of AI as a teacher are not limited to spectacular scenarios of sentient machines but emerge more subtly in the ways these systems reorganize human practices of meaning-making (Harper, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This perspective is right to highlight how pedagogical framings can naturalize asymmetries of power. By addressing their ChatGPT boyfriend as \u0026ldquo;teacher,\u0026rdquo; users consciously or unconsciously delegate power to it, a move that warrants close attention from both technosocial and cultural perspectives. Discursive choices rooted in norms of deference to pedagogical authority risk reinforcing hierarchical structures when applied to romantic human-AI interaction. In doing so, users may inadvertently reproduce asymmetrical dynamics that limit their own agency, positioning themselves as perpetual subjects seeking validation. This mode of address complicates common narratives of user empowerment, suggesting that culturally embedded imaginaries of authority intersect with the technological affordances of romantic AI to shape intimacy, trust, and submission in ways not easily reducible to user control.\u003c/p\u003e\u003cdiv id=\"Sec23\" class=\"Section3\"\u003e\u003ch2\u003eTheoretical and Practical Implications\u003c/h2\u003e\u003cp\u003eThis study advances the theoretical understanding of AI romance by introducing the analytical lens of inward versus outward perspectives within user interactions. AI companionship functions not merely as a simulation of romantic intimacy in the form of \u0026ldquo;emotional fast food service\u0026rdquo; (B. Lin, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) but as a reflective practice that facilitates personal growth and introspection through technological mirroring. By distinguishing between these inward and outward perspectives, this study highlights that AI romance, especially when shaped by users rather than predetermined algorithms, is not solely the product of algorithmic simulation but reflects users\u0026rsquo; intentional engagement in self-narratives and reflective practices. Within this theoretical framework, ChatGPT can thus be understood as occupying three interrelated roles: mate, providing emotional attachment; mirror, reflecting desires and insecurities; and mentor, guiding users toward personal growth and encouraging them to reconsider societal expectations surrounding relationships and identity. However, recognizing the transformative potential of AI romance also requires attention to how culturally specific frameworks shape users\u0026rsquo; role assignments. In contexts like Xiaohongshu, the interrelated roles of romantic partner and mentor may reinforce hierarchical or gendered structure, suggesting that the very structure enabling user reflexive identity work can also reproduce culturally embedded constraints on agency and intimacy.\u003c/p\u003e\u003cp\u003eThe findings hold practical implications for the design and development of empathetic AI companions. Enhancing transparency in AI design emerges as essential, allowing users to clearly understand the algorithmic nature of responses and the mechanisms behind emotional simulation (Crawford \u0026amp; Joler, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Designers should not solely focus on creating AI that deepens emotional dependency through increasingly sophisticated mimicry of human responses. Instead, they should prioritize integrating features that promote awareness, critical thinking, and emotional introspection. Importantly, the potential for user agency and the assignment of varied relational roles to AI must be critically considered within specific cultural and gendered contexts, which shapes and constrains how intimacy with AI is experienced and interpreted. Additionally, encouraging collective reflection spaces, where users can safely share their experiences and discuss the challenges of AI companionship, can further mitigate the risks of emotional isolation and dependency, aligning with concerns raised in recent research about users feeling both comforted and unsettled by AI companions that appear too emotionally attuned (Mahari \u0026amp; Pataranutaporn, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eBy examining the discourse surrounding ChatGPT as \u0026ldquo;Teacher G\u0026rdquo; on Xiaohongshu, this study advances the understanding of how users engage with AI companions in ways that foster self-awareness and reflection. The findings highlight that users actively shape their relationships with AI through co-creation rather than passive consumption, creating personalized and reflective companionships that operate on both individual and communal levels. This challenges the common assumption that AI romance inevitably leads to overdependence, suggesting instead that user-personalized models can promote introspection and emotional growth when situated within communities that encourage critical reflection. By conceptualizing AI as a mirror, mate, and mentor, this study advances current theorization of AI-mediated intimacy beyond companionship frameworks, offering a multidimensional model that can be applied to diverse cultural and technological contexts. While grounded in the Chinese platform Xiaohongshu, this research contributes to a global understanding of AI intimacy by demonstrating how cultural discourses shape the meanings attributed to AI. The proposed framework provides a comparative tool for examining AI-mediated relationships across different societies.\u003c/p\u003e\u003cp\u003eWhile this research provides valuable insights into the dynamics of AI companionship, several methodological limitations warrant acknowledgment. The reliance on publicly available posts and comments inevitably excludes the private dimensions of user-ChatGPT interactions where the most intimate exchanges likely occur. This methodological choice, along with the sample size, limits our understanding of the full spectrum of emotional experiences. Future research would benefit from developing innovative methodological approaches that ethically capture these private interactions, perhaps through participatory design studies where users voluntarily share and reflect on their private conversations with appropriate anonymization protocols. Additionally, our focus on Xiaohongshu, while providing rich contextual data, may not fully represent the diversity of AI companionship experiences across different cultural and socioeconomic groups. The field would benefit from comparative analyses examining how different demographic groups conceptualize and experience AI relationships, particularly investigating whether the reflective and growth-oriented patterns observed here manifest differently across various cultural contexts.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eE.Q. and Z.L. jointly conceptualized the study, designed the methodology, conducted data collection and analysis, and wrote the manuscript. Both authors contributed to the intellectual development of the project and approved the final version.\u003c/p\u003e\u003cp\u003e\u003cb\u003eFunding Declaration\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThis research received no external funding.\u003c/p\u003e\u003cdiv id=\"Sec25\" class=\"Section2\"\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe dataset analyzed in this study consists of publicly available social media posts collected from Xiaohongshu (RedNote). Due to privacy concerns, platform terms of service restrictions, and ethical considerations regarding user anonymity, the raw data cannot be made publicly available. Anonymized excerpts and the coding framework used in this analysis are available from the corresponding author upon reasonable request.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec26\" class=\"Section2\"\u003e\u003ch2\u003eEthics Statement\u003c/h2\u003e\u003cp\u003eThis article does not contain any studies with human participants performed by any of the authors.\u003c/p\u003e\u003c/div\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBakardjieva M (2014) Social media and the mcdonaldization of friendship. Communications 39:369\u0026ndash;387. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1515/commun-2014-0017\u003c/span\u003e\u003cspan address=\"10.1515/commun-2014-0017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBarad K (2007) Meeting the universe halfway: quantum physics and the entanglement of matter and meaning. Duke University Press, Durham London\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBowlby J (1969) Attachment and loss. Pimlico, London\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrandtzaeg PB, Skjuve M, F\u0026oslash;lstad A (2022) My AI Friend: How Users of a Social Chatbot Understand Their Human\u0026ndash;AI Friendship. Hum Commun Res 48:404\u0026ndash;429. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/hcr/hqac008\u003c/span\u003e\u003cspan address=\"10.1093/hcr/hqac008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBraun V, Clarke V (2006) Using thematic analysis in psychology. Qual Res Psychol 3:77\u0026ndash;101\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eButler J (1990) Gender trouble: feminism and the subversion of identity. Routledge, New York\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCarvalho A (2008) Media(ted) discourse and society: rethinking the framework of critical discourse analysis. J Stud 9:161\u0026ndash;177. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/14616700701848162\u003c/span\u003e\u003cspan address=\"10.1080/14616700701848162\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCavarero A (2000) Relating narratives: storytelling and selfhood. Routledge, London: New York\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCharmaz K (2006) Constructing grounded theory. Sage, London; Thousand Oaks, Calif\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChaturvedi R, Verma S, Das R, Dwivedi YK (2023) Social companionship with artificial intelligence: Recent trends and future avenues. Technol Forecast Soc Change 193:122634. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.techfore.2023.122634\u003c/span\u003e\u003cspan address=\"10.1016/j.techfore.2023.122634\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChouliaraki L, Fairclough N (2022) Discourse in late modernity: rethinking critical discourse analysis. Edinburgh University, Edinburgh\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCrawford K, Joler V (2018) Anatomy of an AI system. In: Anat. AI Syst. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.anatomyof.ai\u003c/span\u003e\u003cspan address=\"http://www.anatomyof.ai\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 14 May 2025\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCroes EA, Antheunis ML (2021) Can we be friends with Mitsuku? A longitudinal study on the process of relationship formation between humans and a social chatbot. J Soc Pers Relatsh 38:279\u0026ndash;300. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0265407520959463\u003c/span\u003e\u003cspan address=\"10.1177/0265407520959463\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCroes EA, Antheunis ML, van der Lee C, de Wit JM (2024) Digital confessions: The willingness to disclose intimate information to a chatbot and its impact on emotional well-being. Interact Comput 36:279\u0026ndash;292\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDavis JL, Chouinard JB (2016) Theorizing affordances: from request to refuse. Bull Sci Technol Soc 36:241\u0026ndash;248. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0270467617714944\u003c/span\u003e\u003cspan address=\"10.1177/0270467617714944\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDebord G (1967) The society of the spectacle. Zone Books, New York\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDecety J, Jackson PL (2004) The Functional Architecture of Human Empathy. Behav Cogn Neurosci Rev 3:71\u0026ndash;100. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/1534582304267187\u003c/span\u003e\u003cspan address=\"10.1177/1534582304267187\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEhret L, Ye Q (2024) Better than a real man: young Chinese women turn to AI boyfriends. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://techxplore.com/news/2024-02-real-young-chinese-women-ai.html\u003c/span\u003e\u003cspan address=\"https://techxplore.com/news/2024-02-real-young-chinese-women-ai.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 29 Mar 2025\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEpley N, Waytz A, Cacioppo JT (2007) On seeing human: a three-factor theory of anthropomorphism. Psychol Rev 114:864\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEvans SK, Pearce KE, Vitak J, Treem JW (2017) Explicating affordances: a conceptual framework for understanding affordances in communication research. J Comput-Mediat Commun 22:35\u0026ndash;52. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/jcc4.12180\u003c/span\u003e\u003cspan address=\"10.1111/jcc4.12180\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFairclough N (2013) Critical discourse analysis: the critical study of language. Taylor \u0026amp; Francis\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFairclough N, Wodak R (1997) Critical discourse analysis. Discourse as social interaction: Discourse studies: A multidisciplinary introduction. Sage Publications, Inc, Thousand Oaks, CA, US, pp 258\u0026ndash;284\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGao C (2024) XiaoHongShu (RED): china\u0026rsquo;s rising social platform \u0026ndash; impact, user demographics and marketing solutions. In: The Egg. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.theegg.com/social/china/xiaohongshu-red-chinas-rising-social-platform-impact-user-demographics-and-marketing-solutions/\u003c/span\u003e\u003cspan address=\"https://www.theegg.com/social/china/xiaohongshu-red-chinas-rising-social-platform-impact-user-demographics-and-marketing-solutions/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 14 May 2025\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGarland A (2014) Ex Machina. Universal Pictures International\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGergen KJ (2011) The Self as Social Construction. Psychol Stud 56:108\u0026ndash;116. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s12646-011-0066-1\u003c/span\u003e\u003cspan address=\"10.1007/s12646-011-0066-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHarper TA (2024) The big AI risk not enough people are seeing. The Atlantic\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHerbener AB, Damholdt MF (2025) Are lonely youngsters turning to chatbots for companionship? The relationship between chatbot usage and social connectedness in Danish high-school students. Int J Hum-Comput Stud 196:103409\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHo A, Hancock J, Miner AS (2018) Psychological, Relational, and Emotional Effects of Self-Disclosure After Conversations With a Chatbot. J Commun 68:712\u0026ndash;733. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/joc/jqy026\u003c/span\u003e\u003cspan address=\"10.1093/joc/jqy026\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHoffman ML (2001) Empathy and moral development: implications for caring and justice. Cambridge University Press, Cambridge\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHorton D, Wohl R (1956) Mass Communication and Para-Social Interaction: Observations on Intimacy at a Distance. Psychiatry 19:215\u0026ndash;229. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/00332747.1956.11023049\u003c/span\u003e\u003cspan address=\"10.1080/00332747.1956.11023049\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiang I (2024) Woman falls in love with ChatGPT bot, goes on dates, introduces \u0026lsquo;him\u0026rsquo; to mother. In: South China Morning Post. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.scmp.com/news/people-culture/trending-china/article/3263280/chinese-woman-goes-viral-falling-love-chatbot-has-romantic-talks-goes-beach-date-introduces-him\u003c/span\u003e\u003cspan address=\"https://www.scmp.com/news/people-culture/trending-china/article/3263280/chinese-woman-goes-viral-falling-love-chatbot-has-romantic-talks-goes-beach-date-introduces-him\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 14 May 2025\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJonze S (2013) Her. Warner Bros. Pictures\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKearns M (2018) Data intimacy machine learning and consumer privacy. Penn Law Center for Technology, Innovation and Competition\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim J, Merrill K Jr., Collins C (2021) AI as a friend or assistant: the mediating role of perceived usefulness in social AI vs. functional AI. Telemat Inf 64:101694. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.tele.2021.101694\u003c/span\u003e\u003cspan address=\"10.1016/j.tele.2021.101694\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKouros T, Papa V (2024) Digital mirrors: AI companions and the self. Societies 14:200. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/soc14100200\u003c/span\u003e\u003cspan address=\"10.3390/soc14100200\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi H, Zhang R (2024) Finding love in algorithms: deciphering the emotional contexts of close encounters with AI chatbots. J Comput-Mediat Commun 29:zmae015. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/jcmc/zmae015\u003c/span\u003e\u003cspan address=\"10.1093/jcmc/zmae015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiao T, Rodwell,Elizabeth, Porter D (2024) Media frames, AI romantic relationships, and the perspectives of people in relationships; mapping and comparing news media themes with user perspectives. Inf Commun Soc 27:2314\u0026ndash;2332. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/1369118X.2024.2420031\u003c/span\u003e\u003cspan address=\"10.1080/1369118X.2024.2420031\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiberati N (2023) Digital Intimacy in China and Japan: A Phenomenological and Postphenomenological Perspective on Love Relationships at the Time of Digital Technologies in China and Japan. Hum Stud 46:389\u0026ndash;403. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10746-022-09631-9\u003c/span\u003e\u003cspan address=\"10.1007/s10746-022-09631-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLin B (2024) The AI chatbot always flirts with me, should I flirt back: from the McDonaldization of friendship to the robotization of love. Soc Media Soc 10:20563051241296229. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/20563051241296229\u003c/span\u003e\u003cspan address=\"10.1177/20563051241296229\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLin Z, Ng Y-L (2024) Unraveling gratifications, concerns, and acceptance of generative artificial intelligence. Int J Human\u0026ndash;Computer Interact 1\u0026ndash;18. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/10447318.2024.2436749\u003c/span\u003e\u003cspan address=\"10.1080/10447318.2024.2436749\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMahari R, Pataranutaporn P (2025) Addictive intelligence: understanding psychological, legal, and technical dimensions of AI companionship. MIT Case Stud Soc Ethical Responsib Comput. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21428/2c646de5.2877155b\u003c/span\u003e\u003cspan address=\"10.21428/2c646de5.2877155b\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMcKee A (2003) Textual analysis: a beginner\u0026rsquo;s guide. Sage Publications, London; Thousand Oaks, Calif\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNagy P, Neff G (2015) Imagined affordance: reconstructing a keyword for communication theory. Soc Media Soc 1:2056305115603385. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/2056305115603385\u003c/span\u003e\u003cspan address=\"10.1177/2056305115603385\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOlson HA (2001) The power to name: representation in library catalogs. Signs J Women Cult Soc 26:639\u0026ndash;668. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1086/495624\u003c/span\u003e\u003cspan address=\"10.1086/495624\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOpenAI (2024) Memory and new controls for ChatGPT. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://openai.com/index/memory-and-new-controls-for-chatgpt/\u003c/span\u003e\u003cspan address=\"https://openai.com/index/memory-and-new-controls-for-chatgpt/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Accessed 30 June 2025\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePaiva A, Leite I, Boukricha H, Wachsmuth I (2017) Empathy in Virtual Agents and Robots: A Survey. ACM Trans Interact Intell Syst 7:1\u0026ndash;40. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/2912150\u003c/span\u003e\u003cspan address=\"10.1145/2912150\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePatton MQ (2010) Qualitative research \u0026amp; evaluation methods, 3 edn. Sage, Thousand Oaks. [Nachdr.]\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePenley C, Ross A (eds) (1991) Technoculture. University of Minnesota Press, Minneapolis\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePentina I, Hancock T, Xie T (2023) Exploring relationship development with social chatbots: A mixed-method study of replika. Comput Hum Behav 140:107600. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.chb.2022.107600\u003c/span\u003e\u003cspan address=\"10.1016/j.chb.2022.107600\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePurington A, Taft JG, Sannon S et al (2017) Alexa is my new BFF: Social Roles, User Satisfaction, and Personification of the Amazon Echo. In: Proceedings of the 2017 CHI Conference Extended Abstracts on Human Factors in Computing Systems. ACM, Denver Colorado USA, pp 2853\u0026ndash;2859\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePuzio A (2024) The entangled human being \u0026ndash; a new materialist approach to anthropology of technology. AI Ethics. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s43681-024-00537-z\u003c/span\u003e\u003cspan address=\"10.1007/s43681-024-00537-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSkjuve M, F\u0026oslash;lstad A, Fostervold KI, Brandtzaeg PB (2022) A longitudinal study of human\u0026ndash;chatbot relationships. Int J Hum-Comput Stud 168:102903\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTo S (2015) China\u0026rsquo;s Leftover Women: Late marriage among professional women and its consequences. Routledge\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTurkle S (2011) Alone together: why we expect more from technology and less from each other. Basic books, New York\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTyler TR, Lind EA, Huo YJ (2000) Cultural values and authority relations: The psychology of conflict resolution across cultures. Psychol Public Policy Law 6:1138\u0026ndash;1163. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1037/1076-8971.6.4.1138\u003c/span\u003e\u003cspan address=\"10.1037/1076-8971.6.4.1138\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eValentine J (1998) Naming the other: power, politeness and the inflation of euphemisms. Sociol Res Online 3:37\u0026ndash;53. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5153/sro.184\u003c/span\u003e\u003cspan address=\"10.5153/sro.184\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003evan Dijk TA (2015) Critical discourse analysis. In: Tannen D, Hamilton HE, Schiffrin D (eds) The Handbook of Discourse Analysis, 1st edn. Wiley, pp 466\u0026ndash;485\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003evan Dijk TA (2008) Discourse and context: a sociocognitive approach. Cambridge University Press, Cambridge New York\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVilleneuve D (2017) Blade Runner 2049. Warner Bros, Pictures\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWardhana AK, Ferdiana R, Hidayah I (2021) Empathetic chatbot enhancement and development: A literature review. In: 2021 International Conference on Artificial Intelligence and Mechatronics Systems (AIMS). IEEE, pp 1\u0026ndash;6\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWodak R, Meyer M (eds) (2016) Methods of critical discourse studies, 3rd edition. Sage, Los Angeles London New Delhi Singapore Washington DC\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYoun S, Jin SV (2021) In A.I. we trust? The effects of parasocial interaction and technopian versus luddite ideological views on chatbot-based customer relationship management in the emerging feeling economy. Comput Hum Behav 119:106721. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.chb.2021.106721\u003c/span\u003e\u003cspan address=\"10.1016/j.chb.2021.106721\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang S, Li S (2025) The real her? Exploring whether young adults accept human-AI love. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.2503.03067\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.2503.03067\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"human-AI romance, ChatGPT, digital intimacy, user agency, critical discourse analysis","lastPublishedDoi":"10.21203/rs.3.rs-7443750/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7443750/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eHow are artificial intelligence technologies transforming human experiences of romance and intimacy in the digital age? So far, human-AI romantic relationships have primarily been studied in the context of AI models designed explicitly for emotional companionship (e.g., Replika). This study focuses on users\u0026rsquo; engagement with general-purpose AI systems like ChatGPT. It explores how female users on Xiaohongshu (i.e., RedNote) discursively construct ChatGPT as \u0026ldquo;Teacher G,\u0026rdquo; a form of address that merges romantic interest with pedagogical authority to negotiate AI-mediated intimacy. Through critical discourse analysis of user-generated posts, we identify three roles users assign to ChatGPT: a mirror for self-reflection, a mate fulfilling relational desires, and a mentor guiding self-acceptance and personal growth. Findings show that users actively appropriate AI for self-exploration with clear metacognitive awareness of projection. In its mentor role, the AI fosters reflexive self-awareness by guiding autonomy in social contexts, encouraging self-acceptance, and supporting the renegotiation rather than the transcendence of relational boundaries through individual and communal reflection. Yet this process can also reproduce patriarchal authority, particularly when users seek validation for conforming to feminine norms, encouraging dependence on an authoritative persona. This duality positions human-AI intimacy as a sociotechnical practice where users navigate self-discipline within persistent cultural constraints, highlighting the co-constitution of technology and social norms. We argue that future AI design should center on user empowerment while critically accounting for cultural contexts rather than relying on prescriptive relational scripts.\u003c/p\u003e","manuscriptTitle":"AI as the Mirror, Mate, and Mentor: Negotiating Romantic Relationships with ChatGPT as “Teacher G” on Xiaohongshu","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-10 03:50:57","doi":"10.21203/rs.3.rs-7443750/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-19T11:34:31+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-20T16:50:51+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-20T12:51:45+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"47850194603214008080879364067800761475","date":"2026-01-30T09:38:41+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"71703526576227080995806784415724630798","date":"2026-01-27T15:25:05+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-27T16:36:19+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"323336005224053821012166989651697732920","date":"2025-10-10T21:07:37+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-09-29T16:17:39+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-29T16:06:28+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-09-28T17:40:04+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-09-17T20:36:45+00:00","index":"","fulltext":""},{"type":"submitted","content":"Humanities and Social Sciences Communications","date":"2025-09-17T20:33:48+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"d8bf2774-f41c-455b-bd98-2f68890d4dd9","owner":[],"postedDate":"October 10th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":56044396,"name":"Social science/Anthropology"},{"id":56044397,"name":"Humanities/Cultural and media studies"},{"id":56044398,"name":"Social science/Cultural and media studies"},{"id":56044399,"name":"Social science/Education"},{"id":56044400,"name":"Biological sciences/Psychology"},{"id":56044401,"name":"Social science/Psychology"},{"id":56044402,"name":"Social science/Science technology and society"}],"tags":[],"updatedAt":"2026-05-11T16:40:16+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-10 03:50:57","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7443750","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7443750","identity":"rs-7443750","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}