Computers in Human Behavior 167 (2025) 108625

Contents lists available at ScienceDirect

Computers in Human Behavior

journal homepage: www.elsevier.com/locate/comphumbeh

An assistant or A friend? The role of parasocial relationship of

human-computer interaction
Tiejun Qi * , Hongshen Liu , Zhihui Huang
School of Economics & Management, Changsha University of Science & Technology, Changsha, 410076, PR China

A R T I C L E I N F O A B S T R A C T

Handling editor: Nicolae Nistor The study explored the dual-path mechanism of human-computer parasocial relationships on consumer attitudes
through an experimental approach. Innovatively, ‘assistant’ and ‘friend’ are proposed as the two basic parasocial
Keywords: relationships of human-computer interaction, with the two gender traits of artificial intelligence robots - "male
Human-computer interaction trait" and "female trait" - to stimulate consumers’ perceived competence and warmth through interaction, and
Parasocial relationship
improve consumer attitudes. Research has found that perceived competence and perceived warmth mediate the
Stereotype
impact of two parasocial relationships, "assistant" and "friend", on consumer attitudes. Compared with the
Consumer attitudes
Marketing "friend" type of human-machine parasocial relationship, the "assistant" type of human-machine parasocial rela­
tionship significantly enhances consumers’ perceived competence; Compared with the "assistant" type of human-
machine parasocial relationship, the "friend" type of human-machine parasocial relationship significantly en­
hances consumers’ perceived warmth. Meanwhile, research has demonstrated the moderate role of gender traits
in artificial intelligence robots. This study broadens the research perspective of AI Services marketing, and
provides important enlightenment for enterprises to develop Marketing strategy and AI Intelligent design.

1. Introduction through mechanical intelligence and analytical intelligence, respec­

tively. AI anthropomorphism emphasizes the "EQ" of AI, which relies on
The rapid development of generative AI has led to increasingly empathy to reach cognitive and emotional empathy, which can be
frequent human-computer interaction. Large scale training models achieved through intuitive and empathic intelligence, respectively
allow AI to communicate like real humans. According to the "Mordor (Gladstein, 1983; Huang & Rust, 2018). The theory of relational norms
Intelligence" data, the social robots market size is expected to grow from proposes that there are two basic relationships, public and social, in
USD 4.26 billion in 2023 to USD 17.32 billion by 2028, at a CAGR of human social interactions(Shuqair et al., 2021). Under
32.40% during 2023–2028 (Mordor Intelligence, 2023).The develop­ perception-driven, different social relationships will stimulate different
ment of artificial intelligence technology has changed the landscape of perceptual experiences of consumers, which in turn affects consumer
the service industry. The process of producing and delivering services attitudes (Zhang et al., 2022). It can be seen that human-computer
has changed from human interaction to human-computer interaction interaction has a unique functional cognition and emotional experi­
(Noor et al., 2021; Troshani et al., 2020; Monod et al., 2022). However, ence, which is in line with the path of influence on consumer attitudes
the social interaction skills of AI need to be further improved. Especially and is particularly important for improving consumer attitudes(Park
in the pharmaceutical services industry, consumers are resistant to et al., 2021). Based on the analysis of related literature, the study con­
intelligent service robots(Kitamura, 2023). cludes that the following issues need to be further explored.
There are two important forms of AI to enhance the effectiveness of First, the definition of human-computer parasocial relationship still
human-computer interaction in the field of service marketing: AI so­ needs to be explored. Currently, most of the research on parasocial re­
cialization and AI anthropomorphism(Li & Suh, 2022). AI socialization lationships of AI robots applies the "Computer as a Social Actor (CASA)"
is based on organizational socialization, with a focus on giving con­ paradigm(Reeves & Nass, 1996). During human-computer interaction,
sumers the best solutions and recommendations. This can be achieved humans apply rules of politeness to computers and tend to appreciate

This article is part of a special issue entitled: AI & marketing published in Computers in Human Behavior.
* Corresponding author. 45, Chiling Road, Changsha City, Hunan Province, PR China.
E-mail address: [email protected] (T. Qi).

https://doi.org/10.1016/j.chb.2025.108625
Received 14 March 2024; Received in revised form 1 February 2025; Accepted 27 February 2025
Available online 27 February 2025
0747-5632/© 2025 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

technologies that support them(Jonsson et al., 2017; Nass et al., 1999). Through the difference test and regression analysis, we proved the
At the same time, humans will build relationships with robots and gain different promoting effects of the two paraisocial relationship paradigms
mutual trust, which in turn leads to more authentic and trustworthy on consumer perception and consumer attitude, and proved the exis­
emotional support(Cheng et al., 2022; Meng & Dai, 2021).Recent studies tence of mediation. A total of 137 subjects participated in Experiment 2,
have found that humans label robots with identities(Kim et al., 2021), which further examined the effect of the interaction of parasocial rela­
and Xie et al. (2023) believes that visual cue and identity cue play a tionship and gender traits on consumer perception, and demonstrated
crucial role in interaction satisfaction. Existing studies support the CASA the existence of moderation.
paradigm, however, most scholarly research proves the existence of
human-computer parasocial relationships, and the definition of 2. Literature review
human-computer parasocial relationships still lacks a systematic elab­
oration, as well as ignores its impact on consumer attitudes. 2.1. Parasocial relationships
Second, the path mechanism by which human-computer parasocial
relationships act on consumer attitudes is unclear. With the Stereotype 2.1.1. Computer are Social Actors (CASA) paradigm
Content Model (SCM) being widely used to explain brand perception Most studies on parasocial relationships with AI robots have applied
(Kolbl et al., 2019), corporate image(Gao & Mattila, 2014), trust the "Computer are Social Actors (CASA)" paradigm(Reeves & Nass,
mechanism(Christoforakos et al., 2021), purchase intention(Xue et al., 1996), which has shown that the social rules guiding human interactions
2020), etc. Numerous studies have verified the feasibility of applying are equally applicable to human-computer interactions. Currently, there
stereotype modeling to multicultural, national, and historical experi­ are two main explanations for this phenomenon. One perspective,
ences (Fiske, 2012, 2018; Kotzur et al., 2020). Recent studies have found rooted in evolutionary theory, suggests that humans perceive AI robots
that the level of warmth and competence is predicted by group per­ as social entities comparable to themselves in intelligence, leading them
ceptions, and that perceived warmth and competence together lead to to instinctively respond to the technology as if it were a social actor(Lee
contempt, jealousy, and paternalistic external group evaluations, as well et al., 2006). Another explanation posits that humans exhibit this
as admiring evaluations of alliances and internal groups. Consumer at­ behavior because they perceive themselves as interacting with the cre­
titudes, on the other hand, are a concentration of consumer perceptions, ators or programmers of the machine(Searle, 1981). In recent years, a
emotions, and behaviors. Thus, based on the stereotype model, this wealth of research has supported the CASA paradigm. Kim et al. (2019)
study proposes the consumer cognitive dimensions of perceived found that humans appreciate technologies that support them. Cheng
competence and perceived warmth as the main factors influencing et al. (2022) discovered that humans can establish relationships with
consumer attitudes. robots and develop trust. Meng and Dai (2021) suggest that robots can
Finally, the traits and features of service-based AI that affect con­ serve as authentic and credible sources of emotional support for humans.
sumer perception are yet to be studied. In many studies on stereotypes, Additionally, Kim et al.(2021) found that humans often label robots as
we find that consumers’ perceived competence and perceived warmth "assistants" or "friends".
are affected by gender stereotypes to a certain extent. As a man-made
machine with human-like attributes, AI has yet to be studied in terms 2.1.2. Parasocial relationships in human-computer interactions
of its basic characteristics, and in particular, research related to the The concept of parasocial relationships (PSR) was initially proposed
gender traits of AI robots is still in its infancy. This study focuses on the by Horton and Wohl (1956) and further expanded upon by Rubin et al.
moderating role of gender traits in AI robots, investigates the impact of (1985, 1987). Traditionally, parasocial interaction (PSI) has been
human-computer parasocial relationship interactions on consumer at­ defined as a one-sided form of interpersonal interaction, where the
titudes, and uses this to explore the clues of gender traits setting in AI audience unilaterally establishes a connection with media figures. With
robots. the widespread adoption and advancement of information technology,
Overall, many current studies reveal that AI robots have a significant parasocial relationships have extended into the digital domain,
role in advancing enterprise management, optimizing the service pro­ encompassing entities such as robots, voice assistants, social platforms,
cess, enhancing brand impression, and strengthening consumer experi­ websites, etc.(Labrecque, 2014). From the perspective of the CASA
ence. Some researches have found that AI robots show both paradigm, due to the technological support of virtual information, par­
anthropomorphic characteristics and empathy while interact with con­ asocial relationships are no longer confined to unidirectional commu­
sumer are more likely to be accepted then those with anthropomorphic nication but have gradually evolved into complex bidirectional
characteristics alone(Pelau et al., 2021). However, few have studied the interactions(Daniel et al., 2018; Kassing & Sanderson, 2009). Multiple
important role of parasocial relationship between human-computer in­ factors closely correlate with the formation of parasocial relationships.
teractions and how it benefits consumer attitudes. Therefore the purpose Firstly, the characteristics of information sources become significant
of this study is to measure the role of parasocial relationship between factors influencing parasocial relationships(Sakib et al., 2020). Audi­
human-computer interactions(assistant paradigm & friend paradigm) as ences tend to establish parasocial relationships with reliable and
a motivated tool on consumer attitudes, and explore the mechanism in authoritative sources of information, and this trust is particularly crucial
the whole path. We designed different kinds of AI robots to observe the in virtual environments. Additionally, features of information content
feelings of participants by controlling its appearance, function, and (Ferchaud et al., 2018) and virtual self-attributes(Yuksel & Labrecque,
language style. In terms of mediators, we introduce perceived compe­ 2016) are also closely associated with the formation of parasocial re­
tence and perceived warmth by considering the social factors induced by lationships. Balakrishnan et al. (2022) introduced social self-efficacy as
parasocial relationship. a moderating factor, highlighting how social relationship expectations
Hence, through a 2*2 comparative experiment, we further revealed play a pivotal role in technology-driven services through the study of
that the two parasocial relationship improved consumer attitudes by system factors in human-computer interaction (perceived intelligence
stimulating their perceived competence and perceived warmth to and perceived anthropomorphism). This research reflects people’s ex­
varying degrees. Meanwhile, we take gender traits of AI robots as pectations of parasocial relationships with robots. Skjuve et al. (2021)
moderators which stimulate consumer perception based on the same designed a social chatbot called Replika and found in their investigation
theory(Stereotype Content Model, SCM). We assume that different that some participants regarded Replika as a better friend than humans,
gender traits will increase different consumer perception in specific while others still wished for Replika to serve as an assistant to help with
human-computer parasocial relationship. Two experiments were con­ tasks.
ducted in the study. In Experiment 1, 101 subjects were recruited Overall, research on parasocial relationships has made significant
through an online platform and they were divided into two groups. progress in recent years. Not only has it deepened our understanding of

2
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

its definition, but it has also expanded the scope of research, making it 2.3. Gender traits
more practically applicable in artificial intelligence services. However,
the current definition of quasi-social relationships between humans and 2.3.1. Gender schema in SCM
machines is still unclear, with limited research on its impact on various Martin et al. (2002) introduced the concept of gender schema, sug­
dimensions of consumers and consumer behavior. gesting that gender schema guides an individual’s response to
gender-related information. Gender information relies on societal ideals
2.2. Perceived competence and perceived warmth of femininity or masculinity, including behaviors expected of women or
men. For example, individuals are expected to conform to traditional
2.2.1. Stereotype content model (SCM) gender ideals, such as women being nurturing and men being strong.
In 1999, Fiske et al. introduced the Stereotype Content Model (SCM). Secondly, the concept of schema aids in understanding and recalling
This model has been widely used to study the nature and consequences information, implying a basic consistency in how individuals perceive
of stereotypes, providing a comprehensive framework for understanding gendered information. For instance, household chores are mostly asso­
the psychological reasons behind people’s judgments and categoriza­ ciated with women. This association creates a consistency between tasks
tions of others (Fiske et al., 2002). SCM, through the construction of four and the individuals performing them. Thirdly, gender schema can pro­
underlying assumptions, offers profound insights into the formation and vide a basic repository of information, aiding in judgment when infor­
evolution of stereotypes, contributing to the understanding of social mation is ambiguous. For example, when meeting a new acquaintance,
cognition between individuals and groups. Firstly, within SCM’s dual individuals may infer their interests and personality traits based on their
dimension hypothesis, Fiske et al. identified warmth and competence as gender. If the new acquaintance is female, one might assume she is
universal dimensions of social cognition between individuals and interested in makeup.
groups. This implies that when evaluating others, we often categorize In some related studies, gender stereotypes have gradually come into
them based on their performance in these two dimensions of warmth view. Based on social role theory(Eagly, 1987), women are more likely
and competence. Secondly, under the mixed assessment assumption, than men to be positioned as caregivers for children or the elderly. Men
SCM suggests that warmth and competence can coexist. This means that have a higher propensity for entrepreneurship, considered a masculine
a group or individual may exhibit simultaneous positivity or negativity trait(Gupta et al., 2008). Additionally, when adapting to a job, men are
in both warmth and competence. Thirdly, the social status assumption in perceived as better at handling stressful situations and adapting to
SCM suggests that the higher a group’s social status, the more likely they physical conditions than women. Conversely, women are believed to
are perceived as competent, and conversely, the lower their perceived excel in learning and are expected to adapt to interpersonal and cultural
warmth. Finally, the favoritism assumption posits that people generally relationships(DeArmond et al., 2006). Gender stereotypes can affect the
view in-group performance positively, leading to the derogation of assessment of targets’ competence and warmth(Broverman et al., 1972).
out-groups. This highlights the intergroup differences and opposition Generally, individuals perceive men as more competent than women,
within stereotypes (Baker & Florack, 2021). Fiske et al.(2012, 2018) and women as warmer than men (Ashton-James et al., 2019).
validated the applicability of the SCM across multiple cultures, coun­
tries, and historical experiences. While some underlying assumptions 2.3.2. Gender traits influence on AI robots services
were not fully supported, the fundamental aspects of the model remain Previous research suggests that people tend to project gender sche­
applicable for generalization. mas and stereotypes onto robots, especially when they possess human­
oid features and are placed in social roles(Alesich & Rigby, 2017).
2.2.2. Perceived competence and perceived warmth in AI robot services Studies have found that a robot should be more masculinized when
Based on the SCM, competence and warmth are two crucial di­ performing security-related tasks and more feminized when performing
mensions in social perception. Barchard et al., 2020 integrated the tasks related to guidance(Chee et al., 2013). Similarly, a female robot
concepts of belief, desire, and intention in human-computer interactions with long hair is perceived to be more empathetic than a male robot with
with psychological concepts of behavior, cognition, and emotion, short hair, while a male robot with short hair is seen as more agile than a
creating a 20-item Perceived Social Intelligence (PSI) scale to compre­ female robot(Eyssel & Hegel, 2012). Stereotypes also influence the
hensively measure perceptions of intelligent robot services, including design of robots. For example, a receptionist robot is designed with more
various implementation modes and behaviors, further demonstrating feminine features and evaluated as being more "welcoming." On the
the positive impact of both the competence and warmth of AI robots on other hand, when robots are designed with broader shoulders, they are
consumers. Researchers have also developed social attribute indicators perceived as being more "authoritative"(Trovato et al., 2018). These
for robots, combining measures related to warmth and competence, studies indicate that people believe robots with masculine features are
including assessments of likability and intelligence(Bartneck et al., more capable of completing tasks stereotypically associated with men,
2009). Additionally, the Robot Social Attribute Scale (RoSAS) was such as carrying items or operating machinery, while robots with
developed by comparing the Godspeed scale with social perception feminine features are more capable of completing tasks stereotypically
literature and explicitly includes warmth and competence dimensions associated with women, such as childcare or household chores. Recent
(Carpinella et al., 2017). Many studies have utilized and demonstrated research has also provided evidence. When service robots are highly
support for humanoid personality structures associated with warmth anthropomorphic, female robots will provide more pleasure in service
and competence, such as extroversion and perceived intelligence(Robert compared to male robots (Seo, 2022). However, in specific situations
et al., 2020). Finally, an increasing body of research indicates that such as healthcare, gender stereotypes may have a counterproductive
perceptions of warmth and competence are applicable not only to effect. Due to people’s subconscious concept that robots with female
physical robots but also to virtual artificial intelligence, such as virtual traits are warmer would reduce their perception of usefulness, leading to
environments(Collange & Guegan, 2020), intelligent personal assistants user anxiety (Schillaci et al., 2024). At the same time, research on
(Hu et al., 2021), and chatbots(Huang et al., 2021; Kull et al., 2021). human-robot gender has also entered the public horizon. It is reported
Previous research indicates that perceived warmth is more likely to by the researchers that gender consistency between human and robots
evoke positive attitudes from consumers towards brands, companies, or has a positive direct impact on perceived control, comfort, and brand
employees compared to perceived competence. However, research on attitude(Pitardi et al., 2022), both male and female customers are more
consumer perceptions during human-computer interactions is not as willing to accept female robot services, and male customers have higher
extensive or in-depth. satisfaction with female robot services. On the contrary, female cus­
tomers show lower satisfaction with male robot services((Seo et al.,
2024). Additionally, scholars have found that people’s expectations of

3
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

robots vary depending on their voice or gender. This implies that people Youn & Jin, 2021), which could stimulate human-computer parasocial
expect male robots to have higher intelligence ratings, and are disap­ relationship of "assistant".In the same vein, friend-based robots are used
pointed when male robots have lower intelligence. Another study found to reduce loneliness or treat depression in the elderly, which shows more
that when participants share the same gender as the robot, they have a warmth(Skjuve et al., 2021). These evidence indicate that different AI
higher acceptance of the robot and feel psychologically closer to it robots behaviors could promote different human-computer parasocial
(Eyssel et al., 2012). relationship, and accelerate corresponding consumers perception. For
So, what are the effects of gender of artificial intelligence robots on "assistant" paradigm, consumers perceive more competence than the
consumer attitudes during the service process? What are the underlying "friend" paradigm. To the opposite, consumers perceive more warmth in
mechanisms? These questions are also the focus of this study. the "friend" paradigm.
In summary, the research on parasocial relationships in the field of Hence the hypothesis.
human-computer interaction service marketing is a fusion of multiple
H1. Compared to the "friend" paradigm of human-computer parasocial
disciplines such as social psychology, media communication, and ser­
relationship, the "assistant" paradigm is more likely to enhance con­
vice marketing. Among them, the unique perceived interactivity and
sumers’ perceived competence.
openness (Labrecque, 2014) of computer media promotes the formation
of parasocial interaction, which sets consumers’ social identity and so­ H2. Compared to the "assistant" paradigm of human-computer para­
cial distance through stereotype content models(Giles & David, 2002; social relationship, the "friend" paradigm is more likely to enhance
Tyrväinen & Karjaluoto, 2025). Service providers can establish strong consumers’ perceived warmth.
parasocial relationships with consumers. In this process, the traits of
computers, such as gender, will interact with this parasocial relationship 3.2. The moderating role of gender identity
through gender stereotypes and impression management theory to in­
fluence consumer perception and behavior. Gender stereotype is an important theory for users to perceive AI
personality traits(Ahn et al., 2022). It has implications for career choice,
3. Hypotheses recruitment and selection, with females being perceived as a
high-warmth but low-competence occupational group, that people are
3.1. A relational paradigm for human-computer interaction more likely to associate male traits with competence and female traits
with warmth(He & Kang, 2019).
Donald Horton and R. Richard Wohl first introduced the concept of Some scholars have used gender stereotypes on computers(Lee &
parasocial relationship in 1956, explaining it as an "illusion of intimacy" Nass, 2005). It was found that male traits artificial agents received
between an actor and a TV viewer, unilaterally controlled by the star, higher competence scores than female traits artificial agents and female
similar to an interpersonal relationship. Based on the CASA paradigm, a traits artificial agents received higher warmth scores(Ahn et al., 2022).
parasocial relationship in human-computer interaction is a state of When studied from the perspective of voice, AI machines set to male
perceived intimate interaction with an inanimate "human being", a kind voices were found to be more persuasive actors than AI machines set to
of spiritual relationship. Specifically, when an intelligent robot is iden­ female voices, and conversely, AI machines set to female voices were
tified as a human, it will have a more favorable evaluation if it engages better companions for developing love and relationships. From a career
in higher-intensity social interactions, and vice versa, it will receive a perspective, female surgeons received higher warmth ratings and lower
negative evaluation, from which it can be inferred that the identity cues competence ratings than male surgeons(Kim et al., 2019). This all sug­
set the expectation of AI performance(Go & Sundar, 2019). With the gests that female traits emphasize warmth over competence, showing a
advancement of AI, relationships of a social and emotional nature may stronger priority on perceived warmth, and comparatively, male traits
now also be formed with artifacts-especially social chatbots, which often are more prominent on competence traits.
aim to become partners, friends, or even lovers(Skjuve et al., 2021). Impression Management Theory (IMT) suggests that Individuals in­
Some scholars have classified human-computer interaction behaviors as fluence others’ perceptions by controlling their own behavior and
either task-led or topic-led, with task-led emphasizing the priority of characteristics (Bozeman & Kacmar, 1997). Combining gender-related
completing tasks, while topic-led orientation prioritizes people’s theory, these exhibited gender traits and behaviors will reflect two
self-esteem and well-being(Haugeland et al., 2022). Dautenhahn (2007) perceptual dimensions as competence and warmth (Pino et al., 2020).
proposed that there are two relationship paradigms between humans Considering the background of human-computer parasocial relation­
and intelligent robots - the caretaker paradigm and the companion ships, consumers usually visualize interactions with characters based on
paradigm, with the caretaker paradigm emphasizes the need for robots their identification(Shan et al., 2019). Consumers are more likely to
to recognize and respond to human needs and assist with specific tasks, regard AI robots with male traits as assistants, and those with female
and the companion paradigm needs to ensure consumer satisfaction and traits as friends. Meanwhile, based on gender stereotype, the variation
happiness. Relationship norms proposed that exchange relationship, from different sociological distinctions of male traits and female traits
which focus on comparable benefits or information in return, and influence the judgement of competence and warmth(Friedmann &
communal relationship, which focus more on the concern from others Lowengart, 2019). Shaped by the conceptions of both parasocial rela­
instead of benefits, are two basic types of relationships in social in­ tionship and gender stereotype, consumers perceptions would focus
teractions.(Clark & Mils, 1993).Based on previous research, this study more on competence in the context of "assistant" paradigm and male
categorizes the parasocial relationship between AI robots and consumers traits, and focus more on warmth in the context of "friend" paradigm and
into two paradigm: "assistant" and "friend". The "assistant" paradigm female traits(Hentschel et al., 2019; Spielmann et al., 2021).
considers AI robots as assistants, designed to help people complete Hence the hypothesis.
specific tasks and activities, ensuring high quality service and produc­
H3. In "assistant" human-computer parasocial relationships, male
tion. The "friend" paradigm considers AI robots as friends, designed to
traits are more likely to increase perceived competence than female
provide care and companionship, following with emotional support and
traits.
social value.
According to the Model-Agent-Interaction-Navigation (MAIN) H4. In "friend" human-computer parasocial relationships, female traits
model, consumer perception can be related to the anthropomorphic are more likely to increase perceived warmth than male traits.
behaviors of the AI, which shapes user perception by triggering inter­
action behaviors. It has been found that consumers perceive functional
AIs more competent and have more positive attitudes(Kim et al., 2021;

4
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

3.3. The impact of stereotype on consumer attitude towards AI machines 4.1. Experiment 1

Consumer attitude refers to the emotional response of consumers to Experiment 1 designed two different service scenarios to test the
objects, attributes and interests, that is, consumers have a consistent effects of two human-computer parasocial relationship paradigms, "as­
liking or disliking response tendency to a particular commodity, brand, sistant" and "friend", on consumers’ perceived competence and
or company through learning, which is the harmonious unity of con­ perceived warmth, as well as on consumer attitudes, to validate H1, H2,
sumer emotion, cognition and behavior. Shan et al. (2019) believed that H5, and H6. In Experiment 1, we use the experimental sample to adults
cognition and emotion in consumer attitudes can predict consumer aged 18–55, who are the main target audience for consumer services and
behavioral intentions. Discussed in a narrow sense, consumer attitudes possess both purchasing power and mature cognitive abilities. We paid
include consumers’ cognitive evaluation of objective things or opinions, the "Questionnaire Star" platform to collect experimental data. Table 1
emotional reactions and corresponding behavioral tendencies. Positive illustrate the demographic characteristics of respondents of formal
consumer attitudes are those in which consumers emotionally show experiment 1.
affection for a good or service, and then behaviorally show a preference
for buying that good or service. 4.1.1. Pre-experiments
The Stimulus Biological Response (SOR) framework proposes that Pre-experiments were prioritized before the start of the formal ex­
external stimuli image a person’s response behavior by interfering with periments to verify the validity of the manipulation of two different
their cognition and emotions(Mehrabian & Russell, 1974). Accordingly, human-computer relationships: assistant and friend.
AI machines can influence consumer perceptions and thus consumer The Experiment 1 pre-experiment recruited 67 subjects who were
attitudes through human-computer interaction. According to the ste­ randomly divided into 2 groups to read the self-introductions of two
reotype model, competence and warmth are the core perceived di­ types of AI robots and watch an interaction video in a given service
mensions in the marketing domain(Halkias & Diamantopoulos, 2020), scenario. AI robots would introduce their names, occupations, and other
warmth includes being genuinely friendly helpful and enthusiastic, and information, and engage in a segment of interaction with the consumer
competence includes efficiency, skill, and confidence(Xue et al., 2020). based on the consumer’s requirements. In this case, the control of the
It was found that perceived warmth and competence moderated the human-computer parasocial relationship refers to the control conditions
relationship between service outcomes and consumer satisfaction(Gao & of Munnukka et al., 2022 and Lv et al. (2022), which are mainly based
Mattila, 2014). It has been proposed that human judgments of warmth on the changes to the robot’s appearance, task type and language style.
precede competence and that warmth carries more weight in emotional In the context where an AI robot is ready to establish an assistant rela­
and behavioral responses(Fiske et al., 2007). In studies of brand trust, tionship with a consumer, we designed a travel advisor. It will use
perceived warmth and perceived competence have positively influenced formal and concise language to answer consumer’s questions, make
purchase intentions(Cheng et al., 2022), and warmth is more important travel plans for the consumer, and perform other tasks assigned by the
than competence in stimulating and sustaining consumer attitudes to­ consumer. We chose a 100% mechanically constructed appearance for
ward a brand(Kolbl et al., 2019). Prior research has shown that its appearance and a crisp, sharp, neutral mechanical synthesized sound
perceived warmth is more likely to stimulate positive consumer atti­ for its sound. Its introduction reads "I am your travel advisor, Smith, and
tudes toward brands, companies, or employees, but there is not much I will provide you with information on popular tourist attractions and
research on AI machines. Therefore, our study base on consumer atti­ the best travel plans. I will help you understand the destination well and
tudes toward AI machines service. prepare yourself before traveling."
Hence the hypothesis. In the context where an AI robot is ready to establish a friendship
with a consumer, we designed a queuing entertainment robot. It will use
H5. In AI robot services, perceived competence can stimulate positive
spoken, friendly language to help consumers relieve boredom, fatigue,
consumer attitudes.
and other emotions, and perform a program for consumers. We chose an
H6. In AI robot services, perceived warmth can stimulate positive anthropomorphic robot appearance for its appearance, and a soft, slow,
consumer attitudes. neutral mechanical synthesized sound for its sound. It introduces itself
as "I’m Doris, your happy fruit, and I’ll make you laugh. I can sing, dance
This leads to the proposed research model, which is shown in Fig. 1.
and tell stories. I can help you relieve boredom and exhaustion. While
waiting in line, I will accept your bad temper and make you feel relaxed
4. Methodology
and happy."
After reading the self-introduction and watching the interactive
This study tests hypothesized conjectures by designing scenario ex­
video, the subjects were asked to describe the basic attributes of the AI
periments. We designed an AI robot and observed subjects’ feelings by
robot "Do you feel that the AI robot relates to you more like an assistant
controlling its appearance, function, and language. Fig. 2 shows the
or a friend?". The description of the question was measured using a 7-
experimental materials with different appearances used in the experi­
point Likert scale. The results showed significant control for human-
ment. The experimental data was analyzed using SPSS 26.0 software.
computer parasocial relationships (M assistant = 1.824, SD assistant =
0.968; M friend = 5.910, SD friend = 1.156; t = 15.705, p < 0.001).

Fig. 1. Research model.

5
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

Fig. 2. Comparison of different robot shapes.

and Cheng Chun et al., in 2021, and included the questions, "You liked
Table 1
his/her service" "You would take his/her service again in the future"
Demographic characteristics of respondents of formal
"You would recommend him/her to others" and "You would pay more for
Experiment 1(sample = 101).
his/her services" (Cronbach’s α = 0.815 > 0.8). Measures of perceived
Categories Count (%)
competence, perceived warmth, and consumer attitudes were measured
Gender using a 7-point Likert scale.At the end of the experiment, subjects were
Men 53(52) asked to respond to the pre-experiment questions on human-computer
Women 48(48)
Age (years)
relationships to control the validity of the questionnaire. At the final
18–24 16(16) stage of the experiment, all subjects were asked to fill in information
25–34 35(35) such as gender, age, occupation, education level, and income level.
35–44 31(31)
45–55 19(18)
4.1.3. Results
Income(¥, monthly)
<3000 13(13) The final number of valid questionnaires was screened 101 and the
3001–5000 20(20) data during the study was analyzed using software SPSS 26.0.
5000–8000 47(47)
>8000 21(20) 5. Reliability and validity of scales
Education
High school or lower 25(25)
Bachelor level 47(47) Before conducting data analysis, we conducted reliability and val­
Master/PhD level 29(28) idity analysis on it. The structural validity of the data meets the re­
Occupation quirements(χ2/df = 2.207 < 3, RMSEA = 0.085 < 0.1, CFI = 0.915 >
Functionary 17(17)
0.9). All path factor loads are generally greater than 0.7, AVE value
Public institutions 27(26)
Enterprise employees 30(30) greater than 0.5, Combination reliability greater than 0.8, showing a
Student 13(13) good convergent validity. The discriminant validity of each variable is
Free profession 14(14) significant. Appendix 2 present the results of reliability and validity
analysis.

4.1.2. Formal experiment

2 Manipulation test
Experiment 1 recruited 101 subjects at the university, using a situ­
ational simulation experiment in which all subjects were randomized
In terms of human-computer parasocial relationships, we adapt in­
into 2 groups. The formal experiment followed the experimental mate­
dependent sample T-test. M assistant = 1.612, SD assistant = 0.909, M friend
rials and descriptions that had been validated in the pre-experiment.
= 5.904, SD friend = 0.975; t = − 22.843, p < 0.001. Thus, the manipu­
After reading the self-introduction and watching the interactive video,
lation of the two human-computer parasocial relationship paradigms
subjects were asked to answer several questions adapted to their own
was successful. In subsequent studies, we divide parasocial relationship
feelings based on the scales of perceived competence, perceived warmth,
into grouped variables by adding and subtracting standard deviations
and consumer attitudes. The measures of perceived competence and
from the mean.
perceived warmth were drawn from the scale developed by Fiske et al.
(2007), with perceived competence including the question "You felt that
3 consumer perception
he/she was competent OR confident OR independent OR competitive
OR intelligent during the service you received." 5 items (Cronbach’s α =
Based on the research hypotheses, one-way ANOVA was used to test
0.932 > 0.8), perceived warmth included "You felt that he/she was
the difference in the impact of human-computer parasocial relationships
tolerant/warm/good-tempered/sincere when you received the service."
on consumers’ perceived competence and perceived warmth. The results
4 question items (Cronbach’s α = 0.876 > 0.8). The measure of con­
showed that consumers’ perceived competence was significantly
sumer attitudes was borrowed from the scales of Zeithaml et al. (1985)

6
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

enhanced in the assistant-type human-computer parasocial relationship Table 2

group (M assistant = 6.412, SD = 0.993; M friend = 5.135, SD = 0.964; F = Linear regression analysis of consumer attitude.
43.036, p < 0.001); and consumers’ perceived warmth in the friend-type consumer attitude
human-computer parasocial relationship group was significantly
B t p LLCI ULCI
enhanced (M assistant = 5.480, SD = 1.236; M friend = 6.428, SD = 0.708;
F = 22.710, p < 0.001), see Fig. 3. This result supports H1 and H2, i.e., Perceived competence 0.236 3.197 0.002 0.089 0.382
Perceived warmth 0.291 3.803 0.000 0.139 0.443
compared with the "friend" type of human-computer parasocial rela­
tionship, "assistant" type of human-computer parasocial relationship can
enhance consumers’ perceived warmth more than "friend" type of human-computer parasocial relationship on consumer attitudes. Using
human-computer parasocial relationship. This result supports H1 and the Bootstrap method and the PROCESS 4.1 plugin to test the mediating
H2, i.e., compared with "friend" type human-computer parasocial rela­ effect. In the analysis, Model 4 was selected with a sample size of 5000
tionship, "assistant" type human-computer parasocial relationship can and a confidence interval of 95%. Mediate effect analysis was conducted
enhance consumers’ perceived warmth (H1); compared with "assistant" with consumer attitude as the dependent variable, human-computer
type human-computer parasocial relationship, "friend" type human- parasocial relationship as the independent variable, and perceived
computer parasocial relationship can enhance consumers’ perceived competence as the mediating variable. The results of data analysis
warmth (H2). showed that the mediating effect of perceived competence was signifi­
cant (effect = − 0.411, 95% CI, LLCI = − 0.860, ULCI = − 0.121). The
4 Consumer attitude mediation effect analysis was conducted with consumer attitude as the
dependent variable, human-computer parasocial relationship as the in­
Linear regression was used to test the effects of perceived compe­ dependent variable, and perceived warmth as the mediating variable.
tence and perceived warmth on consumer attitudes. In the analysis The results of data analysis showed that the mediating effect of
process, we regarded perceived competence and perceived warmth as perceived warmth was significant (effect = 0.350, 95% CI, LLCI = 0.144,
independent variables, and consumer attitude as the dependent variable ULCI = 0.630). From the experimental results, it can be seen that the
for linear regression. We observed the significance and slope of the single variable of human-computer parasocial relationship has no sig­
analysis results. The results showed that there was a linear correlation nificant impact on consumer attitudes, indicating that human-computer
between perceived competence(R2 > 0, F = 10.221, p < 0.01), perceived parasocial relationship affects consumer attitudes by influencing their
warmth(R2 > 0, F = 14.463, p < 0.001) and consumer attitudes, and perceived competence and perceived warmth. The results are shown in
perceived competence was positively correlated with consumer atti­ Tables 3 and 4. Similarly, control variables such as gender and age of the
tudes (unstandardized coefficient B = 0.236 > 0, t = 3.197), i.e., the subjects have no significant impact in the path.
higher the perceived competence, the more positive the consumer atti­ The above data results support H1, H2, H5, and H6, while validating
tudes. Perceived warmth is positively correlated with consumer attitude the mediating effect of perceived competence and perceived warmth.
(unstandardized coefficient B = 0.291 > 0, t = 3.803), i.e., the higher Then how to further enhance consumer perception and in this way
the perceived warmth the more positive the consumer attitude. Mean­ further promote positive consumer attitudes? In Experiment 2 we
while, control variables such as gender and age of the subjects have no introduced the moderating variable of AI robot gender traits for in-depth
significant impact on consumer attitudes. The results are shown in study, in order to investigate the intrinsic mechanism of human-
Table 2. This result supports H5 and H6, in which perceived competence computer parasocial relationship affecting consumer attitudes.
can stimulate positive consumer attitudes in AI robot services (H5); and
perceived warmth can stimulate positive consumer attitudes in AI robot 5.1. Experiment 2
services (H6).
Experiment 2 is a 2*2 (assistant/friend*male traits/female traits)
5 Mediation effects test controlled experiment, aiming to test the effects of different kinds of AI
robots on consumers’ perceived competence and perceived warmth, and
Based on the hypotheses and model derivation, this study argues that to prove their effects on consumer attitudes, and to validate H3 and H4.
perceived competence and perceived warmth mediate the effect of In Experiment 2, we also recruited sample to adults aged 18–55 at the

Fig. 3. Influence of human-machine parasocial relationships on consumer perceptions.

7
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

Table 3 two different human-computer parasocial relationship paradigms.

Mediating effects model test for perceived competence. The Experiment 2 pre-experiment recruited 127 subjects who were
Parasocial Perceived R2 F randomly divided into four groups to read the self-introductions of each
relationship competence of the four types of AI robots and to watch videos of the interactions in
Parasocial coeff − 1.278 ​ 0.303 43.036 specific service scenarios. The AI robots introduced their names, occu­
relationship → se 0.195 ​ pations, and other information, and interacted with consumers accord­
Perceived t − 6.560 ​ ing to their needs. The control of human-computer relationship followed
competence p 0.000 ​ the tested experimental materials and descriptions in Experiment I. The
Parasocial coeff − 0.049 0.001 0.074
control of the gender traits of the AI robot is to change the appearance
​
relationship → se 0.180 ​
Consumer t − 0.272 ​ and voice, the "male traits" AI robot adopts the real male appearance and
attitudes p 0.786 ​ timbre; the "female traits" AI robot adopts the real female appearance
Parasocial coeff 0.362 0.322 0.122 6.817 and timbre. The experiment was a 2X2 experiment with four controls:
relationship → se 0.203 0.087
assistant, friend, and male traits and female traits, and four experimental
Perceived t 1.786 3.681
competence → p 0.077 0.000 scenarios were designed. After reading the self-introduction and
Consumer watching the interactive video, subjects were asked to describe the
attitudes gender traits of the AI robot in addition to the interpersonal relationship
"Do you think the AI robot is more masculine or feminine in nature?" The
results showed that the control for human-computer parasocial re­
Table 4 lationships was significant (M assistant = 1.885, SD assistant = 0.777, M
Mediating effect model test for perceived warmth. friend = 5.197, SD friend = 1.099; t = − 19.730, p < 0.001); the control for

Parasocial Perceived R2 F the gender traits of the AI robot was significant (M male traits = 2.623, SD
relationship warmth male traits = 1.213; M female traits = 5.576, SD female traits = 1.177; t =

Parasocial coeff 0.948 ​ 0.187 22.710

− 13.915, p < 0.001).
relationship → se 0.199 ​
Perceived warmth t 4.766 ​ 5.1.2. Formal experiment
p 0.000 ​ Experiment 2 recruited 137 subjects and used the situational simu­
Parasocial coeff 0.049 0.001 0.074
− ​
lation experiment method, all subjects were randomly divided into 4
relationship → se 0.180 ​
Consumer attitudes t − 0.272 ​ groups for a 2*2 (assistant/friend*male/female) controlled experiment.
p 0.786 ​ The formal experiments all used the experimental materials and de­
Parasocial coeff − 0.399 0.369 0.168 9.894 scriptions determined in the pre-experiment, and the problem settings
relationship → se 0.183 0.083 were the same as those in the formal experiments in Experiment I. The
Perceived warmth t 2.185 4.438
purpose was to verify the effects of different kinds of AI robots on con­
−
→ Consumer p 0.031 0.000
attitudes sumers’ perceived competence, perceived warmth, and to demonstrate
their effects on consumers’ attitudes at the same time.

"Questionnaire Star" platform. Table 5 illustrate the demographic 5.1.3. Results

characteristics of respondents of formal Experiment 2. The final number of valid questionnaires was screened to be 137, and
the data from the research process were analyzed using software SPSS
5.1.1. Pre-experiments 26.0.
Pre-experiments were prioritized before the start of the formal
experiment to validate the effectiveness of gender traits manipulation on 1 Reliability and validity of scales

Table 5 Before conducting data analysis, we conducted reliability and val­

Demographic characteristics of respondents of formal idity analysis on it. The structural validity of the data meets the re­
Experiment 2(sample = 137). quirements(χ2/df = 2.004, RMSEA = 0.086, CFI = 0.950). All path
Categories Count (%)
factor loads are generally greater than 0.8, AVE value greater than 0.6,
Combination reliability greater than 0.8, showing a good convergent
Gender
validity. The discriminant validity of each variable is significant. The
Men 73(53)
Women 64(47) results of reliability and validity analysis are also shown in Appendix 2.
Age (years)
18–24 23(17) 2 Manipulation test
25–34 47(34)
35–44 39(29)
45–55 28(20)
In terms of human-computer parasocial relationships, we adapt in­
Income(¥, monthly) dependent sample t-test. M assistant = 1.721, SD assistant = 1.144; M friend
<3000 21(15) = 5.594, SD friend = 1.075; t = − 20.423, p < 0.001; and in terms of AI
3001–5000 29(21) robot gender, M male traits = 1.582, SD male traits = 0.940, M female traits =
5000–8000 63(46)
6.400, SD female traits = 0.769; t = − 32.902. p < 0.001. Thus, the
>8000 24(18)
Education manipulation of the two human-computer parasocial relationships, as­
High school or lower 33(24) sistant and friend, was successful, and the manipulation of the two AI
Bachelor level 61(45) robot genders, male and female, was successful.
Master/PhD level 43(31)
Occupation
Functionary 24(17)
3 Consumer perception
Public institutions 35(26)
Enterprise employees 43(31) Based on the research hypotheses, a two-way ANOVA method was
Student 19(14) used to test the effect of human-computer parasocial relationship & AI
Free profession 16(12)
robot gender traits on consumers’ perceived competence and perceived

8
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

warmth. computer parasocial relationship as the independent variable, AI robot

A two-way ANOVA was conducted with perceived competence as the gender traits as the moderating variable, and perceived competence as
dependent variable and human-machine parasocial relationship and AI the mediator variable. The results of the data analysis showed that the
robot gender traits as the independent variables. The results showed that results of the mediated effects analysis with moderation using perceived
the interaction effect of human-computer parasocial relationship and AI competence as the mediating variable were significant (effect = 0.028,
robot gender traits was significant (F = 4.676, p < 0.05). Meanwhile, the 95% CI, p < 0.05, LLCI = 0.070, ULCI = 1.573), as shown in Table 6.
main effect of human-computer parasocial relationship was significant Specifically, when the gender traits of the AI robot was "male traits", the
(F = 22.688, p < 0.001); the main effect of AI robot gender traits was interaction between the human-computer parasocial relationship and
significant (F = 6.158, p < 0.05). For the "assistant" parasocial rela­ the gender traits of the AI robot had a significant indirect effect on
tionship, "male traits" AI robots are more likely to stimulate consumers’ consumer attitudes through perceived competence (effect = − 1.316,
perceived competence than "female traits" AI robots (M male traits = 95% CI, p < 0.001, LLCI = − 1.853, ULCI = − 0.779). In Fig. 6, we can
6.777, SD male traits = 0.871, M female traits = 5.894, SD female traits = 0.965; also observe that the slope of the moderation effect curve representing
t = 3.958, p < 0.001). For the "friend" type of parasocial relationship, male traits is much steeper than that of the female traits. This further
there was no difference in consumers’ perceived competence of the indicates that male traits have a greater impact on perceived compe­
"male traits" AI robot and the "female traits" AI robot (M male traits = tence, and that assistant-type human-computer parasocial relationship
5.461, SD male traits = 1.411, M female traits = 5.400, SD female traits = 1.131; with male traits lead to a stronger perceived competence for consumers.
t = 0.198, p > 0.05). The results are shown in Fig. 4. A moderated mediated effect analysis was conducted with consumer
A two-way ANOVA was conducted with perceived warmth as the attitude as the dependent variable, human-computer parasocial rela­
dependent variable and human-computer parasocial relationship and AI tionship as the independent variable, AI robot gender traits as the
robot gender traits as the independent variables. The results showed that moderating variable, and perceived warmth as the mediator variable.
the interaction effect of human-computer parasocial relationship and AI The results of the data analysis showed that the results of the mediated
robot gender traits was significant (F = 9.741, p < 0.01). Meanwhile, the effects analysis with moderation using perceived warmth as the medi­
main effect of human-computer parasocial relationship was significant ating variable were significant (effect = 0.058, 95% CI, p < 0.01, LLCI =
(F = 20.059, p < 0.001), the main effect of AI robot gender traits was 0.430, ULCI = 1.920), as shown in Table 7. Specifically, when the gender
significant (F = 4.599, p < 0.05). For the "friend" parasocial relationship, traits of the AI robot was "female traits", the interaction of human-
"female traits" AI robots are more likely to stimulate consumers’ computer parasocial relationship with the gender traits of the AI robot
perceived warmth than "male traits" AI robots(M male traits = 5.689, SD had a significant indirect effect on consumer attitudes through perceived
male traits = 1.470, M female traits = 6.681, SD female traits = 0.550; t = warmth (effect = 1.431, 95% CI, p < 0.001, LLCI = 0.910, ULCI =
− 3.647, p < 0.01).For the "assistant" type of parasocial relationship, 1.951). In Fig. 7, we can also observe that the slope of the moderation
there was no difference in consumers’ perceived warmth towards the effect curve representing female traits is much steeper than that of the
"male traits" AI robot and the "female traits" AI robot (M male traits = male traits. This further indicates that female traits have a greater
5.434, SD male traits = 1.350, M female traits = 5.250, SD female traits = 0.805; impact on perceived warmth, and that friend-type human-computer
t = 0.682, p > 0.05). The results are shown in Fig. 5. parasocial relationship with female traits lead to a stronger perceived
warmth for consumers.
4 Mediation and Moderation Effect Tests The experimental results demonstrate that the interaction between
human-computer parasocial relationships and gender traits affects
Based on the hypotheses and model derivation, this study argues that consumer perceived competence and perceived warmth. Specifically,
perceived competence and perceived warmth mediate the effect of the when gender traits change, the impact of human-computer parasocial
interaction between human-computer parasocial relationship and AI relationship on consumer perception also changes. In other words,
robot gender traits on consumer attitudes. Using the Bootstrap method gender traits moderate the impact of human-computer parasocial rela­
and the PROCESS 4.1 plugin to test mediators and effects. In the anal­ tionship on consumer perception. Meanwhile, the mediating role of
ysis, Model 7 was selected with a sample size of 5000 and a confidence consumer perceived competence and perceived warmth still exists and is
interval of 95%. The mediated effect analysis with moderation was significant under this path. In Fig. 6, we can also observe that the slope
conducted with consumer attitude as the dependent variable, human- of the moderation effect curve representing male traits is much steeper

Fig. 4. Influence of interaction factors on consumer perceptions.

9
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

Fig. 5. Effect of interaction factors on consumer warmth.

Table 6
Mediation model test with moderation (mediated by perceived competence).
Parasocial Gender Parasocial relationship & Perceived R2 F
relationship traits Gender traits competence

Parasocial relationship & Gender traits → Perceived coeff − 0.896 − 0.469 0.822 ​ 0.201 11.181
competence se 0.190 0.190 0.380 ​
t − 4.717 − 2.466 2.163 ​
p 0.000 0.015 0.032 ​
Parasocial relationship → Perceived competence → coeff 0.050 ​ ​ 0.324 0.157 12.513
Consumer attitudes se 0.167 ​ ​ 0.068
t 0.302 ​ ​ 4.751
p 0.764 ​ ​ 0.000

impact in the path. In summary, H1, H2, H3, H4, H5, H6 have all been
validated through data analysis. The summary of model estimation is in
Appendix 3.

6. Discussion

First, the Stereotype Content Model (SCM) is examined in a human-

computer interaction scenario. In order to be more in line with the social
situation of SCM, we set up the experiment under the premise of human-
computer parasocial relationship. The experiments show that "assistant"
human-computer parasocial relationship significantly enhances con­
sumers’ perceived competence, and "friend" human-computer parasocial
relationship significantly enhances consumers’ perceived warmth,
which consists with the bi-dimensional structure and the mixed hy­
pothesis of SCM. We found that consumers could perceive the compat­
ible identity, when a role clue was set to an AI service robot(Cheng et al.,
2022). Thus individual conception could be guided by the specific par­
asocial interaction. The proof is through the distinct design of AI robots,
Fig. 6. Moderation effect on human-computer parasocial relationship and consumers consider themselves developing a certain parasocial rela­
perceived competence. tionship with AI robots, and they have experienced the competence and
warmth of different levels in deed. The fact also prove that it is feasible
than that of the female traits. This further indicates that male traits have to develop a parasocial relationship between human and AI robots.
a greater impact on perceived competence, and that assistant-type Enterprises can also achieve the goal of establishing parasocial re­
human-computer parasocial relationship with male traits lead to a lationships with consumers by controlling the design of AI robots to
stronger perceived competence for consumers. evolve consumer perception and achieve service goals in given situa­
The above data results support H3 and H4, male traits are more likely tions. In this context, the study complement SCM from the perspective of
to enhance perceived competence (H3) compared to female traits in parasocial relationship, and validate it in the context of
assistant-based AI robot services, and female traits are more likely to human-computer interactions. Meanwhile, the study promote CASA
enhance perceived warmth (H4) compared to male traits in friend-based with the structure of parasocial relationship paradigm in
AI robot services, which in turn enhances consumer attitudes. Control human-computer interactions.
variables such as gender and age of the subjects have no significant Second, the moderating role of artificial robot gender traits was only

10
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

Table 7
Mediation model test with moderation (mediated by perceived warmth).
Parasocial Gender Parasocial relationship & Gender Perceived R2 F
relationship traits traits warmth

Parasocial relationship & Gender traits → Perceived coeff 0.856 0.408 1.175 ​ 0.211 11.848
warmth se 0.188 0.188 0.377 ​
t 4548 2.167 3.121 ​
p 0.000 0.032 0.002 ​
Parasocial relationship → Perceived warmth → coeff − 0.499 ​ ​ 0.296 0.135 10.470
Consumer attitudes se 0.168 ​ ​ 0.069
t − 2.971 ​ ​ 4.307
p 0.004 ​ ​ 0.000

while also amplifying their perception of competence. When consumers

are more concerned about hedonistic needs, they prefer to receive care
and pleasure. Therefore, they prefer to have a friend with high warmth
scores, while also amplifying their perception of warmth. This is also
consistent with the perspective of relationship norm theory. In sum­
mary, it is crucial to predict consumer needs and establish matching
service relationships to enhance consumer perception. This is in line
with the Mode-Agent-Interaction-Navigation (MAIN) model. Bastiansen
et al., 2022 did not find any significant impact of warmth, gender traits
of chat robots, or their interactions on trust, helpfulness, and compe­
tence in research, largely due to the lack of establishing parasocial re­
lationships in human-computer interactions.
Finally, the impact mechanism of artificial intelligence robots on
consumer attitudes is further clarified. Through the construction of
theoretical models and empirical testing, we have verified the dual
mediating effect of consumer perceived competence and perceived
warmth. We have demonstrated that the two human-computer para­
social relationships of "assistant" and "friend", as well as their interaction
Fig. 7. Moderation effect on human-computer parasocial relationship and with the gender traits of "male traits" and "female traits" AI robots, have a
perceived warmth. significant impact on consumer perceived competence and perceived
warmth. We have also tested the positive correlation between consumer
partially tested. In the process of human-computer parasocial relation­ perceived competence, perceived warmth, and consumer attitudes. A
ships and gender traits interaction of AI robots. For the "assistant" par­ logically coherent, theoretically reliable, and rigorously argued
asocial relationship, there was no difference in consumers’ perceived moderated mediation model was constructed to answer the impact
warmth between the "male traits" AI robot and the "female traits" AI mechanism of AI robots on consumer attitudes from a social perspective.
robot. For the "friend" parasocial relationship, there was no difference in
consumers’ perceived competence between the "male traits" and "female 7. Conclusion
traits" AI robots. This suggests that the moderating effect of the gender
traits of AI robots is largely influenced by the parasocial relationship 7.1. Theoretical contributions
between humans and robots. This phenomenon is explained by the fact
that in the "assistant" type of parasocial relationship, consumers tend to This study provides a new perspective for service marketing research
perceive the competence of AI robots more, and weaken the warmth of in the context of human-computer interaction. First, it proposes two
AI robots, so the moderating effect of "male traits" is significant. In the parasocial relationships, "assistant" and "friend", as independent vari­
"friend" parasocial relationship, consumers tend to perceive the AI ro­ ables, which broadens the research direction of AI robots in the field of
bot’s warmth more than its competence, so the moderating effect of service marketing. Previous studies have mainly focused on anthropo­
"female traits" is significant. As proposed by the SCM, people’s perceived morphism as the main antecedent variable based on the design factor,
competence and warmth in social relationships may be of the same level task factor, and interaction factor(Davenport et al., 2019; Wirtz et al.,
as high or low. Although previous studies have suggested that gender 2018). In addition to studying the basic attributes of AI robots, this study
stereotype contributes to the establishment of parasocial relationships explores the establishment of parasocial relationships between AI robots
and strengthens consumer perception, experimental results show that and consumers in the process of service marketing, and proves the
gender stereotype do not fully predict people’s perception of compe­ feasibility of the two different kinds of parasocial relationships as the
tence and warmth, while the theory of relationship norms has been independent variables through experiments, which is useful for the
further validated. This indicates that consumers are more concerned service marketing context. feasibility, complementing the related
about their individual requirements in real services, which also leads to research on human-computer interaction in the context of service mar­
expectations for service forms and indirectly affects the establishment of keting. Second, the different paths of the two human-computer para­
parasocial relationships(Cheng et al., 2022). Therefore, the influence of social relationships affecting consumer attitudes are verified, enriching
relationship norms theory takes precedence over the influence of gender the research on consumer attitudes in the context of human-computer
stereotype. To some extent, gender stereotype seems to conflict the interaction. In terms of mediating variables, previous studies have
theory of relationship norms. From the perspective of consumer mainly focused on technology acceptance model and social judgment
behavior, they are more concerned with matching needs with services (Davis, 1989; Gefen et al., 2003). Based on the stereotype model, we
(Martins et al., 2011). For example, when consumers are more con­ choose two social dimension variables, perceived competence and
cerned about utilitarian needs, they prefer to receive specific advice and perceived warmth, to form a unique "dual-path" mediation model. At the
solutions. Therefore, they prefer to have a highly competent assistant, same time, this study further clarifies the research content of consumer

11
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

attitudes in the context of human-computer interaction on the basis of type, applying the appearance and voice of real men, and using language
existing research, and provides solutions in terms of influencing mech­ that leans towards formality and conciseness. If we want to mainly
anisms and scale measurements. Third, it clarifies the influence of stimulate consumers’ perceived warmth, we should design artificial
gender traits attributes of AI robots on consumer attitudes as moderating intelligence robots with "female" characteristics to establish "friend" type
variables, which triggers new thinking about stereotype modeling under parasocial relationships. In practical operations, priority can be given to
the interaction factor. In terms of moderating variables, existing studies choosing companionship or sales as the target task type, using the
basically conduct research from three aspects: technical variables, in­ appearance and voice of real women, and language tends to be friendly
dividual variables and interaction variables(Broadbent et al., 2009; and colloquial.
Hudson et al., 2017; Nomurat et al., 2009; Petersen et al., 2017; Pinx­ Finally, it is necessary to fully consider the differences in cultural
teren et al., 2019). All of these variables are consumer differences, and in backgrounds and technological platforms. The cultural dimension the­
this study, the gender traits of AI robots is taken as a moderating vari­ ory (Hofstede, 1986) provides a suitable framework for this. For
able, and "male traits" and "female traits" gender characters are utilized example, in individualistic cultures, such as North American culture,
innovatively. This variable has the same origin and flow as the mediator people place greater emphasis on personal achievement and indepen­
variable in terms of its pathway of action, and both are based on the dence. In high power distance cultures, such as South Asian or Middle
principle of stereotype modeling, which is highly consistent with the Eastern countries, social relationships are strictly hierarchical. In cul­
research theory, which provides a brand-new idea and perspective for tures accustomed to avoiding uncertainty, social relationships are more
the subsequent research in related fields. formal and traditional. And these cultures tend to establish “assistant”
type parasocial relationships. In collectivist cultures, such as East Asian
7.2. Management insights cultures, people usually emphasize harmony and interdependence
within the group. In cultures with power distance, such as Nordic
This study provides a multi-faceted guidance for the use of AI robots countries, relationships between people are more equal. In cultures that
in the process of service marketing, which has certain practical signifi­ do not avoid uncertainty, social relationships are more flexible and tend
cance for promoting the quality service of artificial intelligence robots to to establish “friend” type parasocial relationships (Pereira et al., 2023).
improve consumer attitudes. Meanwhile, the interactive forms and functional designs of different
First, it is conducive to clarifying the practice scenarios of AI robots platforms also have an impact. In official enterprise websites and
in service marketing. Enterprises can categorize the nature of services e-commerce platforms, users are more inclined towards utilitarian psy­
based on two parasocial relationships, and select different AI robots to chology and tend to establish “assistant” type parasocial relationships.
strengthen the control of the service process and service quality through In visual and social platforms, users are more inclined towards hedonic
the demand for different service scenarios. For environments that psychology and tend to establish “friend” type parasocial relationships
require consumers to feel more capable, AI robots with "male traits" (Liu et al., 2024).
should be selected to stimulate consumer perceived competence by
establishing "assistant" type parasocial relationship, such as hospital 7.3. Research limitations and prospects
suggestion robots, educational guidance robots, legal consultant robots,
etc. For environments that require consumers to feel more warmth, AI There are some shortcomings in this study. The mediator variables of
robots with "female traits" should be selected to stimulate consumer this study on consumer perceived competence and perceived warmth
perceived warmth by establishing "friend" type parasocial relationship, are consumer perceptions on social latitude, and whether they are
such as shopping guide robots, home companion robots, and restaurant influenced by factors such as social fear can be further explored. Second,
reception robots. for the selection and categorization of parasocial relationships, we refer
Second, it is conducive to the continuous improvement of the to the research of Dautenhahn and other scholars, however, parasocial
development direction of AI robots in the field of service marketing. The relationships as a kind of virtual, instantaneous, unidirectional social
application of AI robots in the field of service marketing has to some relationships(Horton & Wohl, 1956), whether it will form "lovers",
extent weakened the four major service characteristics of intangibility, "relatives" or even "rivals", and what is the impact of human-computer
heterogeneity, inseparability, and perishability. At present, people still parasocial relationships on consumer attitudes in this case? What is
tend to use it as an auxiliary tool. By using artificial intelligence robots to their mechanism of action? The above issues can be further explored in
establish parasocial relationships with consumers, especially when future research. Finally, in terms of experimental settings, we chose two
establishing "friend" type parasocial relationships, they can greatly typical scenarios: tourism consultants and queues, which can basically
replace employees in the marketing process, thereby reducing cost in­ represent consumers’ hedonic and utilitarian needs. However, there are
vestment for internal employees to a certain extent. Focusing on the various service scenarios in reality, and the demands for services are also
service itself, due to the particularity of artificial intelligence robot different. Research could continue to explore the different impacts of
services, this "friend" type of parasocial relationship not only enhances human-computer interaction in various service contexts.
the service commitment before service, but also retains the service at­
mosphere and customer perception during the service process to a CRediT authorship contribution statement
certain extent, stimulates consumer attitudes after service, and reduces
waiting time, which is very suitable for the service marketing model of Tiejun Qi: Writing – review & editing, Writing – original draft,
the "fast-moving consumer" era. Innovated the functional positioning of Visualization, Methodology, Funding acquisition, Formal analysis, Data
artificial intelligence robots in the service process, promoting their curation, Conceptualization. Hongshen Liu: Writing – review & editing,
transformation from tools to employees. Supervision, Methodology, Conceptualization. Zhihui Huang: Writing –
Third, it is conducive to enriching the design concept of service- review & editing, Methodology, Funding acquisition, Data curation,
oriented AI robots. According to the control and processing of the Conceptualization.
experiment, controlling the robot’s appearance, sound, task type, and
language style can complete the manipulation of interpersonal quasi Funding
social relationships and artificial intelligence robot gender. If we want to
mainly stimulate consumers’ perceived competence, we should design Tiejun QI acknowledge the support from the Hunan Province Grad­
artificial intelligence robots with "male" characteristics to establish uate Research Innovation Project (QL20230219). Hongshen Liu
"assistant" type parasocial relationships. In practical operations, priority acknowledge the support from the Base Project of Philosophy and Social
can be given to selecting suggestion or consultant tasks as the target task Science Fund in Hunan Province (24JD014). Zhihui Huang acknowledge

12
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

the support from the National Natural Science Foundation of China Research Innovation Project (QL20230219). Hongshen Liu acknowledge
(72302025), Natural Science Foundation of Hunan Province the support from the Base Project of Philosophy and Social Science Fund
(2023JJ40057), Changsha Natural Science Foundation (kq2202216), in Hunan Province (24JD014). Zhihui Huang acknowledge the support
Hunan Provincial Department of Education Youth Project (21B0297). from the National Natural Science Foundation of China (72302025),
Hunan Provincial Natural Science Foundation (2023JJ40057), Chang­
Declaration of competing interest sha Natural Science Foundation (kq2202216), Hunan Provincial
Department of Education Youth Project (21B0297). If there are other
The authors declare the following financial interests/personal re­ authors, they declare that they have no known competing financial in­
lationships which may be considered as potential competing interests: terests or personal relationships that could have appeared to influence
Tiejun QI acknowledge the support from the Hunan Province Graduate the work reported in this paper.

Appendix 1. Items used in construct measurement (7-point scale)

Construct Scale

Parasocial relationship type I feel that the AI robot relates to me more like an assistant or a friend? Youn and Jin (2021)
Perceived competence I felt that he/she was competent during the service I received. Fiske et al. (2007)
I felt that he/she was confident during the service I received. ​
I felt that he/she was independent during the service I received. ​
I felt that he/she was competitive during the service I received. ​
I felt that he/she was intelligent during the service I received. ​
Perceived warmth I felt that he/she was tolerant when I received the service. Fiske et al. (2007)
I felt that he/she was warm when I received the service. ​
I felt that he/she was good-tempered when I received the service. ​
I felt that he/she was sincere when I received the service. ​
Consumer attitudes I liked his/her service Zeithaml et al. (1985)
I would take his/her service again in the future ​
I would recommend him/her to others ​
I would pay more for his/her services ​
Gender traits I think the AI robot is more masculine or feminine in nature ​

Appendix 2. Reliability and validity analysis results

Statistics of latent variables used in Experiment

Construct Correlation

Mean SD AVE CR 1 2 3

Experiment 1 1.Perceived competence 5.754 1.166 0.740 0.934 ​ ​ ​

2.Perceived warmth 5.968 1.103 0.649 0.880 0.356** ​ ​
3.Consumer attitudes 6.322 0.898 0.585 0.847 0.399*** 0.38*** ​
AVE square root ​ ​ ​ ​ 0.860 0.805 0.765
​ ​ ​ ​ ​ Correlation ​ ​
​ ​ ​ ​ ​ 1 2 3
Experiment 2 1.Perceived competence 5.879 1.230 0.670 0.910 ​ ​ ​
2.Perceived warmth 5.777 1.226 0.720 0.911 0.401*** ​ ​
3.Consumer attitudes 6.396 0.982 0.716 0.910 0.565*** 0.368*** ​
AVE square root ​ ​ ​ ​ 0.818 0.849 0.846
Notes: **p < 0.05, ***p < 0.01.

Measurement items used for latent variables

Measurement items Factor loadings

Experiment 1 Experiment 2

Perceived competence ​ ​
I felt that he/she was competent during the service I received. 0.887 0.799
I felt that he/she was confident during the service I received. 0.846 0.847
I felt that he/she was independent during the service I received. 0.800 0.803
I felt that he/she was competitive during the service I received. 0.855 0.811
I felt that he/she was intelligent during the service I received. 0.909 0.831
Perceived warmth ​ ​
I felt that he/she was tolerant when I received the service. 0.724 0.835
I felt that he/she was warm when I received the service. 0.871 0.858
I felt that he/she was good-tempered when I received the service. 0.818 0.864
I felt that he/she was sincere when I received the service. 0.802 0.837
Consumer attitudes ​ ​
I liked his/her service 0.760 0.801
I would take his/her service again in the future 0.856 0.817
(continued on next page)

13
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

(continued )
Measurement items Factor loadings

Experiment 1 Experiment 2

I would recommend him/her to others 0.587 0.880

I would pay more for his/her services 0.828 0.884

Appendix 3. Summary of model estimation

Estimated paths β Hypothesis LLCI ULCI Supported

Direct effects
human-computer parasocial relationship → perceived competence − 1.311*** H1 − 1.693 − 0.930 ✓
human-computer parasocial relationship → perceived warmth 0.912*** H2 0.517 1.306 ✓
Perceived competence → consumer attitudes 0. 236*** H5 0.089 0.382 ✓
Perceived warmth → consumer attitudes 0. 291*** H6 0.139 0.443 ✓
Mediate effects
human-computer parasocial relationship → perceived competence → consumer attitudes − 0.411*** ​ − 0.860 − 0.121 ​
human-computer parasocial relationship → perceived warmth → consumer attitudes 0.350*** ​ 0.144 0.630 ​
Moderate effects
human-computer parasocial relationship & gender traits → perceived competence 0.028** H3 0.070 1.573 ✓
human-computer parasocial relationship & gender traits → perceived warmth 0.058*** H4 0.430 1.920 ✓
Notes: β = standardised coefficient; **p < 0.05, ***p < 0.01.

Data availability Collange, J., & Guegan, J. (2020). Using virtual reality to induce gratitude through
virtual social interaction. Computers in Human Behavior, 113, Article 106473.
Daniel, E. S., Crawford Jackson, E. C., & Westerman, D. K. (2018). The influence of social
Data will be made available on request. media influencers: Understanding online vaping communities and parasocial
interaction through the lens of taylor’s six-segment strategy wheel. Journal of
Interactive Advertising, 18, 109–196.
References Dautenhahn, K. (2007). Socially intelligent robots: Dimensions of human–robot
interaction. Philosophical Transactions of the Royal Society B: Biological Sciences, 362,
Ahn, J. Y., Kim, J., & Sung, Y. (2022). The effect of gender stereotypes on artificial 679–704.
intelligence recommendations. Journal of Business Research, 114. Davenport, T. H., Guha, A., Grewal, D., & Breßgott, T. (2019). How artificial intelligence
Alesich, S., & Rigby, M. (2017). Gendered robots: Implications for our humanoid future. will change the future of marketing. Journal of the Academy of Marketing Science, 48,
IEEE Technology and Society Magazine, 36, 50–59. 24–42.
Ashton-James, C. E., Tybur, J. M., Grießer, V., & Costa, D. (2019). Stereotypes about Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of
surgeon warmth and competence: The role of surgeon gender. PLoS One, 14(2), information technology. MIS Quarterly, 13, 319–340.
Article e0211890. DeArmond, S., Tye, M. G., Chen, P. Y., Krauss, A. D., Rogers, D. A., & Sintek, E. (2006).
Baker, A., & Florack, A. (2021). Uncovering men’s stereotype content (warmth and Age and gender stereotypes: New challenges in a changing workplace and workforce.
competence) associated with a representative range of male body size categories. Journal of Applied Social Psychology, 36, 2184–2214.
Body Image, 37, 148–161. Eagly, A. H. (1987). Sex differences in social behavior : A social-role interpretation.
Balakrishnan, J., Abed, S. S., & Jones, P. (2022). The role of meta-utaut factors, perceived Eyssel, F., & Hegel, F. (2012). (S)he’s got the look: Gender stereotyping of Robots1.
anthropomorphism, perceived intelligence, and social self-efficacy in chatbot-based Journal of Applied Social Psychology, 42, 2213–2230.
services? Technological Forecasting and Social Change, 180. Eyssel, F., Kuchenbrandt, D., Bobinger, S., Ruiter, L. E., & Hegel, F. (2012). ‘If you sound
Barchard, K. A., Lapping-Carr, L., Westfall, R. S., Fink-Armold, A., Banisetty, S. B., & Feil- like me, you must be more human’: On the interplay of robot and user features on
Seifer, D. (2020). Measuring the perceived social intelligence of robots. ACM human-robot acceptance and anthropomorphism. In 2012 7th ACM/IEEE
Transactions on Human-Robot Interaction (THRI), 9, 1–29. international conference on human-robot interaction (HRI) (pp. 125–126).
Bartneck, C., Kulić, D., Croft, E., et al. (2009). Measurement instruments for the Ferchaud, A., Grzeslo, J., Orme, S., & LaGroue, J. (2018). Parasocial attributes and
anthropomorphism, animacy, likeability, perceived intelligence, and perceived YouTube personalities: Exploring content trends across the most subscribed YouTube
safety of robots. Int J of Soc Robotics, 1, 71–81. channels. Computers in Human Behavior, 80, 88–96.
Bastiansen, M. H. A., Kroon, A. C., & Araujo, T. (2022). Female chatbots are helpful, male Fiske, S. T. (2012). Journey to the edges: Social structures and neural maps of inter-group
chatbots are competent? Publizistik, 67, 601–623. processes. British Journal of Social Psychology, 51(1), 1–12.
Bozeman, D. P., & Kacmar, K. M. (1997). A cybernetic model of impression management Fiske, S. T. (2018). Stereotype content: Warmth and competence endure. Current
processes in organizations. Organizational Behavior and Human Decision Processes, 69 Directions in Psychological Science, 27, 67–73.
(1), 9–30. Fiske, S. T., Cuddy, A. J., & Glick, P. (2007). Universal dimensions of social cognition:
Broadbent, E., Stafford, R. Q., & MacDonald, B. A. (2009). Acceptance of healthcare Warmth and competence. Trends in Cognitive Sciences, 11, 77–83.
robots for the older population: Review and future directions. International Journal of Fiske, C. S. T., Glick, A. J. C., PeterXu, & Jun. (2002). A model of (often mixed)
Social Robotics, 1, 319–330. stereotype content: Competence and warmth respectively follow from perceived
Broverman, I. K., Vogel, S. R., Broverman, D. M., Clarkson, F. E., & Rosenkrantz, P. S. status and competition. Journal of Personality and Social Psychology, 82(6), 878–902.
(1972). Sex-role stereotypes: A current appraisal. Journal of Social Issues, 28(2), Friedmann, E., & Lowengart, O. (2019). Gender segmentation to increase brand
59–78. preference? The role of product involvement. The Journal of Product and Brand
Carpinella, C. M., Wyman, A. B., Perez, M. A., & Stroessner, S. J. (2017). The robotic Management, 28, 408–420.
social attributes scale (RoSAS): Development and validation. 2017 12th ACM/IEEE Gao, Y. L., & Mattila, A. S. (2014). Improving consumer satisfaction in green hotels: The
International Conference on Human-Robot Interaction, HRI, 254–262. roles of perceived warmth, perceived competence, and CSR motive. International
Chee, B. T., Park, T., Jung, Y., Tan, Y. K., & Wong, A. H. (2013). When stereotypes meet Journal of Hospitality Management, 42, 20–31.
robots: The effect of gender stereotypes on people’s acceptance of a security robot. Gefen, D., Karahanna, E., & Straub, D. W. (2003). Trust and TAM in online shopping: An
Interacción, 38, 75–84. integrated model. MIS Quarterly, 27, 51–90.
Cheng, X., Zhang, X., Cohen, J. F., & Mou, J. (2022). Human vs. AI: Understanding the Giles, & David, C. (2002). Parasocial interaction: A review of the literature and a model
impact of anthropomorphism on consumer response to chatbots from the perspective for future research. Media Psychology, 4(3), 279–305.
of trust and relationship norms. Information Processing & Management, 59, Article Gladstein, G. A. (1983). Understanding empathy: Integrating counseling, developmental,
102940. and social psychology perspectives. Journal of Counseling Psychology, 30, 467–482.
Christoforakos, L., Gallucci, A., Surmava, T., Ullrich, D., & Diefenbach, S. (2021). Can Go, E., & Sundar, S. S. (2019). Humanizing chatbots: The effects of visual, identity and
robots earn our trust the same way humans do? A systematic exploration of conversational cues on humanness perceptions. Computers in Human Behavior, 97,
competence, warmth, and anthropomorphism as determinants of trust development 304–316.
in HRI. Frontiers in Robotics and AI, 8. Gupta, V. K., Turban, D. B., & Bhawe, N. (2008). The effect of gender stereotype
Clark, M. S., & Mils, J. (1993). The difference between communal and exchange activation on entrepreneurial intentions. Journal of Applied Psychology, 93(5),
relationships: What it is and is not. Personality and Social Psychology Bulletin, 19, 1053–1061.
684–691.

14
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

Halkias, G., & Diamantopoulos, A. (2020). Universal dimensions of individuals’ Nomura, T., Kanda, T., Suzuki, T., & Kato, K. (2009). Age differences and images of
perception: Revisiting the operationalization of warmth and competence with a robots: Social survey in Japan. Interaction Studies, 10, 374–391.
mixed-method approach. International Journal of Research in Marketing, 37, 714–736. Noor, N. A., Hill, S. R., & Troshani, I. (2021). Artificial intelligence service agents: Role of
Haugeland, I. K., Følstad, A., Taylor, C., & Bjørkli, C. A. (2022). Understanding the user parasocial relationship. Journal of Computer Information Systems, 62, 1009–1023.
experience of customer service chatbots: An experimental study of chatbot Park, N., Jang, K., Cho, S., & Choi, J. (2021). Use of offensive language in human-
interaction design. International Journal of Human-Computer Studies, 161, Article artificial intelligence chatbot interaction: The effects of ethical ideology, social
102788. competence, and perceived humanlikeness. Computers in Human Behavior, 121,
He, J. C., & Kang, S. K. (2019). Covering in cover letters: Gender and self-presentation in Article 106795.
job applications. Academy of Management Journal, 64(4). Pelau, C., Dabija, D., & Ene, I. (2021). What makes an AI device human-like? The role of
Hentschel, T., Heilman, M. E., & Peus, C. (2019). The multiple dimensions of gender interaction quality, empathy and perceived psychological anthropomorphic
stereotypes: A current look at men’s and women’s characterizations of others and characteristics in the acceptance of artificial intelligence in the service industry.
themselves. Frontiers in Psychology, 10. Computers in Human Behavior, 122, Article 106855.
Hofstede, G. (1986). Cultural differences in teaching and learning. International Journal of Pereira, M. C., Coelho, F., & Silva, G. M. (2023). Is there a happy culture? Multiple paths
Intercultural Relations, 10, 301–320. to national subjective well-being. Kyklos, 76(4), 613–641.
Horton, D., & Wohl, R. R. (1956). Mass communication and para-social interaction; Petersen, D., Andrew, J., Doorn, J. V., Mende, M., Noble, S. M., Hulland, J.,
observations on intimacy at a distance. Psychiatry, 19(3), 215–229. Ostrom, A. L., Grewal, D., & Petersen, J. A. (2017). Domo Arigato Mr. Roboto.
Hu, Q., Lu, Y., Pan, Z., Gong, Y., & Yang, Z. (2021). Can AI artifacts influence human Journal of Service Research, 20, 43–58.
cognition? The effects of artificial autonomy in intelligent personal assistants. Pino, G., Zhang, C. X., & Wang, Z. (2020). “(S)he’s so hearty”: Gender cues, stereotypes,
International Journal of Information Management, 56, Article 102250. and expectations of warmth in peer-to-peer accommodation services. International
Huang, Y., Gursoy, D., Zhang, M., Nunkoo, R., & Shi, S. (2021). Interactivity in online Journal of Hospitality Management, 91, Article 102650.
chat: Conversational cues and visual cues in the service recovery process. Pinxteren, M. M., Wetzels, R., Rüger, J., Pluymaekers, M., & Wetzels, M. (2019). Trust in
International Journal of Information Management, 60. humanoid robots: Implications for services marketing. Journal of Services Marketing,,
Huang, M., & Rust, R. T. (2018). Artificial intelligence in service. Journal of Service 33(4), 507–518.
Research, 21, 155–172. Pitardi, V., Bartikowski, B., Osburg, V., & Yoganathan, V. (2022). Effects of gender
Hudson, J., Orviská, M., & Huňady, J. (2017). People’s attitudes to robots in caring for congruity in human-robot service interactions: The moderating role of masculinity.
the elderly. International Journal of Social Robotics, 9, 199–210. International Journal of Information Management, 70, Article 102489.
Jonsson, I., Nass, C., Harris, H., & Takayama, L. (2017). Matching in-car voice with driver Reeves, B., & Nass, C. (1996). The media equation - how people treat computers, television,
state : Impact on attitude and driving performance. and new media like real people and places.
Kassing, J. W., & Sanderson, J. (2009). “You’re the kind of guy that we all want for a Robert, L. P., Alahmad, R., Esterwood, C., Kim, S., You, S., & Zhang, Q. (2020). A review
drinking buddy”: Expressions of parasocial interaction on Floydlandis.com. Western of personality in human‒robot interactions. Now Foundations and Trends.
Journal of Communication, 73(2), 182–203. Rubin, R. B., & McHugh, M. P. (1987). Development of parasocial interaction
Kim, A., Cho, M., Ahn, J., & Sung, Y. (2019). Effects of gender and relationship type on relationships. Journal of Broadcasting & Electronic Media, 31, 279–292.
the response to artificial intelligence. Cyberpsychology, Behavior, and Social Rubin, A. M., Perse, E. M., & Powell, R. A. (1985). LONELINESS, parasocial interaction,
Networking, 22(4), 249–253. and local television news viewing. Human Communication Research, 12, 155–180.
Kim, J., Merrill, K., & Collins, C. (2021). AI as a friend or assistant: The mediating role of Sakib, N., Zolfagharian, M., & Yazdanparast, A. (2020). Does parasocial interaction with
perceived usefulness in social AI vs. functional AI. Telematics and Informatics, 64, weight loss vloggers affect compliance? The role of vlogger characteristics, consumer
Article 101694. readiness, and health consciousness. Journal of Retailing and Consumer Services, 52.
Kitamura, F. C. (2023). ChatGPT is shaping the future of medical writing but still requires Schillaci, C. E, de Cosmo, L. M., Piper, L., Nicotra, M., & Guido, G. (2024).
human judgment. Radiology, Article 230171. Anthropomorphic chatbots’ for future healthcare services: Effects of personality,
Kolbl, Ž., Arslanagić-Kalajdžić, M., & Diamantopoulos, A. (2019). Stereotyping global gender, and roles on source credibility, user satisfaction, and intention to use.
brands: Is warmth more important than competence? Journal of Business Research, Technological Forecasting and Social Change, 199.
104. Searle, J. R. (1981). Analytic philosophy and mental phenomena. Midwest Studies In
Kotzur, P. F., Veit, S., Namyslo, A., Holthausen, M., Wagner, U., & Yemane, R. (2020). Philosophy, 6, 243–265.
“Society thinks they are cold and/or incompetent, but I do not”: Stereotype content Seo, S. (2022). When female (male) robot is talking to me: Effect of service robots’
ratings depend on instructions and the social group’s location in the stereotype gender and anthropomorphism on customer satisfaction. International Journal of
content space. British Journal of Social Psychology, 59(4), 1018–1042. Hospitality Management, 102, 103166.
Kull, A. J., Romero, M., & Monahan, L. (2021). How may I help you? Driving brand Seo, S., Wu, X., & Choi, J. (2024). Is her (his) gender matched or not matched with me?
engagement through the warmth of an initial chatbot message. Journal of Business Gender (dis)match effect between customers and gendered service robots.
Research, 135, 840–850. International Journal of Hospitality Management, 119.
Labrecque, L. I. (2014). Fostering consumer–brand relationships in social media Shan, Y., Chen, K. J., & Lin, J. (2019). When social media influencers endorse brands:
environments: The role of parasocial interaction. Journal of Interactive Marketing, 28, The effects of self-influencer congruence, parasocial identification, and perceived
134–148. endorser motive. International Journal of Advertising, 39, 590–610.
Lee, K. M., & Nass, C. (2005). Social-psychological origins of feelings of presence: Shuqair, S., Pinto, D. C., So, K. K., Rita, P., & Mattila, A. S. (2021). A pathway to
Creating social presence with machine-generated voices. Media Psychology, 7, 31–45. consumer forgiveness in the sharing economy: The role of relationship norms.
Lee, K. M., Peng, W., Jin, S. A., & Yan, C. (2006). Can robots manifest personality? : An International Journal of Hospitality Management, 98, 1–8.
empirical test of personality recognition, social responses, and social presence in Skjuve, M., Følstad, A., Fostervold, K. I., & Brandtzaeg, P. B. (2021). My chatbot
human-robot interaction. Journal of Communication, 56, 754–772. companion - a study of human-chatbot relationships. International Journal of Human-
Li, M., & Suh, A. (2022). Anthropomorphism in AI-enabled technology: A literature Computer Studies, 149, Article 102601.
review. Electronic Markets, 32, 2245–2275. Spielmann, N., Dobscha, S., & Lowrey, T. M. (2021). Real men don’t buy “Mrs. Clean”:
Liu, X., Ren, P., Lv, X., & Li, S. (2024). Service experience and customers’ eWOM Gender bias in gendered brands. Journal of the Association for Consumer Research, 6,
behavior on social media platforms: The role of platform symmetry. International 211–222.
Journal of Hospitality Management, 119. Troshani, I., Hill, S. R., Sherman, C., & Arthur, D. (2020). Do we trust in AI? Role of
Lv, X., Luo, J., Liang, Y., Liu, Y., & Li, C. (2022). Is cuteness irresistible? The impact of anthropomorphism and intelligence. Journal of Computer Information Systems, 61,
cuteness on customers’ intentions to use ai applications. Tourism Management, (Jun), 481–491.
90. Trovato, G., Lucho, C., & Paredes, R. (2018). She’s electric - the influence of body
Martin, C. L., Ruble, D. N., & Szkrybalo, J. (2002). Cognitive theories of early gender proportions on perceived gender of robots across cultures. Robotics, 7, 50.
development. Psychological Bulletin, 128(6), 903–933. Tyrväinen, O., & Karjaluoto, H. (2025). Unrequited love? : A mixed-methods study of
Martins, J. M., Yusuf, F., & Swanson, D. A. (2011). Perspectives on consumer behaviour. parasocial engagement with social media influencers. International Journal of
In Consumer demographics and behaviour. The springer series on demographic methods Information Management, 80, Article 102845.
and population analysis (Vol. 30). Dordrecht: Springer. Wirtz, J., Patterson, P. G., Kunz, W. H., Gruber, T., Lu, V. N., Paluch, S., & Martins, A.
Mehrabian, A., & Russell, J. A. (1974). An approach to environmental psychology. (2018). Brave new world: Service robots in the frontline. Journal of Service
Meng, J., & Dai, Y. (2021). Emotional support from AI chatbots: Should a supportive Management, 29(5), 907–931.
partner self-disclose or not? J. Comput. Mediat. Commun., 26, 207–222. Xie, Y., Zhu, K., Zhou, P., & Liang, C. (2023). How does anthropomorphism improve
Monod, E., Watson-Manheim, M. B., Qi, I., Joyce, E., Mayer, A., & Santoro, F. M. (2022). human-AI interaction satisfaction: A dual-path model. Computers in Human Behavior,
(Un)intended consequences of AI sales assistants. Journal of Computer Information 148, Article 107878.
Systems, 63, 436–448. Xue, J., Zhou, Z., Zhang, L., & Majeed, S. (2020). Do brand competence and warmth
Mordor Intelligence M. (2023). Social robot market - growth, trends and forecasts (2023- Always influence purchase intention? The moderating role of gender. Frontiers in
2028). Mordor Intelligence. Available from: Social Robots Market - Size, Share & Psychology, 11.
Growth (mordorintelligence.com. Youn, S., & Jin, S. V. (2021). "In A.I. we trust?" The effects of parasocial interaction and
Munnukka, J., Talvitie-Lamberg, K., & Maity, D. (2022). Anthropomorphism and social technopian versus luddite ideological views on chatbot-based customer relationship
presence in Human-Virtual service assistant interactions: The role of dialog length management in the emerging "feeling economy". Computers in Human Behavior, 119,
and attitudes. Computers in Human Behavior, 135, Article 107343. Article 106721.
Nass, C., Moon, Y., & Carney, P. (1999). Are respondents polite to computers? Social
desirability and direct responses to computers. Journal of Applied Social Psychology,
29(5), 1093–1110.

15
T. Qi et al. Computers in Human Behavior 167 (2025) 108625

Yuksel, M., & Labrecque, L. I. (2016). “Digital buddies”: Parasocial interactions in social Zhang, H., Bai, X., & Ma, Z. (2022). Consumer reactions to AI design: Exploring consumer
media. Journal of Research in Interactive Marketing, 10, 305–320. willingness to pay for AI-designed products. Psychology and Marketing,, 39(11).
Zeithaml, V., Parasuraman, A., & Berry, L. L. (1985). Problems and strategies in services
marketing. Journal of Marketing, 49, 33–46.

16