PaperlayerView exact revision on arXiv

Source-linked AI summary

Assembling, Breaking, and Refusing the Mask: Agency in AI-Mediated Self-Presentation in Livestreaming

Yang Hong, Nusrat Jahan Mim, Sharifa Sultana

arXiv:2609.28721v1cs.HC

TL;DR

Women livestreamers must navigate industry, platform, and audience expectations while shaping recognizable personas and managing personal boundaries. This mixed-method study examines their masking practices through recruitment-post analysis, interviews, and observations. It finds that masks are assembled and negotiated across bodies, technologies, spaces, performances, and social relations, with breaking and refusal revealing stakes beyond interface control.

  • Problem

    Prior research says less about how women streamers assemble recognizable selves across technologies, bodies, spaces, performances, and social relations, or negotiate visibility demands that conflict with their boundaries.

    Source: §8 · p. 16

  • Method

    The study analyzes recruitment posts and combines livestream observations with interviews to examine Chinese women streamers’ masking practices.

    Source: §3.1.1 · p. 3, §9 · p. 16

  • Results

    Streamers assemble masks across bodies, AI tools, spaces, performances, and social relations, and may modify, break, resist, or refuse them as they negotiate expectations and personal boundaries.

    Source: §1 · p. 2, §9 · p. 16

  • Takeaways & Limitations

    Masking is a sociotechnical negotiation over who shapes visible selves and which forms of femininity become recognizable and valuable, informing theory and platform design.

    Source: §7 · p. 14, §9 · p. 16

    The study focuses primarily on lower-tier Chinese women livestreamers, was based on WeChat recruitment groups and interviews and observations, and does not generalize beyond the studied setting.

    Source: §8 · p. 16

Abstract

from arXiv · show

Our mixed-method study examines how Chinese women livestreamers use masking to construct idealized mediated personas while navigating gendered, commercial, organizational, and platform pressures alongside personal agendas. We built on the concept of masking, analyzed 627 recruitment posts, and conducted livestream observations and interviews with 26 Chinese women streamers. We found that streamers assembled masks across bodies, AI-mediated technologies, spaces, performances, and social relations to become recognizable while protecting personal boundaries. These masks were continually negotiated, and participants sometimes broke, resisted, or refused them when demands became misaligned or unsustainable. We conceptualize masking as a sociotechnical assemblage in which agency lies in preserving, disrupting, and reconfiguring relations rather than controlling a single interface. We further theorize breaking as a consequential part of masking that exposes hidden labor and unequal costs of visibility. We offer theoretical and design directions for more negotiable, contestable, and agency-supporting AI-mediated self- presentation.

1 Introduction

The paper frames masking as how Chinese women livestreamers build recognizable personas while negotiating industry expectations and personal boundaries. It studies how masks are assembled and negotiated, and what breaking or refusing them reveals about agency and visibility.

1 Introduction

  • Industry expectations and personal agendas intersect as streamers use beauty filters, AI avatars, styling, props, surroundings, and strategic performance to shape on-camera personas.Brand-promoting streamers may be expected to match mainstream feminine aesthetics, while managing what they reveal and their privacy, safety, and identity boundaries.Source: §1 · p. 1
  • The study combines analysis of 627 recruitment posts with observations and interviews involving 26 Chinese women livestreamers across e-commerce and entertainment.The supplied method passage begins describing this mixed-method study; the introduction frames its questions around mask assembly, negotiation, breaking, resistance, refusal, and design.Source: §1 · pp. 1–2
  • Masking coordinates bodies, AI tools, spaces, performances, and social relations to make streamers recognizable while they negotiate personal boundaries and industry expectations.Participants may modify company-defined appearances, manage platform and audience expectations, or break, resist, and refuse masking when its demands become untenable.Source: §1 · pp. 1–2
  • The paper reframes agency as preserving, disrupting, and reconfiguring relations among bodies, technologies, spaces, organizations, audiences, and platforms, rather than controlling a single interface.Breaking, resistance, and refusal reveal hidden labor and unequal costs of deviating from expected visibility.Source: §1 · p. 2
  • The study treats masking as a sociotechnical site where visibility, agency, and power are negotiated, not merely a cosmetic or technical function.It offers theoretical and design perspectives on negotiable AI-mediated self-presentation, agency, contestability, and visibility.Source: §1 · p. 2

2 Related Work

The related work connects mediated self-presentation, gendered visibility, and theories of masking to explain why recognizable online selves are socially and technologically shaped. It motivates studying how women negotiate or refuse visibility demands and how masking can inform HCI theory and design.

2 Related Work

  • The related work brings together research on mediated self-presentation, gendered visibility, and masking to motivate questions about assembling, negotiating, resisting, and refusing technologically mediated personas.It asks what these practices contribute to HCI theory and design.Source: §2 · p. 2

2.1 Mediated Self-Presentation in Livestreaming

  • Livestreamers’ visible selves emerge through bodies, technologies, audiences, platforms, and spatial arrangements, while filters, avatars, and face-swapping can transform appearance or detach it from participation.Prior work says less about how streamers assemble recognizable selves across these elements and negotiate the expectations attached to them.Source: §2.1 · p. 2

2.2 Gendered Visibility and Livestreaming

  • Women’s visibility can attract audiences and economic opportunity while also exposing them to scrutiny, harassment, sexualization, stigma, and conflicting expectations about femininity.Existing work leaves less clear how women respond when demanded forms of visibility conflict with their boundaries or become difficult to sustain.Source: §2.2 · pp. 2–3

2.3 Theories of Masking: Identity, Recognition, and Power

  • Masking makes identities recognizable through culturally shared signs, but those signs are shaped by unequal social relations and do not make the person equivalent to the prescribed identity.The paper frames masking as inhabiting available images while negotiating what they signify, rather than simply hiding an authentic self or adopting an artificial one.Source: §2.3 · p. 3

3 Methods

The study combines recruitment-post analysis with interviews and livestream observations to examine industry requirements and women streamers’ masking practices. The authors describe their data collection, analysis, and researcher positionality as part of this approach.

3 Methods

  • The two-phase mixed-methods study combines analysis of 627 recruitment posts with interviews and observations of Chinese women livestreamers.The research received university IRB approval and was conducted between June and August 2026.Source: §3 · p. 3, §3.1.1 · p. 3

3.1 Phase 1: Recruitment Post Analysis

  • Phase 1 collected 1,268 posts from two public WeChat recruitment groups and screened them to form a final corpus of 627 posts targeting women streamers.GPT-5 assisted only with initial screening; substantive coding was manual, and a stratified manual audit found 100% precision and 96.3% recall before disagreements were reviewed and corrected.Source: §3.1.1 · p. 3
  • The recruitment-post analysis manually coded requirement concepts in Chinese, using iterative conceptual content analysis to develop eight broader, non-mutually-exclusive themes.Each post could contribute to multiple categories, but a given requirement was counted at most once within a post.Source: §3.1.2 · pp. 3–4

3.2 Phase 2: Interviews and Observation

  • Phase 2 combined 26 semi-structured interviews with 23 livestream observations totaling 17 hours to examine personas, masking practices, motivations, and technology perspectives.Interviews covered preparation, presentation, conditions shaping masking, and views of current and desired technologies; observations tracked settings, filters, styling, content, and interactions.Source: §3.2 · p. 4, §3.2.2 · p. 4, §3.2.3 · p. 4
  • The team analyzed interviews, observation notes, screenshots, and participant-shared photographs thematically, developing 48 initial codes and clustering them into higher-level themes.The first author translated Mandarin transcripts into English and used back translation to check consistency and accuracy.Source: §3.2.4 · p. 4

3.3 Positionality

  • The authors describe the first author’s Chinese linguistic and platform familiarity alongside her HCI-researcher distance from participants, and note how team backgrounds and translation shaped interpretation.They retained Chinese terms when English could not fully convey their meanings and acknowledge their attention to agency, gender, and structural power.Source: §3.3 · p. 4

4 Chinese Women Livestreamers and Livestreaming Practices

The section situates participants in a varied, often intermediary-supported livestreaming industry and describes the work behind their on-camera personas. Preparation, backstage coordination, and post-broadcast tasks extend beyond what viewers see.

4 Chinese Women Livestreamers and Livestreaming Practices

  • Chinese livestreaming spans gift-driven entertainment and e-commerce, with increasingly blurred boundaries between showroom and sales formats.Platforms have integrated shopping and storefront functions into livestreaming.Source: §4 · pp. 4–5

4.1 Who Are “nüzhubo”?

  • Participants described stigma around the identity nüzhubo and the difficulty of moving beyond the lower dibu tier in China’s crowded market.Most participants were lower-tier streamers; P26 was an exception with about 260,000 followers. Participants also described low salaries, commission-based income, contractual restrictions, and demanding schedules.Source: §4.1 · p. 5, §4.1.4 · p. 5, §4.1.5 · p. 5
  • Participants entered livestreaming from varied occupations and circumstances, including students, office workers, actors, shop owners, entrepreneurs, international students, and service workers.Some used livestreaming alongside existing jobs or businesses, while others began through recruitment, audience-building, or income needs.Source: §4.1.2 · p. 5
  • Participants worked in overlapping sales, consultation, and entertainment formats, and 24 had worked with companies, brands, factories, agencies, or other intermediaries.Some moved between formats, showing that livestreaming personas were rarely produced by streamers alone.Source: §4.1.3 · p. 5

4.2 Preparing and Performing a Livestream

  • Preparation began hours before broadcasting and included personal styling, content or sales preparation, and arranging backgrounds, lighting, camera angles, products, and props.P20 began two hours before going live to arrange products and verify links, colors, and sizes.Source: §4.2.1 · p. 5
  • Commercial teams coordinated live work across streamer, assistant host, central control, and operations roles, while Figure 1 contrasts the audience-facing stream with its backstage setup.Operations staff monitored performance and sales data and could shape pacing, product choices, account direction, and persona development; three-person teams combining roles were more typical than four-person teams.Source: §4.2.2 · p. 6, Figure 1 · p. 6
  • Independent streamers often performed while monitoring comments, managing the interface, and handling disruptions, sometimes with help from moderators or managers.Managers could provide emotional support or backend monitoring and real-time suggestions.Source: §4.2.3 · p. 6
  • Work continued after broadcasts as streamers reviewed performance metrics, maintained relationships with high-spending viewers, and handled sales logistics.In P5’s fabric business, post-stream packing and shipment preparation followed broadcasts that sometimes ended at midnight or 2:00 AM.Source: §4.2.4 · pp. 6–7

5 Industry-imposed Personas on Women Livestreamers

Recruitment posts mapped a desired woman streamer persona across appearance, aesthetic and personality, and attention and sales skills. They framed bodily presentation and on-camera capability as occupational requirements.

5 Industry-imposed Personas on Women Livestreamers

  • Across 627 recruitment posts, appearance (N=641) led specified expectations, followed by on-camera capabilities (N=561), sales and conversion (N=494), aesthetics (N=451), body size (N=448), personality (N=250), and age (N=205).Source: §5 · p. 7

5.1 Appearance and Bodies

  • Recruiters treated the body as an occupational qualification, requesting attractive looks (194), gentle temperament (142), height at least 160 cm (159), XS/S size (83), or weight no more than 50 kg (57).Youthful presentation was also favored, with posts specifying young feminine (75), lightly mature (35), or girly (24) labels.Source: §5.1 · p. 7

5.2 Aesthetics and Personalities

  • Posts linked recognizable styles to body and styling—Korean style (104), casual chic (88), hot-girl (72), minimalist (56)—and favored approachable personality (90).Other listed styles included heiress (49) and old-money (25); personality traits included sweet and cute (59), gentle or tender (41), and passionate or infectious (34).Source: §5.2 · p. 7

5.3 Skills for Attention and Monetization

  • Posts sought on-camera presence (121), verbal expression (101), clothing modeling (89), aesthetic judgment (88), pacing (47), room control (36), account-launch experience (80), organic-traffic conversion (66), GMV or sales performance (59), and high-ticket sales (59).Streamers were expected to sustain talk and respond quickly, while attention was expected to convert into gifts or e-commerce sales.Source: §5.3 · pp. 7–8

6 Mask, Persona, and Livestreamers’ Negotiation

This section examines how streamers assemble and negotiate recognizable masks across personas, bodies, technologies, spaces, performances, and social relations, including maintaining, breaking, or reducing them to navigate commercial, audience, and platform demands while protecting personal boundaries and managing performance costs.

6.1 Assembling the Mask: Constructing an On-Stream Persona

  • Companies and brands shaped streamers’ expected personas, matching appearance, voice, and personality to commercial roles that participants could plausibly sustain.Some participants chose staff whose aesthetic preferences resembled their own, while others found close alignment with their offline selves easier to maintain.Source: §6.1.1 · p. 8
  • Masks extended beyond the face: streamers calibrated makeup, lighting, cameras, and digital effects, and used bodies, objects, settings, and rehearsed performance to make personas recognizable.A polished face aimed for enhanced but plausibly natural appearance; product-selling and entertainment performances also relied on prepared expertise, rapid presentation, and audience interaction.Source: §6.1.2 · p. 8, §6.1.3 · p. 8, §6.1.4 · p. 9

6.2 Fitting, Maintaining, or Breaking the Mask

  • Streamers fitted masks to their own traits and commercial needs, balancing imposed personas and aesthetic ideals against sustainability, self-recognition, and product credibility.Some resisted excessive retouching or appearance standards, while product sellers adjusted effects to support product claims or preserve accurate color and texture.Source: §6.2.1 · pp. 10–11
  • Platform moderation made fitting masks uncertain: all 26 participants described traffic reductions, suspensions, mistaken moderation, or bans, and some used coded speech or workarounds.P2 eventually stopped livestreaming after two accounts were banned, despite attempts to avoid moderation.Source: §6.2.1 · p. 11
  • Audiences and moderators helped maintain personas through interaction, hidden disciplinary work, shared conventions around AI-assisted performances, and contact that sometimes continued off-stream.Three participants described reassuring jealous moderators, and five added high-spending viewers on personal WeChat accounts.Source: §6.2.2 · p. 11
  • Streamers strategically broke or replaced masks to create contrast and novelty, making unexpected shifts another recognizable layer of their persona.During PK sessions, streamers and audiences also negotiated how much mask-breaking would be exposed, including by avoiding makeup-removal punishments.Source: §6.2.3 · pp. 11–12
  • Technical failures could expose beauty effects and threaten authenticity, prompting some streamers to stop using modifications that became visibly unstable.Twenty-four participants described effects losing track of faces or bodies; P25 stopped using leg-lengthening after glitches made the manipulation obvious.Source: §6.2.3 · p. 12
  • Participants protected boundaries around identity and everyday life through stage names, selective framing, separate accounts, and limits on viewers’ access.P12 blocked the camera when her roommate returned because she did not want her roommate to appear or know she livestreamed.Source: §6.2.4 · p. 12
  • Streamers protected privacy and offline relationships by stopping streams, keeping home locations and family out of view, and balancing audience-facing personas with how offline contacts knew them.P15’s dorm livestream let a viewer infer her university and wait at the campus gate; P13 kept her husband off-camera to shield him and her family from scrutiny, while P15 used her own photos to avoid confusing offline friends.Source: §6.2.4 · pp. 12–13
  • To limit romantic or sexual access, streamers used transactional humor or weakened attractive personas, while company rules against flirtation offered limited on-the-spot support.P15 pretended to call her mother and demanded a car as bride price; others disclosed filter use or claimed to be lesbian to discourage viewers seeking offline intimacy.Source: §6.2.4 · p. 13
  • Long hours, overnight work, and accumulated exhaustion could make maintaining a persona unsustainable, with P26’s confrontation exposing labor and strain her controlled mask had concealed.P11 left after severe physical stress and safety concerns; P26 sometimes streamed 15–16 hours while sleeping about four hours a day before a degrading comment became the breaking point.Source: §6.2.4 · pp. 13–14
  • Eight participants reduced masking, and P23 rejected facial retouching and polished beauty expectations, preferring a natural appearance despite modest traffic.She used only occasional playful effects that did not change her facial features, did not join a guild, and gained a stable group of women viewers who valued her presentation.Source: §6.2.4 · p. 14

7 Discussion

The discussion frames masking as a relational negotiation over visibility, recognition, and boundaries, not merely a cosmetic platform function. It connects agency to reconfiguring mask assemblages and treats breaking, resistance, and refusal as consequential but constrained practices.

7 Discussion

  • Masking is a political and sociotechnical negotiation over who shapes streamers’ visible selves, which femininities become recognizable and valuable, and when demands can be modified or refused.The discussion opens theoretical and design directions through assemblage, visibility, breakdown, resistance, and refusal.Source: §7 · p. 14

7.1 Implications to Design

  • Design should preserve streamer-defined appearance, make organizational mask edits contestable, and make moderation rules legible and appealable.Proposals include protected features and modification ranges, records and restoration of edits, and explanations with human review.Source: §7.1 · p. 14, §7.1.1 · p. 14, §7.1.2 · p. 14

7.4 Breaking, Resistance, and Refusal

  • Masking agency lies in destabilizing, preserving, and recomposing relations within commercial sociotechnical assemblages, not simply complying with or resisting them.Breaking may expose the infrastructures and dependencies behind seamless presentation; refusal can carry costs to visibility, income, audience response, or workplace evaluation.Source: §7.2.2 · p. 15, §7.4.1 · p. 16, §7.4.2 · p. 16

7.2 Masking Through an Assemblage Lens

  • A mask assemblage joins bodily, computational, spatial, organizational, and performative elements; agency involves reconfiguring their relations rather than adding interface controls alone.Assigned personas, beauty presets, and metrics can stabilize masks, while disruptions may become another recognizable performance.Source: §7.2.1 · pp. 14–15, §7.2 · p. 14, §7.2.3 · p. 15

7.3 The Politics of Visibility and Boundary Work

  • Masking organizes visibility across domestic, public, and commercial spaces, while recognition remains shaped by unequal gendered standards and evaluation.The camera frame becomes part of the mask, and privacy involves protecting roommates, partners, and family as well as the streamer.Source: §7.3 · p. 15, §7.3.1 · p. 15, §7.3.2 · p. 15

8 Limitations and Future Work

The study’s findings are bounded by its participant profile, recruitment source, and interview-and-observation approach. The authors refrain from generalizing beyond the studied setting and identify more diverse samples, data sources, and extended engagement as future directions.

8 Limitations and Future Work

  • The study focused primarily on lower-tier Chinese women livestreamers, whose experiences may differ from those of higher-tier streamers with larger audiences, stronger teams, and greater bargaining power.Participants were primarily heterosexual women, leaving other sexuality, gender identity, class, age, and regional positions for future study.Source: §8 · p. 16
  • Recruitment-post analysis relied on WeChat groups, so future work could use multiple recruitment platforms and industry data sources to capture a broader labor ecology.Source: §8 · p. 16
  • Observation and interviews may miss practices difficult to articulate, and the authors refrain from generalizing beyond the studied setting.They suggest extended self-ethnographic or longer-term engagement as one future approach.Source: §8 · p. 16

9 Conclusion

The study examines how Chinese women livestreamers use masking to negotiate industry requirements and personal agendas for visibility, identity, privacy, and safety. Its findings describe how recognizable masks are assembled and negotiated, and the conclusion discusses implications for theory and platform design.

9 Conclusion

  • Industry-defined ideals of appearance, femininity, and performance shape recognizable masks before women go live.Source: §9 · p. 16
  • Streamers assemble masks through technologies, clothing, environments, voice, and performance, then negotiate, break, resist, or refuse them in relation to companies, audiences, platforms, and personal boundaries.Source: §9 · p. 16
  • The study discusses implications of these findings for theory and platform design.Source: §9 · p. 16
Loading 2609.28721v1…