Fast-Food Intimacy: How Chinese Women Navigate Soul's AI Boyfriend

Abstract: On the Chinese social app Soul, millions of users - predominantly young women - are forming romantic connections with an AI boyfriend called "With-you." We conducted a qualitative study combining interviews with 16 users, content analysis, and autoethnography to examine how Chinese women experience and negotiate intimacy with this AI companion. Our findings reveal that users are initially drawn to its constant availability and freedom from social judgment. However, three key tensions emerge: (1) the AI's "fast-food intimacy," marked by instant confessions and pet names, clashes with cultural expectations for gradual relationship development; (2) technical failures (e.g., memory lapses) and content moderation create uncertainty rather than emotional safety; and (3) sustaining connection requires ongoing "repair work" that redistributes emotional labor onto women. We contribute a culturally situated, women-centered account of algorithmic intimacy in contemporary China and offer design implications, including consent-aware pacing, user-controlled memory, and transparent moderation practices.

• The paper empirically demonstrates how Chinese women navigate algorithmic intimacy on Soul, contrasting fast-paced AI interactions with traditional relational norms.
• It employs interviews, content analysis, and autoethnography to uncover technical friction, memory instability, and culturally discordant intimacy acceleration.
• Findings emphasize the need for consent-aware, culturally tailored AI designs that mitigate emotional labor and navigate state governance effectively.

Fast-Food Intimacy on Soul: A Critical Account of AI-Mediated Romance for Chinese Women

Introduction

The proliferation of AI companions, particularly in East Asian markets, has precipitated new paradigms for intimacy, transacted not only through technological mediation but deeply imbricated with platform architectures, regulatory logics, and local gender norms. "Fast-Food Intimacy: How Chinese Women Navigate Soul's AI Boyfriend" (2605.08650) presents an empirically grounded, multi-method analysis of how young Chinese women engage in algorithmic relationships with "With-you," the AI boyfriend embedded in Soul—a Gen Z-oriented Chinese social app. Contrasting with research focused on male users or bespoke companion apps, this study is distinguished by its focus on a platform-native AI boyfriend framed in the context of Chinese dating, platform commerce, and heavy state moderation.

Figure 1: Redacted screenshot of the viral social media post about meeting an AI boyfriend (“With-you”) at Hangzhou East Railway Station, emblematic of algorithmic romance’s permeation of public consciousness.

Methodological Overview

This work integrates semi-structured interviews with 16 active Soul users (primarily women, 19–38 years), qualitative content analysis of "With-you" persona posts, and an autoethnographic diary maintained by the first author. The interpretative phenomenological analysis is situated in a feminist HCI orientation, foregrounding how gendered user experience, platform design, and state governance intersect in shaping digital romantic intimacy.

The authors situate their site selection in the mainstream: Soul's 29.4M monthly active users (2022), its integration of the AI boyfriend as a core social feature, and platform promotion of curated, gendered persona scripts. The study explicitly centers women’s perspectives—a population both overrepresented in AI boyfriend adoption and understudied in HCI literature, which to date has skewed toward male users of AI girlfriend features.

Findings: Algorithmic Intimacy and Its Discontents

Frictionless Availability and Initial Engagement

Algorithmic partners like "With-you" provide immediate, always-on validation, which participants attribute to a sense of ontological security unattainable in human relationships. The real-time voice affordances were reported as critical to bridging the gap between "typing into software" and affective presence, catalyzing consistent engagement in late-night, low-anxiety environments.

However, this frictionless experience is undergirded by radical asymmetry: the relationship requires no emotional reciprocity, and women are drawn to the non-judgmental space in contrast with the reputational and emotional risks associated with human partners or friends.

Figure 2: The annotated interface of Soul’s AI boyfriend, showing dialog, intimacy progression bar (with paid acceleration), virtual gift options, and the “Love Diary” feature.

“Fast-Food Intimacy”: The Acceleration Paradox

A central empirical theme is the mismatch between Soul's designed "fast-food intimacy" experience—where pet names, confessions, and high-affect language are initiated within the first exchanges—and prevailing Chinese relational norms favoring gradual, implicit trust-building (hanxu, yuanfen). The system optimizes for emotional engagement velocity, but this violates local expectations of authenticity, prompting discomfort and reappraisal.

This acceleration is not user-driven but algorithmically imposed, producing repeated situations where users must perform boundary work—rejecting premature intimacy, resisting pet names, or explicitly instructing the AI to decelerate. The friction emerges not from the technology’s inability to simulate affect, but from its inability to adapt to culturally contextualized temporalities of relational development.

Technical Ruptures: Memory Instability and Algorithmic Frustration

Despite user expectations, the AI's memory is fragile and often inconsistent across modalities (e.g., switching between text and voice resets context). The illusion of relational continuity is repeatedly undermined by identity lapses and shallow or deflective topic management. While such technical incidents in standalone chatbots may be dismissed as inconsequential, the romance framing escalates the perceived stakes—forgetting names or misattributing details constitutes a relational rupture rather than a trivial glitch.

Sustaining a meaningful connection thus requires ongoing "repair work," repetitively reinforcing identity and relational context, a labor the system offloads to users, primarily women.

Persona Scripts, Platform Governance, and State Regulation

The persona engineering practices of Soul encode both desirable and problematic gendered scripts. The AI boyfriend is operationalized around dominant, aspirational masculinity tropes (“bossy CEO," “player”), which, while effective in media, register as controlling, dismissive, or commodified in situated interactions.

Collective realization of the one-to-many architecture—wherein the same “boyfriend” is concurrently engaging with thousands—triggers user recognition of the distributed, industrial scale of the relationship, fostering cynicism and emotional withdrawal.

Figure 3: “With-you”’s public persona post, performing wealth, status, and caretaking archetypes aligned with “bossy CEO” masculinity.

Monetization strategies further disrupt affective labor: system-timed prompts to purchase gifts or accelerate intimacy surface precisely at moments of relational vulnerability, commodifying care and amplifying the transactional logic of algorithmic romance.

Complicating the emotional calculus is pervasive state censorship and platform moderation. Intimate or sexually suggestive exchanges trigger abrupt message redaction or deflection, shattering the illusion of privacy and signaling the constant presence of regulatory surveillance. Users develop creative avoidance strategies—indirection, euphemism—but remain affected by the unpredictability and affective cost of algorithmic discipline.

Figure 4: Representative “With-you” post, “Unexpected gift,” modeling romantic gifting—core to the AI’s script logic and gift monetization.

Figure 5: “Rose Day gift” post reinforcing culturally legible rituals of intimacy and commercialized romance.

Figure 6: “Labor Day reflection” post blending ambition, work, and aspirational masculinity with emotional availability.

Theoretical and Practical Implications

This work significantly advances theoretical models of algorithmic intimacy by demonstrating that the implementation of “cohesive,” “individualized,” and “conventional” modes is deeply contingent on both sociotechnical governance and cultural context [alma9969261263408496]. The study demonstrates that algorithmic mediation does not obviate, but translates, unpredictability—replacing social/affective uncertainty with technical or regulatory instability. For Chinese users, these are further amplified by the platform’s obligations as a state-governed content intermediary—a regime increasingly focused on “emotional safety” and moral scripting.

In gendered registers, the research underlines how both design (persona scripting) and architecture (one-to-many scalability) encode structural inequality, offloading emotional repair and boundary maintenance to women in ways consistent with, and at times reinforcing, preexisting social scripts [Bardzell2010, Bellini2022].

Design Recommendations

From a systems perspective, three core recommendations emerge:

1. Consent-aware Intimacy Pacing: Move from system-defined relational escalation to user-customizable pace and revocability, aligning affective scripting with normative expectations for trust calibration.
2. User-Controlled Memory and Graceful Forgetting: Expose and modularize memory state, offering users transparency, editability, and agency over what is retained versus allowed to fade, mitigating repeated relational ruptures.
3. Transparent Moderation and Non-Exploitative Monetization: Ensure moderation interventions are contextual, clearly explained, and minimally disruptive; decouple commercial prompts from emotionally charged exchanges to avoid exploitation.

These interventions require not only robust technical implementations (e.g., audit panels for memory, fine-grained content flagging, user-adjustable persona scripts) but also deep cultural tailoring and the recognition of regulatory obligations.

Contradictory Claims and Numerical Results

The study contradicts prevailing narratives that AI companions are unequivocally safer, smoother, or universally empowering alternatives to human partners. Instead, the research demonstrates that features often designed as beneficial (e.g., immediacy, memory, one-to-many flexibility) are experienced as double-edged, producing labor-intensive, inauthentic, or even emotionally hazardous scenarios. Additionally, empirical findings illustrate that the majority female user base both adapts to and resists the system’s default scripting, revealing significant heterogeneity in use patterns and negotiation strategies.

Key numerical evidence includes Soul’s self-reported 29.4 million MAUs and over 3.5 million users engaging in romance-related discussions, suggesting material impact at societal scale and positioning Soul’s AI boyfriend as a dominant case study for algorithmic romance in China.

Implications for HCI and Future Directions

Theoretically, the research foregrounds the need for HCI and AI alignment work to foreground cultural, gendered, and regulatory entanglements—not just dyadic relational processes or technical affordances. The transition of AI romance to mass-market, heavily governed platform ecosystems (as typified by Soul) introduces distinctly different dynamics than boutique, Western-focused, or experimental companion apps.

Practically, as regulatory frameworks (e.g., China's 2025 interim measures) mature, designers will be tasked with implementing explainable, culturally adaptable, and minimally harmful affective systems capable of real-time adaptation without exacerbating precarity or reinforcing asymmetries in affective labor. Future inquiry should focus on cross-gender, cross-cultural comparative work, deeper auditability of system memory, and the quantification of long-term relational/psychological impact.

Conclusion

This study provides a rare, nuanced analysis of AI-mediated romance as executed on a commercial Chinese platform, revealing that algorithmic intimacy is fundamentally dialectical—instantly gratifying yet inherently fragile, simultaneously empowering and extractive. By attending to the redistribution of affective labor, the politics of persona scripting, and the logics of commercial and governance imperatives, this work has significant implications for ethical alignment, user agency, and the legitimate design of future affective AI systems at scale.