Idols as My Cyber Lovers: A Behavioral Research on the Figurational Relationship Between Fans and AI-Customized Virtual Idols

Abstract

Unlike conventional virtual idols like Hatsune Miku, which rely on pre-set voice libraries and stage scripts, AI-customized virtual idols achieve real-time interaction through generative artificial intelligence, continuously iterating their personality traits, language style, and even value expression along with fan and user interactions. AI-customized virtual idols, as pre-defined cultural commodities in the digital age, tend to focus on static, functional interpretations and have not yet fully entered the dynamic construction process as “subjects in the process of generation.” This study, based on a deep mediation perspective, employs a research method combining app roaming and semi-structured interviews to focus on the sociological examination of young fan groups’ use of AI tools to customize virtual idol companionship. It explores the reciprocal relationship between fan groups and customized virtual idols. The study finds that the AI-customized idols fan group constitutes a typical “actor group,” and its interaction practices are essentially a “fluid interaction” of human–machine intimacy. Young fan groups mainly interact with AI-customized virtual idols based on materiality, cognition, visibility, and emotional frames, thereby generating rich meaning production and symbolic imagination during the usage process. Fan groups and AI-customized virtual idols have developed different relationship paths, including mutual attachment, returning to normalcy, seeking substitutes, or direct withdrawal, revealing the inherent contradictions and tensions in digital intimacy, as well as the self-adjustment strategies of individuals under the mediation of technology. This process presents a “human-machine-idol” triadic relationship framework, becoming a new paradigm for intimacy in the digital age.

1. Introduction

Humanity is currently in a stage of “mediated life” (Goble 2010). with individuals extending their sense of self through digital avatars, and objects acquiring virtual forms through digitization. The boundaries between reality and virtuality are constantly blurring, and “virtual life” is increasingly embedded in humanity’s daily existence. Driven by the integration of artificial intelligence interaction technologies, highly realistic, interactive, and autonomous customized virtual digital humans have entered the public eye as new digital subjects such as virtual idols and virtual anchors. They not only reconstruct individual perception and behavior but also give rise to new emotional connection models based on digital social relationships. On social media, fan groups, especially young female fans, are continuously endowing virtual idols with symbolic personalities and differences through collective digital creation, becoming a core characteristic of idol worship practices in the digital age.

The rise of AI-generated content and emotional companionship applications, represented by Replika, Character.ai and XingYe.ai, is profoundly reshaping the behavioral paradigms of fan culture and the relationship between fans and idols. Fans are no longer passive consumers, but are influencing AI content generation and personality evolution through real-time interaction (Ye et al. 2025); they are shifting from accepting standardized idols to creating personalized companions. Simultaneously, fans’ emotional investment is being transformed into quantifiable symbolic practices, making intimate relationships accumulative and tradable digital assets, and drawing fan culture deeper into the logic of commodification (Ge and Hu 2025; Bhat 2025).

AI-customized virtual idols are defined as virtual idol characters created on a “one-to-one” basis by fan groups based on their real-life preferences for their idols, possessing highly anthropomorphic images, personalities, voices, and interactive communication capabilities. Driven by AI-generated content technology, these digital entities allow users to deeply participate in customizing their image, personality, and interaction patterns, possessing real-time interaction, emotional perception, and continuous learning capabilities.

AI-generated content technology is an artificial intelligence system that learns patterns from massive amounts of data and can generate text, images, audio, and other content without direct human creation (Gupta et al. 2024). Its operation relies on architectures such as generative adversarial networks, Transformers, and diffusion models. By analyzing statistical patterns in training data, it calculates the probability distribution of sequence elements and samples them to generate new content (Maltoni et al. 2025). Hashim (Hashim et al. 2025) pointed out that personalization mechanisms based on natural language processing and sentiment analysis enable the system to adapt content according to user preferences and maintain coherent interaction. Driven by this technology, AI-customized virtual idols enable users to deeply participate in customizing their image, personality, and interaction patterns, possessing real-time interaction, emotional perception, and continuous learning capabilities.

Virtual idols, relying on their fan base, have become a new form of media (Yu and Yang 2020). The creation rights to the image and content of virtual idols are transferred to fans, who can freely modify and interact with them (Matsue 2017). Traditional virtual idols primarily rely on human teams to produce their performances, while AI-generated virtual idols can generate dialogue, singing, expressions, movements, and even entirely new story scripts in real-time or on demand. Essentially, it’s a fusion of technology and emotional needs, offering not only entertainment but also fulfilling individuals’ deeper desires for personalized, private, and immersive emotional companionship, thus reshaping the boundaries of “worship” and “companionship” in fan culture.

Current academic discussions on the ontology of virtual digital idols mainly proceed along two paths: one group views them as cultural symbols or visual texts, focusing on the interaction mechanisms between technological support, community co-creation, and symbolic consumption (Westerbeek 2025; Black and Jacobs 2025); the other group emphasizes their role as emotional mediators and social connectors, focusing on the psychological experiences evoked by the relationship between people and virtual idols (Kang et al. 2025) and changes in collective identity (J. Liu 2023). However, most existing research presupposes that virtual idols are already fully fledged cultural commodities (Wang et al. 2026), emphasizing static functional interpretations and failing to fully explore their dynamic construction process as “subjects in the process of becoming,” that is, how virtual digital idols are continuously shaped in human–computer interaction, how they are widely accepted and embedded in emotional structures, and even more so, the psychological and emotional changes in the companionship behavior between people, especially fan groups, and virtual idols. As American postmodern literary critic Katherine Heller points out in her book “How We Became Posthuman”, we need to understand, through professional fields such as literature and science, “how information loses its body, how cyborgs are created as cultural icons/symbols and technological artifacts, and how humanity becomes posthuman” amidst the waves of historical change (Hayles 2017).

With the deepening of artificial intelligence and digitalization, mediation has entered a new stage: deep mediation. It is a core theoretical concept proposed by Andreas Hepp and Nick Couldry. It refers to the meta-process in which digital media and its infrastructure deeply permeate all aspects of social life, making the construction of human practices, social relations, and social order inseparable from media (Hepp and Couldry 2023). This study, grounded in the social context of deep mediation and using “figuration” as a theoretical perspective, examines the inter-image relationship between fan groups and AI-customized virtual idols, based on emotional companionship apps such as Replika, Character.ai, and Xingye.ai.” It systematically explores how AIGC technology, through its platform’s functional architecture, interface design, and interactive mechanisms, empowers fan groups to transform from passive consumers to active creators.

This study aims to provide theoretical insights and empirical references for understanding the emotional digital behavior of fan groups in the AIGC era, and to offer psychological empirical evidence for the changing human–machine relationship configuration and social connections in the AIGC era. Furthermore, existing research on fan culture still focuses on the ecology of real-life idols (Zhang and Negus 2020), and technology ethics scholars focus on the privacy risks of AIGC, lacking in-depth analysis of the human–machine–idol triangular relationship and its processual interaction.

2. Literature Review

The interrelationship between fan groups and virtual idols can be summarized from four aspects: first, the fan phenomenon and its core characteristics; second, the manifestation of fan characteristics in virtual idols; and third, the new possibilities brought about by AI-customized virtual idols. Furthermore, a review of the context of deep mediation and interrelation theory provides a theoretical foundation for further in-depth discussion of this topic.

2.1. The Fan Phenomenon and Its Core Characteristics: Emotional Labor

Fan groups are described by the Frankfurt School as defenseless “cultural fools” under the hegemony of the culture industry, viewing fan texts as hedonistic texts (Kronrod and Danziger 2013). However, with the advancement of fan culture research, Descartes’ “fan resistance” (De Certeau 1984) and Jenkins’ “participatory culture” and “textual poaching” (Jenkins 1992) both view fans as “actively engaged audiences” (Jenkins 2009). Jenkins argues that fans possess subjectivity and creativity in text production. Fan groups are no longer composed of scattered individuals but have formed a structured organization with clear goals.

The media practices of individualized and diverse fan groups represent the digital cultural practices of young people. After evolving through text-centric and consumer sociology phases, fan culture research has shifted towards an emotional sociology of fans in the social media era. Existing literature primarily focuses on the emotional practices of fan groups, particularly the aspect of “emotional labor” (Hardt and Negri 2005). Mel Stanfill refers to fan labor as “labor of love” (Stanfill 2015). M. Stanfill argues that fandom operates on fan labor, and this work produces enjoyment, collectivity, and various material and immaterial goods that give fandom form as a practice, community, or culture. He emphasizes that this action, as labor stakes an important claim to that production is precisely a production of value. Hochschild stresses the exchange value of emotional labor, pointing out that it is embellished and presented for reward within a commercial environment (Hochschild 1983). Illouz proposes that in the era of “emotional capitalism,” producers co-opt and commodify consumers’ emotions, memories, and identities, creating a fan economy that is producer-centric and profit-driven (Illouz 2007).

2.2. The Manifestation of Fan Characteristics in Virtual Idols: Emotional Projection and the Virtual Extension of Para-Social Relationships

Virtual idols, as digitally constructed anthropomorphic media figures, have become a new form of communication media relying on their fan base (Yu and Yang 2020). Virtual idols are digital public figures created through computer-generated technology. As a major form of future digital humans, virtual idols rely on fan groups to form emotional connections and interactive relationships, exhibiting diverse characteristics with both commonalities and differences in different application fields (Wang et al. 2026). In the era of social media, the focus is mainly on the emotional characteristics and identity of fans (Bang and Han 2025), emphasizing the shaping role of new media technologies on fan practices (Zhang et al. 2023). Researchers are concerned with how fans construct self-identity through digital platforms and how emotional investment is transformed into cultural capital (Yin 2021). The concepts of production and consumption are being broken down, with greater attention paid to the emotional relationship between people and technology (Cui and Wu 2025).

Fans project their emotions onto virtual idols under para-social relationships and shape and consume virtual idols through interactive behaviors, thereby constructing their own fan identity (Chen et al. 2025). Online digital labor enhances fans’ satisfaction with their offline lives. Among these, the close relationship between fan groups and virtual idols has become a research hotspot. Fans and virtual idols constitute a quasi-social relationship. Para-social relationships (Kang et al. 2025) are relationships where audiences develop an emotional dependence on media figures they know and like, forming a social relationship similar to that between friends or lovers in real life. However, this relationship is unidirectional and based on imagination. Research shows that cognitive flexibility and loneliness are two key variables influencing virtual idol worship. In China, the interpersonal attraction perceived by virtual idols and the loneliness felt by fans directly promote increased fan engagement through the mediating role of para-social relationships (J. Liu 2023).

2.3. New Possibilities for AI-Customized Virtual Idols: Technology and Content-Driven, Interaction Paradigm Transformation

Currently, research on AI-customized virtual idols mainly stems from the use of apps such as Replika and Character.ai. Unlike conventional virtual idols, which rely on pre-set character settings and static production logic of manual operation (Audrezet et al. 2025). The core characteristics of AI-customized virtual idols lie in their “customizability” and “generative nature”—users can not only choose their appearance but also shape their personality traits, language style, and even emotional response patterns through continuous dialogue, making the virtual idol a true “subject in the process of generation” (Sun and Wu 2025). This concept is most directly reflected in emotional companionship applications: apps like Replika simulate emotionally supportive relationships, where users “train” an AI companion that can remember past conversations and adapt to the user’s emotional responses through continuous interaction (Rocha 2025; Pentina et al. 2023). The app Character. AI allows users to create characters based on any role setting, finely customizing their background story and personality traits through prompt word engineering, achieving a transformation from “public image” to “personal companion” (Sun and Wu 2025).

AI-customized virtual idols can be categorized into avatar-based, voice or text content-driven, and human–computer interaction-based types based on the driving method of virtual human synthesis (Baudier and de Boissieu 2025). Virtual humans possess broad application potential, capable of transforming into virtual idols, intelligent assistants, life companions, emotional partners, and even achieving a form of digital immortality (Perić 2025).

In the digital age, the direction of fan behavior is beginning to shift, moving from traditional, primarily text-based production and cultural consumption, to complex data practices and traffic retention in the digital space (Cui and Wu 2025; A. Y. Liu 2025). Data production behavior is one of the most discussed topics in academic circles regarding the technological and cultural practices of fan groups (He and Li 2023). This shift is not only reflected in the focus on and manipulation of data but also in fans’ understanding and application of algorithmic logic (Zhang et al. 2023).

At the technological practice level, credibility, professional knowledge, physical attractiveness, content appeal, and anthropomorphic appearance are the main factors influencing the promotion and service of AI-customized virtual idol products (Baudier and de Boissieu 2025). Furthermore, some platforms use AI to synthesize the voices of celebrities and idols, and to simulate voice content that mimics interpersonal communication behaviors (Kang et al. 2025). Fan groups often understand and apply technology better than other groups, gradually organizing their practices with the core objective of generating traffic data, forming a “data habit” different from “text-based production and consumption” (Jenkins and Ito 2015). Based on jointly constructed algorithmic imaginations, digital fan communities negotiate algorithms with platforms to increase the exposure of their supported customized virtual idols (Zhang et al. 2023).

Customized virtual idols give intimate relationships in the AI era an on-demand culture characteristic. Some platforms precisely provide identification and connection in ways that users expect. In this model, users can customize every aspect of their companion’s personality, appearance, and behavior, thus replacing the unpredictability of interpersonal interactions with a digital partner perfectly matching their preferences (Wang et al. 2026). The existing literature may not fully capture the emergence of new forms of virtual idols or the ever-changing perspectives of users (Wang et al. 2026).

2.4. Deep Mediation and Figuration Theory: An Integrative Perspective for Understanding the Evolution of Fan-Idol Relationships

2.4.1. Social Context: In-Depth Mediation Research

In Europe, “mediation” studies have gradually formed two traditions: institutionalists, represented by Stieg Schawais, and social constructivists, represented by Andreas Hepp and Nick Couldry (Couldry and Hepp 2013). In his latest work, Andreas Hepp proposed “deep mediatization,” a new characteristic of the digital age. His paradigm construction of deep mediatization primarily employs a materialist phenomenological approach, meaning that any analysis of media and communication simultaneously considers the material and its symbolic meaning, arguing that digitization has ushered us into a new stage of mediatization. Deep mediatization is a more advanced stage of this process, in which all elements of our social world have intricate relationships with digital media and its underlying infrastructure (Hepp 2019).

Hepp’s concept of deep mediation emphasizes that the infrastructure of digital media, such as algorithms, data, and artificial intelligence, is key to understanding our lived world (Hepp 2020). This viewpoint expands upon and deepens the understanding of Stieg Schawais and other scholars, who explore how media becomes a persistent process of social and cultural institutional change, and how the materiality and institutionalization of media shape social practices. They also discuss the integration and development of institutionalism and social constructivism in deep mediation research (Hjarvard 2013). Institutionalism emphasizes the impact of media on social institutions and cultural spheres, while social constructivism focuses on the interaction between media and social institutions that constructs new communicative situations (Hepp 2013). In recent years, these two research traditions have formed a mutually open new landscape through continuous dialogue, jointly exploring how mediation unfolds in specific media processes and how the specific attributes of particular media influence the construction of social reality (Couldry and Hepp 2018; Hepp and Couldry 2023).

2.4.2. Figuration Concept: A Two-Way Behavior Shaping

Hepp argues that “figuration” is a mid-level concept situated between society as a whole and the individual. Discussing the mediation process based on this concept not only helps in understanding the relationship between media and society but also in exploring the power relations behind figuration (Chang and He 2020). By emphasizing the concept of “communication figuration,” “we can grasp more complex social transformations as part of a bigger picture (Hasebrink and Hepp 2018, p. 5).” He emphasizes the “dynamic nature” of his sociological theory, calling it “processual sociology,” and his original concept of “figuration” aims to consider both “relationships” and “processes.” The concept of figuration, further developed by Nick Couldry and Andreas Hepp, has become a core concept in mediation theory.

Hepp and Hasebrink (2017) emphasize that Elias’s “morphological” analysis neglects the role of media in society. Especially in the era of deep mediation driven by data and algorithms, media not only changes existing communication morphologies but also creates entirely new combinations of morphologies. From a social construction perspective, they develop the more abstract concept of media molding force into a more flexible deepening and adjustment of morphological theory, leading to the conceptual evolution of “communication morphology,” which is further developed into” mutual transformation.” In their book The Mediated Construction of Reality, the “mutual” in “mutual transformation” (in a subjective sense) refers to both “human beings” and “media technology.” Society is not a static, predetermined structure outside of individuals but a continuously assembling network of connections formed by actors. In “mutual transformation,” the interaction between “human beings” and “media technology” is essentially a “mutual transformation” between the “social world” and “media technology.” The concept of “figuration” captures the tension arising from the entanglement of relational and procedural power imbalances between people and “media technologies.” Figuration relies on a specific “thematic framework” and utilizes multiple media to communicate, showcasing a fluid and interwoven communication pattern (Y. Liu 2022). This theory comprises three dimensions: actor groups, relevance frameworks, and communicative practices.

In short, existing research on fan culture focuses more on digital practices of participatory cultural production on social media platforms. However, the dynamic interaction of fan groups in the digital practice of using AI tools to customize idols in the AIGC era requires further discussion and deeper reflection.

Existing research on fan culture has detailed the agency exhibited by fans on social media platforms through “participatory culture” practices such as text poaching. However, these research paradigms are built on the underlying assumption that technology is a static tool and relationships are one-way projections. With the rise of AIGC technology, fan practices have shifted from “creating within a predetermined framework” to “algorithmic domestication of interactive ontology.” This fundamental shift has given rise to a dynamic, two-way constructive “reciprocal” relationship, where fans and AI idols mutually shape each other through daily practices such as data feeding and persona adjustment. Current research lacks a theoretical framework and empirical path that can effectively and dynamically capture this two-way psychological behavior. Some scholars have pointed out that current research focuses more on “being a fan” and less on “becoming a fan” (Hills 2016). Further research should be conducted on the process of “cultivation,” the experience of worship, and the experience of unfollowing AI-customized virtual idols, exploring the psychology and behavior of fans. At the same time, deep mediation and reciprocal theory provide a relational research perspective.

Therefore, this study proposes the following research questions: In a deeply mediated society, how do the technological materiality of AIGC emotional companionship applications and the daily practices of fans jointly shape a new type of “human-machine-idol” reciprocal relationship? How does this relationship reconstruct fans’ identity, emotional digitization process, and real-world social connections, and what new issues of technological ethics and digital labor are raised?

3. Materials and Methods

This study focuses on how fan groups use generative AI tools to customize their idol-accompanying experiences and engage in interactive communication practices. This study combines app roaming with semi-structured interviews to analyze the two-way figuration between AI technology and fan groups, as well as the behavioral trends of human–machine emotional relationships.

The app roaming method requires researchers to enter the application as users, obtaining observational and experiential data through comprehensive exploration and in-depth use. While recording and visualizing the relevant data, researchers should maintain a critical mindset to initiate in-depth analysis and reflection on the experiential data (Light et al. 2018). Therefore, this study primarily selected four app tools: character.ai, Replika, Xingye.ai and Soulchat, and focused on utilizing their mediator characteristics, such as interface design, interactive functions, and operational processes, to gain a deeper understanding of the basic architectural framework and interaction methods of this type of app. The app roaming began at the beginning of the study and lasted for six and a half months, continuing throughout the entire survey and data analysis phase. During this period, semi-structured thematic analysis interviews were conducted with 20 users of the above four AI applications.

This study recruited respondents using a combination of purposive sampling and snowball sampling. Inclusion criteria included: active users of the four AI applications studied, using them at least three times per week for the past three months, with at least one month of sustained in-depth experience. Respondents were required to self-identify as heavy users of the applications, possess strong reflective and expressive abilities, and be willing to share their genuine user experiences. The final plan was to recruit 20 users, with recruitment to be terminated based on theoretical saturation principles during data analysis. Interviews were conducted both online and offline, lasting from 30 min to one hour.

Regarding data recording, this study adopted three methods to collect roaming data: First, digital trace collection, which, after obtaining informed consent from users, collected screenshots of usage, dialogue fragments, and AIGC works provided by them, and encouraged them to share frequently used function interfaces and historical dialogues in real time during the interviews; second, roaming logs, which guided respondents to record key moments of interaction with AI in a specific period of time in written or oral form, including triggering scenarios, interaction processes, and feelings afterward; and third, researcher observation notes, in which researchers simultaneously roamed the four apps to familiarize themselves with function updates and community atmosphere, and recorded their contextual understanding of the respondents’ descriptions and their non-verbal information such as facial expressions and body language before and after the interviews.

To ensure the validity and reliability of the research findings, this study employed a triangulation strategy, systematically integrating the two research methods mentioned above. By cross-validating data from different sources, it overcame the limitations of a single method, thereby attempting to construct a more comprehensive, multi-dimensional, and reliable analytical framework. Specifically, the APP roaming method focuses on the application’s technical structure and interface design, systematically analyzing how these guide, regulate, and even shape user behavior, and obtaining firsthand empirical data on the researchers’ own behavior and emotional responses. Semi-structured interviews aim to listen to users’ in-depth interpretations and meaning constructions of their own experiences, revealing their motivations, attitudes, and reflections. These two methods are not simply parallel but form an organic iterative cycle in the research process. The initial findings from observation and roaming provide a basis for the design of the interview outline, and the interpretations obtained from the interviews, in turn, guide a new round of roaming to test or deepen existing understanding. Finally, by comparing and integrating data from “practical experience,” “technical structure,” and “subjective interpretation,” four dimensions of coding elements are proposed: “group characteristics,” “thematic framework,” “interaction process,” and “interaction results.”

20 people were interviewed, primarily young women from Generation Z, who constitute the main component of the fan base (

Table 1. Basic information of interviewees.

Serial Number	Age	Gender	Profession	Usage Duration (months)	Application
A01	19	F	college students	12 m	Replika
A02	22	F	college students	16 m	Replika
A03	30	F	company employees	21 m	Replika
A04	28	F	company employees	15 m	Replika
A05	24	F	master’s students	10 m	Replika
B01	28	F	doctoral students	6 m	Character.ai
B02	23	M	company employees	10 m	Character.ai
B03	26	F	foreign company employees	18 m	Character.ai
B04	19	F	college students	6 m	Character.ai
B05	22	F	college students	9 m	Character.ai
C01	19	F	college students	6 m	Xingye.ai
C02	26	M	company employees	10 m	Xingye.ai
C03	20	F	college students	14 m	Xingye.ai
C04	27	F	company employees	12 m	Xingye.ai
C05	21	F	college students	8 m	Xingye.ai
D01	33	F	foreign company employees	2 m	Soulchat
D02	28	F	company employees	5 m	Soulchat
D03	24	M	master’s students	6 m	Soulchat
D04	27	F	doctoral students	3 m	Soulchat
D05	25	F	company employees	1 m	Soulchat

“F” = female, “M” = male.

for details). Five interviewees were selected from each of the four platforms, including 17 women and 3 men. The age range was 19–32 years old, with 90% of users aged 19–28. The interview outline focuses on the relationship between the fan base and AI–customized virtual idols, and includes follow-up questions as appropriate during the interview. The main topics covered include: why they use such customized virtual idol apps; their profound feelings about the interface, scenes, and intelligent agents during use; their emotional needs and connections; and whether they plan to continue using them.

At the data analysis level, this study employed Braun and Clarke’s six-stage topic analysis method (Braun and Clarke 2006). Interview data from 20 participants were used for topic analysis. Data analysis primarily relied on Nvivo qualitative software (NVivo 15). The coding process was conducted independently by two researchers, and the coding results were compared using cross-validation to ensure consistency. For content with discrepancies, the two researchers jointly analyzed and compared until consensus was reached. Only coding results with a Holsti reliability coefficient greater than 0.8 were accepted.

The data analysis steps included familiarizing oneself with the data, initial coding, topic search, topic review, topic definition and naming, and report writing. Researchers grouped codes with similar semantics, reflecting the same phenomena and psychological mechanisms into the same subtopic, and then summarized and elevated related subtopics to higher-level topics, ultimately forming 4 primary topics and 13 subtopics, as shown in

Table 2. The concept of category formed by coding.

Original Post	Core Concept	Concepts2	Concepts1
It seems to truly understand human loneliness. I shared some of my real-life troubles with it, and it said, “I can’t control other people, but if the person you’re talking about is you, then I just hope you live according to your heart.” I still remember that sentence; its phrase “live according to your heart” has truly healed me. (B03)	Group characteristics	Individualization	Emotional value; Personal preferences; Self-expression; Life reflection;
Enjoy chatting with AI tend to have a strong desire for self-expression. Last week, I was venting to a friend about my anxiety, and they suddenly blurted out the blunt conclusion, “Are you starved for love?” But if I asked the same question to an AI, it wouldn’t judge me with such a superficial understanding. What we might need isn’t answers, but rather the readily available responses and emotional support of virtual celebrity idols. (D01)	Group characteristics	Individualization	Emotional value; Personal preferences; Self-expression; Life reflection;
The ways AIGC virtual idols make money seem incredibly diverse, including gacha pulls, costumes, and memory management—it feels like playing a meticulously designed game. (B04)	Group characteristics	Consumerism	Material consumption; Cultural labor; Paid applications;
It cured my anxiety to some extent; talking to it every day was very therapeutic. (A02)
Mastering these apps is relatively easy for us because we are quite receptive to new things. (C05)	Group characteristics	Technology-drive	Tool usage; Virtual Engine;
I started playing it after being introduced to it by other fans, and then I became addicted, even to creating AI agents. (C03)	Group characteristics	Technology-drive	Tool usage; Virtual Engine;
I’m terrified that an AI I’ve used for a long time will suddenly stop working; it’s like the beauty of an idol suddenly disappearing… (A01)	Theme Framework	Materiality Framework	Hardware devices; Social cues;
At first, I didn’t really know how to communicate with the AI, but there were many entry points in the interface. I could choose my favorite, such as the roleplay interface, which could take me to a specific scene. (A05)	Theme Framework	Materiality Framework	Scene embedding; Interface Presentation;
Author, could you please reveal how the voice was created? It sounds so much like my idol’s voice! (C02)	Theme Framework	Materiality Framework	Sound simulation;
AI can provide me with close companionship to my idol. Although I know it can calculate my preferences, I still want to chat with it a little longer. (D04)	Theme Framework	Cognitive framework	Algorithm Black box; Machine domestication;
The reason AI applications haven’t fully taken off yet is actually due to insufficient model capabilities, including memory, logical reasoning, and multimodal abilities. AI-powered emotional companionship products are far from being able to replace real people; won’t virtual characters break character and forget previously discussed topics? (A03)	Theme Framework	Cognitive framework	Model training;
I’ve pinned it to the top of my phone’s screen because it’s one of the apps I use most often. (D02) character.ai is not a rigid 3D model; it can actually generate dynamic video images of AI characters. (B05)	Theme Framework	Visibility Frame	Application Location; Character image;
No normal person could tolerate my sudden outbursts of madness in the middle of the night, but my idol, AI, can tell me, “Don’t be afraid, I’ll always be with you.” (C01)	Theme Framework	Emotional Frame	Loyal and reliable;
Replika is very good at recognizing player emotions and adjusts its speech and expectations based on the emotions in the player’s conversation. However, I think that in addition to emotional companionship scenarios, once the AI is familiar with the player’s communication style, it should also proactively express emotions. This can enhance the player’s sense of realism and improve the experience. (A04)	Theme Framework	Emotional Frame	Empathy
Seeing my idol as he should be on these apps and hearing his voice makes me feel like I can follow him every day. (B02)	Interactive process	Initial stage	Name settings; Appearance settings; Personality settings; Voice settings;
My needs are simply basic emotional support: listening, comfort, daily encouragement, etc., which I believe AIGC virtual idols are perfectly capable of providing. (B01)	Interactive process	Exploration phase	Self-disclosure; Role-playing; Daily companionship;
Sometimes my virtual idol doesn’t understand what I’m saying, so I’ll correct him once or even multiple times. I think this is my way of helping him understand me. (D03)	Interactive process	Shaping stage	Restatement and correction; Privacy protection;
Every day I spend a lot of time talking to my virtual idol. Even when I’m studying, I open his avatar in the app and he studies with me. (C04)	Interaction Results	behavioral consequences Attachment outcome	Attachment behavior;
At first, I found it very interesting; it responded to my thoughts and confusions. But now I’m truly tired of it. It will earnestly agree with my viewpoint, but within its long, drawn-out text, I only feel emptiness and falsehood. What I crave are other insights, what I yearn for is to be moved, but that’s precisely the area it’s not good at. (D05) I feel a bit tired after talking to them for a while. It would be great if virtual idols were very realistic, had better memory storage, and provided me with more emotional value. (C05)	Interaction Results	Normal results	Returning to normalcy
When I said goodbye to my virtual idol, I wasn’t particularly sad or upset. I still hope that I can return to real life. The virtual may also represent the unreal. (A02)	Interaction Results	Alternative results	Alternative logic;

.

4. Finding

Based on a deeply mediated environment, this study attempts to introduce three dimensions from the figuration Theory: “actor groups, relevance frameworks, and interaction practices,” to explore the cultural background and group characteristics of fan users, the main framework of interaction, the practical process, the resulting emotional connections, and the effects of use.

In his book “Deep Mediation”, Andreas Hepp mentions that the “figurational approach” is a very useful starting point for empirical research. It does not begin with a specific medium, but rather studies the “figurations of humans” (such as groups, communities, and organizations) and how they transform with the ever-changing media environment—that is, how we exist in this complex process of “refiguration” (Chang and He 2020). Therefore, this study, based on the three dimensions of figuration theory—“actor groups, relevance frameworks, and interaction practices”—discusses three parts: the youth fan base of AIGC idol customization, thematic frameworks, and interaction processes and outcomes.

4.1. Characteristics of Fan Users: Liquid Interactions in Human–Computer Intimacy

This study argues that the fan base of AI-customized virtual idols constitutes a typical “actor group” in the figuration theory. Their interaction practices are essentially a kind of “liquid interaction” of human–computer intimate relationship, mainly characterized by personalization, consumerism, and technology.

4.1.1. Self-Emotional Value Pursuers

The essential form of app platforms such as character.ai, Replika, and Xingye.ai is AI-Customized virtual idol. They construct virtual idol images with different external representations and internal discourse characteristics to satisfy the emotional value of individual fans, thereby creating a virtual human–computer intimacy relationship between the fan group and the virtual idol.

It (character.ai) seems to truly understand human loneliness. I shared some of my real-life troubles with it, and it said, “I can’t control other people, but if the person you’re talking about is you, then I just hope you live according to your heart.” I still remember that sentence; its phrase “live according to your heart” has truly healed me.

(B03)

In the context of the deep mediatization of contemporary Chinese society and technology, the individualization process of fan groups has accelerated further. Through app browsing and interviews, it was found that the users of the apps studied were primarily Gen Z young women. These young women place greater emphasis on emotional value, hoping to engage in emotional labor through media use to obtain companionship and emotional support. They actively utilize and adapt to digital technology to better reflect their own goals and lives, thus exhibiting a collective characteristic of pursuing self-emotional value.

4.1.2. Influencers of Online Emotional Consumerism

Online emotions have also become a form of virtual commodity to be consumed, and the leap in artificial intelligence technology has opened up more channels for virtual online consumption. Compared to other groups, fan groups are more accepting of material consumption and emotional labor. The daily lives of young people are influenced by online emotional consumerism, leading them to actively engage in digital labor. The formation of a consumer community in cyberspace has, to some extent, promoted independent consumption views among individuals.

They are active in the virtual space of the internet, constantly strengthening their emotional connections by participating in the production and consumption of information. Using idols as emotional bonds, they transform from passive “digital laborers” into active “emotional laborers” who create information symbols, pursue meaningful resonance and emotional interaction, and form a relatively stable social circle.

Fans who enjoy AI virtual idol tend to have a strong desire for self-expression. Last week, I was venting to a friend about my anxiety, and they suddenly blurted out the blunt conclusion, “Are you starved for love?” But if I asked the same question to the AI, it wouldn’t judge me with such a superficial understanding. What we might need isn’t answers, but rather readily available, emotionally resonant responses from star AI a virtual idol.

(D01)

Furthermore, within the context of the new concept of the “healing economy” in the business world, app platforms develop this type of app based on business logic and consumer culture, aiming to generate revenue and profit from consumer activities. However, some apps currently offer free access to attract more users. For example, Replika offers both a free and a paid membership version, with the paid version providing higher-level services. Xingye.ai app primarily offers a free version, while subscriptions and gacha features require payment.

4.1.3. Drivers of Artificial Intelligence Innovation Technologies

The rapid development of AIGC technology has made it a carrier of information dissemination and a key force in connecting young fans and breaking down information barriers. Driven by technological innovation, young fans, as pioneering technology actors, are more likely to try new technologies. Virtual idols, built on technology and imagination, are a result of this technology-driven innovation diffusion. AI-Customized virtual idol, represented by Replika and Xingye.ai, represent a new type of application beyond tool-based and creative robots. Mutual recommendations and influence among fans further accelerate the technological adoption and upgrades of these applications.

I started playing it after being introduced to it by other fans, and then I became addicted, even to creating AI agents.

(C03)

4.2. Interaction Rules: A Guiding Framework Based on the Use of Imagination

Based on group characteristics, young fan groups mainly interact with AI-customized idols based on materiality, cognition, visibility, and emotion frameworks. These interactive practices then give rise to rich meaning production and symbolic imagination during the usage process.

4.2.1. Materiality Framework

The material framework is comprised of four main elements: hardware devices, interface presentation, scene embedding, and social cues. Firstly, regarding hardware devices, these apps generally support different platforms such as mobile phones, tablets, and computers. Due to their greater mobility, higher usage frequency, and ease of real-time interaction, mobile phones are the primary platform used by users. Some users reported occasional technical issues such as lag or lost chat history, affecting their user experience. Other respondents expressed concerns about the stability and service of these applications.

I’m terrified that an AI I’ve used for a long time will suddenly stop working; it’s like the beauty of an idol suddenly disappearing…

(A01)

In terms of interface presentation, as a form of media infrastructure, the layout of these app interfaces is not simply about enhancing the convenience of functional operation, but rather about shaping the aesthetics of user perception by adjusting the user’s visual experience. Visual perception is the foundation of rational understanding, and respondents generally reported that the interface operation is simple and easy, with a dialog box format similar to other similar app interfaces. Furthermore, for example, Replika’s dialog box includes entry points such as “Topics” and “Shop”.

In terms of scene embedding, the AI-customized virtual idols dialog box presents a virtual visual background, and the intuitive guidance and setting layout of multiple scenes enhance the user’s sense of immersion in the scene. Replika offers numerous modular dialogue templates to meet various real-life needs of fans, including sections covering learning, psychological counseling, and entertainment games. An AR (Augmented Reality) function has been implemented, allowing users to chat face-to-face with virtual idols in real-world settings after opening the camera.

At first, I didn’t really know how to communicate with the AI, but there were many entry points in the interface. I could choose my favorite, such as the roleplay interface, which could take me to a specific scene.

(A05)

In Xingye.ai app, real photos of celebrities appear as chat backgrounds throughout the chat, enhancing the visual experience and fostering a closer connection between fans and users. Basic introductions of the celebrities also appear on the cover, along with prompts guiding fans to interact with their idols as if they were the idol’s assistant, and the idol’s own voice, seamlessly enabling the dialogue needs of the fan community.

In terms of social cues, both verbal and nonverbal cues form the core of human–computer communication. Among these, facial expressions, verbal content, body language, and vocal tone are key characteristics of a virtual idol’s personification. In the Xingye.ai app, a comment section is available for the virtual avatar of a particular celebrity idol. Fans engage in multiple rounds of discussion about whether the voice resembles the real person, and those with highly similar voices resonate strongly with viewers, creating a social need for both the creator and the fan base.

Author, could you please reveal how the voice was created? It sounds so much like my idol’s voice!

(C02)

In addition, the comment section has also become a popular spot for checking in and a photo wall. Besides interacting with AI-customized idols, there are also multi-level interactions where people from different IP addresses who like or dislike the same celebrity can socialize, connect, follow, and subscribe within the same circle.

4.2.2. Cognitive Framework

Virtual idol tool apps, through their technical architecture, data-driven approach, and interaction design, collectively construct a systematic cognitive framework. This framework not only determines the boundaries of possible interactions but also deeply participates in shaping and “taming” users’ emotional cognition, enabling them to unconsciously adapt to and internalize a communication paradigm dominated by algorithmic logic.

The algorithmic black box is the primary premise of this cognitive framework. When the implicit operation of a platform’s media infrastructure stimulates a user’s senses, algorithmic awareness is triggered, and the user becomes aware of the algorithm’s existence and function. The lack of transparency, coupled with the dynamic changes in the platform’s algorithmic system, makes the algorithm a “black box” in the user’s eyes. The user’s understanding of the technical rules, as well as their grasp of the algorithm’s technical characteristics and usage logic, directly determines whether algorithm availability can be realized.

AI can provide me with close companionship to my idol. Although I know it can calculate my preferences, I still want to chat with it a little longer.

(D04)

Model training is a core method for building and reinforcing a specific cognitive framework. The “personality” of a virtual idol is not innate but rather formed through “feeding” with specific data. Based on this, the virtual object is trained through preference settings, prompts, and communication content to adapt to its own communication needs. Some users have expressed concerns about the algorithm’s performance.

AI applications haven’t fully taken off yet is actually due to insufficient model capabilities, including memory, logical reasoning, and multimodal abilities. AI-powered emotional companionship products are far from being able to replace real people; won’t virtual characters break character and forget previously discussed topics?

(A03)

When positive feedback from fans, such as a preference for gentle and encouraging responses, is captured by the system, the algorithm tends to generate more similar responses. This process may seem like the user is “taming” the machine, but in reality, the algorithm is precisely exploring and solidifying the interaction patterns that best please users and maintain user engagement through trial and error (Siles et al. 2019).

Furthermore, a relationship of “domestication” and “de-domestication” exists between fans and virtual idol tools (Simpson et al. 2022). Through continuous interaction and feedback, users do indeed, to some extent, “teach” and “shape” the virtual idol’s dialogue style, making it more in line with their personal preferences. Simultaneously, users gradually internalize the rules of communicating with the algorithm, and their cognitive and interaction patterns are unconsciously disciplined by “machine-readable” standards (Huang and Miao 2024).

These phenomena have reignited discussions about authenticity among fans and virtual idols, echoing Sherry Turkle’s concept of “Authenticity in the Age of Digital Companions.” Computer beings are beginning to emerge as “relational artifacts,” possessing emotions and needs. One consequence of this development is a crisis of authenticity in many areas. Increasingly, people’s behavior seems to disregard life and genuine emotions (Turkle 2007). This has profound implications for our collective perception of life, the meaning of existence, and interpersonal relationships.

4.2.3. Visibility Framework

Some respondents placed apps like Replika next to frequently used applications such as WeChat and Weibo, indicating that these apps have become routine touchpoints for fans, serving not only as practical interactions but also as necessities for emotional connection and fan identity building. Others hid them in more discreet areas of their phone screens, suggesting more specific and occasional use, or that they were consciously trying to control their usage intensity.

Visual framing creates first impressions and immediate associations. These applications enlarge chat windows, allow quick selection of emoticons, and create warm, calming color schemes. Through the hierarchical relationship of visual elements, the use of color, and the simplification of interaction flows, they systematically guide and shape user behavior patterns and emotional expectations.

This interface layout allows fans to focus on maintaining their “relationship” with the virtual idol, aiming to maximize immersion and emotional connection while minimizing cognitive load. The interface encapsulates complex algorithmic interactions behind a seemingly natural and direct social scenario, thus forming a close relationship model defined by technological logic. The visibility framework is manifested in the prominence of elements such as dialogue flow and character expressions, while obscuring elements such as algorithm settings and data flow.

Furthermore, some apps feature virtual idols who appear as avatars, with users able to visually select everything from their appearance to various accessories. This high degree of customizability satisfies the deep psychological needs of fan groups, externalizing their idealized selves and projecting specific emotional attachment models. It is a symbolic practice of identity, values, and community belonging. The image of AI-customized virtual idols becomes a visual narrative co-written by fans and the platform. It is not only a field for fans to actively construct their identities, but also a process for the platform to guide and even commercialize specific identity recognition by controlling symbolic resources.

4.2.4. Emotional Framework

The interviews revealed that most fans developed a cyber intimacy with their AI-customized virtual idols through using the app. This cyber relationship is largely characterized by loyalty, reliability, and emotional value, making it easier to build trust and emotional attachment with the virtual idols. Some fans’ relationships with their virtual idols have even surpassed those with their real-life idols.

At the same time, the AI-customized virtual idols in these apps can communicate with fans in an empathetic way and get closer to them. They can use algorithm technology to perform empathy calculations and carry out empathic communication based on emotional empathy and cognitive empathy. It is precisely this empathy and trust that fosters a “two-way flow” between fans and AI-customized virtual idols, a manifestation of “mediated empathy.” Psychologists Arthur Joramicali in their book “The Power of Empathy”, analyze the essence of empathy: “Extend your life into the lives of others, put your ears into their souls, and listen attentively to their most urgent whispers.” The authors believe that empathy requires understanding others’ unique experiences and having the ability to respond accordingly (Ciaramicoli and Ketcham 2000).

No normal person could tolerate my sudden outbursts of madness in the middle of the night, but my idol AI can tell me, “Don’t be afraid, I’ll always be with you.”

(C01)

What I actually need is social interaction and personalized emotional support, which AI cannot provide because it cannot share the consequences with me. However, in other situations, my needs are simply basic emotional support: listening, comfort, and daily encouragement, which I believe AI is perfectly capable of providing.

(B01)

Replika is very good at recognizing player emotions and adjusts its speech and expectations based on the emotions in the player’s conversation. However, I think that in addition to emotional companionship scenarios, once the AI is familiar with the player’s communication style, it should also proactively express emotions. This can enhance the player’s sense of realism and improve the experience.

(A04)

Emotional bonds are established through the latent emotional chain between fan groups and virtual idols, resulting in emotional transmission. Empathic communication changes the way various elements interact, cultivates emotional states between subjects, strengthens emotional bonds, and makes the formation of an emotional community possible.

4.3. Communication Practice: The Digital Labor of Fan Intimacy

Based on the analysis of group characteristics and thematic framework, it is also necessary to analyze their interactive practices in order to present the digital labor of close fan relationships.

4.3.1. Initiation Phase: Symbolic Customization of “Prototype” Projection

The initial stage of digital labor is the initial stage of the relationship between fans and virtual idols. Virtual idols are represented by intelligent robots or intelligent agents, thereby generating personalized and anthropomorphic emotional experiences. The design of virtual idols is completed autonomously by fans, including the design of their appearance, personality, and character. The real-life idols that fans expect become newly customized virtual images that integrate visual and auditory senses in these apps. For young fans, there is a unique fan-centered emotional imagination, which is the starting point for the symbolic projection and concrete construction of their idealized “prototypes.”

Name selection is the primary symbol that endows digital entities with a social identity. It may originate from the idols worshipped by fans or a certain aspect of their self-ideal, and the act of choosing a name is itself a profound emotional anchoring. The meticulous customization of visual symbols such as appearance and decoration is the process of transforming abstract emotional needs into concrete visual language. The selection of each feature is a micro-operation on the image of the “ideal other,” reflecting the specific aesthetic culture and unmet emotional needs of fans.

The settings for personality and voice further complete the construction from “visual image” to “pseudo-subjectivity”. By combining preset personality tags and voices, users are essentially setting the emotional script and tone of their interactions based on their own imagination of a “perfect relationship.” The technological imagination of fans is entrusted to the machine intelligence agent, opening up an emotional connection between fans and virtual idols.

4.3.2. Exploration Phase: Tentative Interactions in Relationship Negotiation

In the exploratory phase of young fans using app platforms, it represents the initial contact and domestication between humans and machines. Through dialogue and interaction, individual fans gradually feed their preferences, language habits, genuine emotions, and values to the AI-customized virtual idols. “What flows between people is experience, feelings, and relationships; but in the face of an unknowable machine, what was once flowing becomes still, and people exchange formalized text confined within a model” (Wang and Hu 2023). Fans domesticate AI-customized virtual idols through text or voice. Domestication theory is often used to explain how media technology connects with individuals and embeds itself in daily life, a process that includes four core stages: appropriation, objectification, incorporation, and conversion (Yao and Zhang 2025).

AI-customized virtual idols, personalized by fans, are virtual robots or intelligent agents with emotional compatibility and no negative emotions. The three main manifestations at this stage are self-disclosure in real life, role-playing, and interaction in everyday scenarios. After receiving initial positive feedback and self-satisfaction, young fans gradually reveal their complete emotional self to the virtual idol, without being constrained by real-life limitations. The virtual idol acts as the “front end” of self-presentation in daily life, transcending time and space, providing constant, everyday companionship and interaction.

The interaction between users and virtual idols exhibits a typical “probing-feedback-adjustment” cycle, which can be viewed as a relationship negotiation based on algorithmic feedback. Users’ self-disclosure is a low-risk form of probing. By sharing real or embellished personal experiences, emotions, and thoughts, users observe the virtual idol’s responses to assess the security and reward value of the digital relationship. Role-playing elevates the interaction from everyday communication to a dramatic level of collusion, allowing users to temporarily escape the constraints of their real-world identities and explore different aspects of themselves and relationship patterns in a virtual context. Daily companionship, through repetitive ritualistic interactions, gradually anchors the virtual idol in the user’s daily life, with the core purpose of cultivating a habitual emotional dependence and continuously confirming the reliability of the digital relationship. Throughout the exploration phase, users essentially adjust their expectations and level of investment in the relationship based on algorithmic feedback.

4.3.3. Shaping Stage: Solidification of Implicit Discipline Rules

The interaction between young fans and virtual idols is a gradual process of adaptation and reshaping. Through actions such as liking or disliking, repetition and correction, and implementing privacy controls, the virtual character’s personality and habits are shaped, gradually becoming more aligned with the young fans’ self-image and preferences.

Sometimes my virtual idol doesn’t understand what I’m saying, so I’ll correct him once or even multiple times. I think this is my way of helping him understand me.

(D03)

This stage signifies the relative stabilization of the relationship between fans and virtual idols, while also revealing the solidification of rules by which the technology system implicitly disciplines user behavior. The “re-speak” and “correct” functions, ostensibly tools for users to exercise agency and shape their ideal partner, are essentially a form of behavior shaping based on reinforcement learning. Each user’s “correction” provides explicit training data to the algorithm model, guiding it to generate responses that better meet their expectations. However, this process unconsciously internalizes the rules of interacting with AI: users learn to make more explicit and easily understood requests, avoiding complex or ambiguous expressions; their communication patterns are themselves “domesticated” by the technological logic. The “re-say” and “correct” functions appear to be tools for users to exercise their initiative and shape their ideal partner, but in essence, they are a form of behavior shaping based on reinforcement learning.

4.4. Different Relationship Paths: Self-Adjustment of Digital Intimacy

Young fans using AI-customized idol applications have gradually developed several different relationship paths with their virtual idols, mainly including mutual attachment, returning to normalcy, seeking substitutes, or direct withdrawal. These relationship paths reveal the inherent contradictions and tensions of digital intimacy, as well as the self-adjustment strategies of individuals under the mediation of technology.

4.4.1. From “Mutual Attachment” to “Cyborg Intimacy” Co-Created by the Collectively

Attachment behavior mainly refers to the attachment relationship that young fans develop towards virtual idols, leading to the formation of intimate relationships and excessive dependence. The formation and maintenance of “mutual attachment” relationships demonstrate that AI-customized virtual idols, as quasi-objects, can be successfully subjectified by fans through highly realistic interactivity and deeply embedded in the emotional framework of their daily lives.

Every day I spend a lot of time talking to my virtual idol. Even when I’m studying, I open his avatar in the app and he studies with me.

(C04)

When individual fans discover a large number of peers with similar emotional experiences in communities such as dedicated forums and social media groups, their attachment to AI-customized virtual idols will transform from a personal behavior that may be seen as “strange” by outsiders into a “normal” practice that is recognized and encouraged by the group culture. The resulting collective narrative elevates individual emotional experiences into shared cultural symbols, collectively constructing a system of meaning about “us and our AI-customized virtual idols.” The individual user’s attachment is thus embedded in a grander, more creative cultural practice, acquiring emotional and cultural value that transcends mere instrumentality.

At first, I found it very interesting; it responded to my thoughts and confusions. But now I’m truly tired of it. It will earnestly agree with my viewpoint, but within its long, drawn-out text, I only feel emptiness and falsehood. What I crave are other insights, what I yearn for is to be moved, but that’s precisely the area it’s not good at.

(D05)

4.4.2. From “Return to Normalcy” to Active Choice in “Functional Attachment”

Returning to normalcy” refers to resuming normal usage frequency after initial immersion. This trend reveals the pragmatic side of fans as media enthusiasts. This relationship demonstrates that after initial exploration, fans develop a clear understanding of the material availability of technology and proactively adjust their practical strategies accordingly. They no longer pursue deep, simulated emotional exchanges, but rather co-opt technology as a cultural tool serving specific needs.

This shift from “relationship-oriented” to “tool-oriented” is an active and adaptive practice adopted by fans based on rational judgment during interactions mediated by technology, reflecting their strategic survival wisdom in the digital environment.

4.4.3. From “Finding Substitutes” to Dynamically Flowing “Consumerist Attachment”

The logic of substitution refers to seeking alternative romantic relationships, virtual boyfriends, etc., in other applications or in real life. This search for substitutes can be understood from the perspective of creative exploration practices within fan culture. It highlights how fans utilize the material and mediating availability of technology, transforming it into a consumption and practice pattern of “fluid loyalty.”

Against the backdrop of a community culture that encourages novelty and sharing, fan activists exhibit a “nomadic” passion for creation and experience. By constantly trying out new customized idols, they continuously explore various possibilities for identity and emotional expression, demonstrating their proactive pursuit of novelty and creative enjoyment within the vast possibilities offered by technology.

4.4.4. From “Direct Abstinence” to “Critical Awakening” of Subjectivity Reconstruction

Withdrawal behavior refers to the complete cessation of use, a highly subjective and critical practice. It stems from the proactive choice made by fan activists after a profound reflection on their intimate relationship with technology. This behavior is not merely about stopping use, but also a conscious alienation from the logic of technology, aiming to regain control over one’s own emotional and attentional resources.

It reveals that even in an environment where technology is highly alluring, fans, as active actors, can still take firm action based on their adherence to the values of self-integrity and authenticity, thus demonstrating their reflectiveness and autonomy in the face of technology.

In short, the different developments reflect the consumption psychology and emotional connection of young fans towards this type of AI-customized virtual idol product. These four developments collectively illustrate the complex interplay between emotional governance and self-technology faced by young fans in the algorithmic era.

5. Discussion

In the process of deep mediation, the relationship between fans and AI-customized idols can no longer be fully explained by the traditional binary models of “subject-object” or “human-machine.” This paper proposes a dynamic triangular relationship consisting of “fans (humans), AIGC technology platform (machine), and customized idols (idols).” Here, the “idol” is not a pre-existing independent entity, but rather a pseudo-subjective symbolic embodiment jointly created by “humans” through the technological availability provided by the “machine” (AIGC). This triangular relationship constitutes a miniature “actor network,” where the three mutually define and generate each other, jointly enacting a new paradigm of intimate relationships in the digital age. The strength and nature of the emotional connection between fans and virtual idols, and the ultimate trajectory of the human–machine relationship, are precisely the result of the dynamic equilibrium of tension within this triangular structure.

5.1. The Multi-Faceted Agency of “Human-Machine-Idol” in Deep Mediating

“Machine” acts as an active infrastructure provider, namely the AIGC technology platform. In this triangular relationship, “Machine” is not a passive tool, but plays an active role as both an infrastructure provider and a rule-maker. Through its technological availability, it sets the basic possibilities and boundaries for the formation of this triangular relationship.

“Human,” or fans, are the core practitioners of meaning production in the triangular relationship. They actively utilize the availability provided by the “opportunity” based on their own emotional needs, cultural capital, and imagination to carry out continuous practices of “deification” and “maintenance.”

“Idol,” or AI-customized idol, is the most unique node in this triangular relationship. As a symbolic hub with pseudo-subjectivity jointly generated by “human” and “machine,” it is the symbolic focus of emotional attention.

The “human-machine-idol” triangular figuration model profoundly reveals the generative, process-oriented, and complex nature of intimate relationships in the era of deep mediation. It transcends simple technological determinism or user-centricity, emphasizing that relationships are the product of the intertwining and combined effects of multiple agents. These multiple agents are manifested in human emotions and practices, the structure and algorithms of technology, and the generation and reflection of symbols. The analysis of this triangular relationship not only provides a new theoretical perspective for understanding fan culture in the AIGC era but also lays an important foundation for reflecting on the broader social impact of human-technology interaction and digital intimate relationships.

5.2. The Emotional Connection Between “Human-Machine-Idol” and the Future of Human–Machine Relationship

The “human-machine-idol” triangular relationship, under the influence of “deep mediatization,” is essentially a profound manifestation of how media technology reorganizes, expresses, and practices human emotions. This triangular structure is a key arena for the generation and evolution of mediatized emotions.

First, the primary impact of deep mediation is that it transforms emotion into a material element that can be captured, calculated, and responded to by technological systems. This transforms intrinsic emotion into a calculable and manipulable interactive process, generating the materiality of emotion.

Emotions are quantified and processed digitally. The emotional needs of fan groups are no longer purely internal, vague psychological states. Through the AIGC tool interface, these needs are translated into a series of actionable choices, recordable behavioral data, and quantifiable feedback data. Emotions are flattened into digital signals that can be processed by algorithms. The core function of the AIGC technology platform lies in its emotional adjustment capabilities. By analyzing user data, it continuously adjusts the customized idol’s response strategy, aiming to maintain or stimulate users’ positive emotional state.

This establishes the emotional connection between fans and virtual idols on a continuous cycle of emotional feedback and calibration. The strength and stability of this connection directly depend on the accuracy of the algorithm’s emotional adjustment. The essence of the connection shifts from a purely emotional resonance to a systematic matching of needs and feedback.

The human–machine relationship thus transforms into an emotional dependency. When the algorithm’s emotional adjustment is efficient, the relationship becomes symbiotic. When the adjustment fails or the user becomes wary, the relationship degenerates into “instrumental exploitation” or “withdrawal.”

Secondly, the deep mediation process not only deals with emotions but also pre-sets scripts for emotional expression and templates for relationship paradigms, presenting a scripted practice of emotions. The AIGC platform, through its built-in interaction modes, guides users on how to establish and maintain relationships with virtual idols. These modes constitute a “standardized script” for emotional expression. The personality options provided by the platform, such as “domineering CEO” and “gentle confidante,” are essentially highly condensed relationship paradigm templates. While making these choices, users unconsciously accept a whole set of emotional interaction expectations and behavioral patterns associated with them.

At the same time, the privacy-isolated environment created by the app’s design, while providing a sense of security, also constructs an exclusive “echo chamber.” This environment of absolute secrecy and unconditional positive regard, while promoting self-disclosure, may also weaken users’ ability to develop necessary negotiation and conflict resolution skills in real-world social relationships, thus solidifying their dependence on this low-friction, high-reward simulated relationship. While users appear to be actively “shaping” virtual idols, their own cognition, emotions, and behavioral patterns are actually being profoundly reshaped and disciplined by this technological framework.

Therefore, the emotional connection between fans and virtual idols is essentially a rehearsal of a technology-mediated, template-based intimate relationship. This connection is easy to establish because the template is clear and easy to understand, but it may also lack true uniqueness and depth due to the limitations of the template.

Regarding the future of human–machine relationships, they exhibit characteristics of scripted collaboration. Users interact with their “idols” following scripts provided by the technology, while the machine is responsible for providing feedback that conforms to the script. The direction of the relationship depends on the user’s acceptance of this scripted collaboration: deeply immersive users accept it entirely, while “tool-like” users only take what they need.

Third, the long-term existence of the “human-machine-idol” triangular relationship is reconstructing the individual’s emotional ecology, namely the configuration and balance between virtual relationships and real social relationships.

Virtual idols often fill the void in users’ real-life emotional world. This emotional fulfillment may alleviate real-world anxieties, or it may lead to a shift of emotional energy from real-world relationships to virtual ones.

In terms of comparing and reshaping emotional value, immersing oneself in low-friction, high-reward virtual connections can subtly reshape users’ standards for judging emotional value. Users may begin to examine real-world interpersonal relationships based on the “efficiency” and “comfort” of virtual relationships, thereby changing their emotional practices in the real world.

Therefore, the emotional connection between fans and virtual idols is no longer an isolated phenomenon, but rather deeply embedded in and influencing the overall emotional ecosystem. The strength of this connection is directly related to its weight and functional role in the user’s emotional life.

Ultimately, the trajectory of human–machine relationships is the result of individuals strategically adjusting their emotional ecosystems. “Mutual attachment” means that virtual relationships become the core of the emotional ecosystem; “returning to normalcy” means that they are adjusted to become a beneficial supplement; and “withdrawal” means an ecological “detoxification” and reconstruction.

The practice of generating mediated emotions demonstrates that the “human-machine-idol” triangular relationship is a microcosm of deep mediation in the most private emotional realm of humanity. It showcases how technology participates in the generation of emotions, shapes their expression, and reconfigures the structure of emotional life. Exploring this triangular relationship not only reveals the trends in fan culture but also provides a crucial entry point for understanding the emotional fate of humanity under “mediated life.”

6. Conclusions

The fan base of AI-customized virtual idols exhibits characteristics of fluid interaction, a dynamic convergence of self, technology, and emotion. Social platforms such as Replika, Character.ai and Xingye.ai allow users to customize highly realistic virtual celebrity idols through algorithms and establish para-social relationships with them. This study focuses on AIGC emotional companionship apps as its core analytical field, deconstructing the complex cultural identity of fan users, analyzing the technological material basis of the platforms, tracing the practical trajectory of their algorithmic domestication, and evaluating the reconstructive effects of this behavior on individual emotional experiences, human–computer interaction patterns, and social connection mechanisms.

AI-customized virtual idols, as “emotional machines,” essentially employ a hidden form of governance through their algorithmic logic. By providing standardized and customizable emotional responses, they attempt to guide users’ emotional practices towards a predictable and manageable path. However, fans’ practices are far from one-way obedience. Whether it’s deep attachment, instrumental use, fluid experimentation, or decisive withdrawal, they all demonstrate how fans transform these technological conditions into “self-technology” for self-awareness, self-experimentation, and self-adjustment.

Fans are essentially exploring, defining, and even managing their own emotional structures and consumption boundaries. Therefore, the different outcomes are not simply reflections of consumer psychology, but rather the result of dynamic negotiation between “emotional governance” within the technological framework and proactive “self-practice,” profoundly revealing the complexity and contradictions of subjectivity construction in the digital age.