Journal of Retailing and Consumer Services 77 (2024) 103690

Contents lists available at ScienceDirect

Journal of Retailing and Consumer Services

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

Human vs. AI: The battle for authenticity in fashion design and
consumer response
Garim Lee a, *, Hye-Young Kim b
a
Merchandising, Indiana University Bloomington, 117 Kirkwood Hall, 130 S Woodlawn Ave., Bloomington, IN, 47405, USA
b
Retail Merchandising, University of Minnesota, 240 McNeal Hall, 1985 Buford Ave., St. Paul, MN, 55108-6136, USA

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

Keywords: Generative Artificial Intelligence (AI) empowers the AI design process. Then, how do consumers respond to AI-
Generative AI designed fashion products? Building on schema theory, this research investigated the extent to which AI-
AI-Assisted design designed clothing is perceived as authentic through three online experiments. Study 1 (n = 121) and Study 2
Schema theory
(n = 161) showed consumers generally respond more favorably to human-designed (vs. AI-designed) clothing,
Authenticity
AI customization
which is driven by perceived authenticity and expected product quality. Study 3 (n = 156) confirmed that
negative responses toward AI-designed clothing can be mitigated when consumers have the option to provide
input to customize the design because it enhances perceived authenticity. Study findings offer a theoretical
understanding of how and why consumers respond to AI-designed products and practical guidelines for retailers.

1. Introduction subsequent evaluation. While a few studies investigated consumer re­

sponses to AI versus human design, a consensual agreement building on
The retail industry is experiencing unparalleled transformation theoretical knowledge of how people evaluate AI-designed fashion
accelerated by the development of Artificial Intelligence (AI) and products has not been sufficiently made. Furthermore, because con­
computing efficiency. A promising area of AI applications is fashion sumer responses to AI design technology are yet underresearched from
design. In the AI-assisted design process, designers can develop new various perspectives, plenty of room remains to investigate how and
designs inspired by the future fashion trends predicted by AI-powered why generative AI in the fashion design process impacts consumers’
search engines (Harreis et al., 2023). Furthermore, the recent growth product and brand evaluation.
of generative AI powered by advanced machine learning and easier To fill the research gap, this study aims to advance the understanding
access to big data opened the door for companies for AI-driven fashion of how people evaluate AI-designed fashion products. Specifically,
design. Generative AI refers to the AI techniques and models that learn a under the framework of schema theory and the authenticity literature,
representation of objects in input data and generate new, original output three online experiments examined consumer responses to fashion
that keeps the identity of the original data (Murphy, 2022). Leveraging products designed by AI by comparing those designed by a human
the power of generative AI to create thousands of renderings, companies designer. In addition, the mediation of perceived authenticity and ex­
have recently started to use AI to understand customer needs and design pected product quality was tested to see if they explain the effect of
better-suited apparel items with increased efficiency (Figoli et al., 2022; design entity (AI vs. human) on consumer responses. Lastly, the role of
Lee, 2022). For example, Acne Studio showcased AI-assisted fashion AI customization that changes consumer responses to fashion products
designs for its ready-to-wear collection in Paris Fashion Week 2020. Levi designed by AI is examined.
Strauss & Co. used AI algorithms to create new design alternatives for Investigating consumer evaluation of AI-designed fashion products is
denim jackets as well as to optimize their production and manufacturing timely and crucial. The recent interest in generative AI (e.g., GPT-X,
processes. Midjourney, DALL⋅E) is incredibly surging due to technological prog­
While AI in the fashion design process is becoming more prevalent, ress, and empirical findings on consumer responses to generative AI in
not many studies have been conducted focusing on the way AI in the fashion design are essential to help establish the direction of the relevant
design process affects consumers’ perception of the product and the business strategies. From the schema theory perspective, the findings of

* Corresponding author.
E-mail addresses: garilee@iu.edu (G. Lee), hykim@umn.edu (H.-Y. Kim).

https://doi.org/10.1016/j.jretconser.2023.103690
Received 22 September 2023; Received in revised form 12 November 2023; Accepted 15 December 2023
Available online 22 December 2023
0969-6989/© 2023 Elsevier Ltd. All rights reserved.
G. Lee and H.-Y. Kim Journal of Retailing and Consumer Services 77 (2024) 103690

the three online experiments provide important theoretical and practical consumers are willing to pay more for AI-designed (vs. human-designed)
implications by enlightening why and how consumers form particular products because of curiosity. More recently, Pieconka (2023) found
responses to fashion products designed by AI, compared to human- that consumers are less willing to pay for AI-designed products because
designed (Study 1, Study 2) and AI customization (Study 3). Conse­ of little personal benefits from already finalized AI-designed products.
quently, the study findings offer implications for practitioners seeking
strategies to interact with consumers about generative AI usage in their
design process. 2.2. Schema theory

2. Literature review & hypotheses development This study draws from schema theory to conceptualize the effect of
design entity (AI vs. human) on consumer responses. Schema theory
2.1. AI in fashion design process explains that people categorize information in schemata, which are
memory-stored cognitive frameworks that represent the organization of
AI allows companies to collect and analyze massive amounts of sales their background knowledge or experience for specific domains (Fiske,
and consumer data. By leveraging such capabilities, companies utilize 1982). Schemata facilitate people’s decision-making processes as newly
various AI-powered techniques including virtual agents and personal­ incoming information is processed in an individual’s perception based
ized services to offer better customer experience (Alimamy and Kuhail, on existing schema (Bem, 1981). That is, people retrieve information
2023; Kim et al., 2023; Li et al., 2023; Yu et al., 2023). Furthermore, and categorize new input based on their expectations about a particular
generative AI creates new image-based outputs contingent on input data domain (Davvetas and Diamantopoulos, 2016; Sujan and Bettman,
or prompts, bringing about innovations in fashion design. GAN-based 1989). People may have schemata of AI that has a low ability to possess
models such as a Deep Convolutional GAN (DCGAN) can synthesize artistic intention and create artwork, which may affect their responses to
design sketches, garment patterns, and texture information to generate AI-created artwork (Chamberlain et al., 2018; Hong and Curran, 2019).
realistic design alternatives (Guo et al., 2023; Xian et al., 2018; Yan Consumers’ perceptions of products and brands are significantly
et al., 2022). After input data, such as previous company designs, affected by schemata (Davvetas and Diamantopoulos, 2016; Lee and
budget, materials, and design features, is entered, algorithms generate, Kim, 2022; Meyers-Levy and Tybout, 1989). When consumers view
evaluate, refine, and optimize design alternatives. The design outputs product information from retailers or brands, they use previously
can then be evaluated and regenerated by modifying the input data or accumulated relevant knowledge to interpret the information. The
fine-tuning the algorithm models. Fig. 1 illustrates the AI-assisted design activated schemata settle the initial configuration of defining or evalu­
process. Fashion brands can use AI to produce new profitable designs ating the products or brands. When new product information is consis­
with enhanced time and labor efficiency while identifying market trends tent with an individual’s preexisting schema, the information is easily
and patterns (Figoli et al., 2022; Harreis et al., 2023; Lee, 2022). processed and often leads to a basic sense of liking without evoking
While the studies on people’s perceptions of AI-generated creative fluctuating responses toward the product (Halkias and Kokkinaki,
works mainly focused on text-based or non-commercial arts (see Ap­ 2014). Similarly, an individual may assimilate or accommodate the in­
pendix A for review), only a few studies have investigated how con­ formation which moderately diverges from the previous knowledge and
sumers respond to AI-designed fashion products. Some focused on modify existing schemata (Mandler, 1981), and such experiences can
consumer evaluation and attitude toward the products. In a seminal positively influence the evaluation of the stimuli (Meyers-Levy and
work, Lee (2022) confirmed that human-generated fashion design is Tybout, 1989). In contrast, new product or brand information that
more favorably evaluated than AI-generated design. Similarly, Xu and highly conflicts with one’s schema prevents the individual from being
Mehta (2022) demonstrated that consumers form more negative brand willing to accommodate and thus evokes reactance and negative eval­
attitudes toward AI-designed (vs. human-designed) luxury products due uation (Boush and Loken, 1991; Halkias and Kokkinaki, 2014; Mandler,
to perceived brand essence, especially for fashion. Others focused more 1981).
on willingness to pay, resulting in mixed findings. Sohn et al. (2020) Using the schema theory as the theoretical base, the next sections
confirmed that young consumers are more willing to pay for describe consumers’ perceptions of AI as a design entity, which repre­
GAN-generated fashion design. Zhang et al. (2022) found that sents discrepant information from people’s existing schemata. Specif­
ically, the subsequent evaluations of the AI-designed clothing are

Fig. 1. Illustration of AI-assisted design process.

2
G. Lee and H.-Y. Kim Journal of Retailing and Consumer Services 77 (2024) 103690

proposed to be explained by perceived authenticity and expected quality judgments about the product (Rao and Monroe, 1989). Impor­
product quality. tantly, research has found that consumers’ authenticity attribution
positively affects the quality perception of stimuli (Cinelli and LeBoeuf,
2.3. Perceived authenticity and consumer responses toward AI-designed 2020; Moulard et al., 2016). For example, Cinelli and LeBoeuf (2020)
clothing confirmed that the intrinsic motivation of a firm for producing products
increases anticipated product quality via perceived authenticity. Such a
This study proposes that information about design entity, either AI or relationship between perceived authenticity and product quality is
humans, serves as the critical determinant that affects perceived product attributed to the perception of quality commitment, which is one of the
authenticity. Authenticity is defined as what is genuine, real, true, and dimensions of perceived authenticity (Gilmore and Pine, 2007; Napoli
unique (Beverland and Farrelly, 2010; Moulard et al., 2014). Consumers et al., 2014). Consumers are likely to perceive higher authenticity from a
perceive specific levels of authenticity from products along various di­ product as they believe a quality commitment has been made to the
mensions, including product information, influencer characteristics, product. Consequently, perceived quality of the product is enhanced,
advertising messages, production methods, materials, origins, and con­ ultimately increasing positive consumer responses.
nections to firm value (Audrezet et al., 2020; Campagna et al., 2023; Extending such reasoning based on the literature to AI-designed
Newman and Dhar, 2014). When authenticity is perceived based on clothing, perceived authenticity and expected product quality are
whether a product is made with the appropriate techniques and pro­ likely to explain the negative responses to AI-designed fashion. The
cesses or whether a passionate creator makes a product, they are also product designed by genuine and passionate human designers compared
known as craft authenticity and artist authenticity (Littrell et al., 1993; to AI will be deemed to be of quality commitment in overall product
Moulard et al., 2014), respectively. For example, a souvenir product is design and thus more authentic. Given that the perceptions of dedication
perceived to be highly authentic if it is associated with uniqueness, to design are the critical determinant for assessing designed product
workmanship, artistry, aesthetics, and cultural and historical integrity. quality (Zhu et al., 2009), the clothing with a higher expected quality
Perceived authenticity has been found to positively affect consumer commitment will be considered to be high quality.
responses, such as overall product valuation (Newman and Dhar, 2014)
H3. Perceived authenticity and expected product quality will serially
and attitudes toward the products or brands and purchase intention
explain the negative responses to AI-designed (vs. human-designed)
(Carsana and Jolibert, 2018; Moulard et al., 2014; Napoli et al., 2014).
clothing.
It is expected that the discrepancy between people’s preexisting
fashion design schema and AI (vs. humans) is large that perceived
2.5. The role of AI customization
authenticity is lowered from AI-designed fashion products than human-
designed ones. Fashion design has been conventionally considered a
How can consumers’ negative responses to AI-designed clothing be
human-dominant area. People expect fashion design elements to
mitigated? This study proposes that real-time customization in AI design
encompass designers’ intentions, knowhows, personal characteristics,
process will change the extent to which consumers perceive authenticity
and sophisticated skills (Lamb and Kallal, 1992). This implies that
from AI-designed clothing, and in turn, reduce negative responses. Based
fashion design schemata are associated with human work and charac­
on the existing literature and available technologies (Aggarwal et al.,
teristics. In contrast, AI as a design entity in fashion represents the in­
2020; Harreis et al., 2023; Yan et al., 2022), AI customization is
formation that highly conflicts with the existing schemata, considering
conceptualized as the process through which consumers provide input to
the long history of intersection between humans and fashion design
personalize their products in real-time based on particular design com­
works and the relevant schemata. Generative AI is developed to create
ponents and AI generates designs to actualize customization accord­
original content, but it is true that many people question the reliability
ingly. Mass customization in product design utilizes flexible
and authenticity of the output from AI (Bang et al., 2023; Harper, 2023),
computer-aided systems. The benefits of mass customization have
which potentially indicates low perceived authenticity. Because gener­
been well documented. Key market performance metrics such as sales
ative AI is trained on data lakes and question snippets to create content
revenues and gross margin, as well as consumer evaluations, outperform
responding to prompts, the originality of the content creation is often
when product design pipelines encompass customer involvement
doubted (Appel et al., 2023). While no research empirically tested the
(Moreau and Herd, 2010; Nishikawa et al., 2013; Stevens et al., 2017).
authenticity perception of AI-generated design, Jago (2019) showed
Furthermore, a positive connection between personalization and
that algorithm-made decisions across domains are perceived to be low in
authenticity is speculated in that both draw favorable consumer atti­
authenticity and sincerity and thus less favored, compared to human
tudes and intentions toward AI, as implied in the literature (Alimamy
decisions. Such a pattern is predicted to be amplified for fashion design
and Kuhail, 2023; Chu et al., 2022). For example, Alimamy and Kuhail
because fashion is especially considered to be the domain where the
(2023) found that both perceived authenticity and personalization of a
importance of the human touch cannot be overlooked. Thus, consumers
virtual assistant predict consumer commitment and trust. When it comes
will perceive higher authenticity from human-designed clothing
to AI design process, AI customization is likely to increase perceived
compared to AI-designed clothing, which enhance consumers’ favorable
authenticity by inducing the values of mass customization and enhanced
responses.
benefits of real-time personalization. Therefore, it is expected that the
H1. Consumers will form more favorable responses (i.e., attitude and negative effect of AI as a design entity on perceived authenticity will be
purchase intention) toward human-designed clothing than AI-designed alleviated by AI customization, consequently mitigating the negative
clothing. responses to AI-designed clothing.
H2. Perceived authenticity will explain the negative responses to AI- H4. AI customization will mitigate the negative effect of AI as design
designed (vs. human-designed) clothing. entity on consumer responses (i.e., attitude and purchase intention).

2.4. Perceived authenticity and expected product quality 3. Research model & overview of the studies

If perceived authenticity on the basis of one’s schemata of AI explains Fig. 2 presents the research model along with an overview of the
why consumers would respond differently to human-designed versus AI- studies and the variables tested. This study adopted a quantitative
designed clothing, consumer expectations of product quality will likely approach, and three online experiments were conducted to test the hy­
serve as the second serial mediator. Consumers rely on extrinsic signals potheses. As a scenario-based experiment using real brand names, Study
such as product information, warranty, or brand reputation to make 1 examined the main assumptions of this research by testing whether

3
G. Lee and H.-Y. Kim Journal of Retailing and Consumer Services 77 (2024) 103690

Fig. 2. Research model and overview of the studies.

design entity (AI vs. human) affects product attitude and purchase 4. Study 1
intention (H1) and whether perceived authenticity mediates such an
effect (H2). Study 2 replicated Study 1 using stimuli of a product image 4.1. Methods
and a fictitious brand name. Further, Study 2 examined whether
perceived authenticity and expected product quality serially explain the 4.1.1. Study design and procedure
effects of design entity (H3). Lastly, Study 3 explored the role of AI A scenario-based online experiment was conducted where design
customization by testing whether AI design process with versus without entity (AI vs. human) and two brand replicates were entered as between-
customization features affects product attitude, brand attitude, and subject factors. One hundred and forty-two U.S. participants were
purchase intention (H4). The mediations of perceived authenticity (H2) recruited from Amazon Mturk. After deleting those who failed multiple
and the authenticity–product quality chain (H3) explaining the miti­ qualification and attention check questions, 121 participants remained
gating effect of AI customization were also tested in Study 3. The study for further analysis (Mage = 38.84, SDage = 11.59; 66.9% male; see
design, data analysis, results, and discussion of each study are described Appendix B and C for details). Once participants agreed to participate in
below. the study, they were randomly given one of the four shopping scenarios

Table 1
Measurement items.
Item Content Study 1 Study 2 Study 3

M (SD) Loadings M (SD) Loadings M (SD) Loadings

Perceived Authenticity α = .91 α = .90 α = .88

Ingenuine–Genuine 5.64 (1.52) .90 5.51 (1.28) .78 4.26 (1.59) .90
Not authentic–Authentic 5.74 (1.45) .94 5.50 (1.30) .81 4.26 (1.66) .90
Not original–Original 5.62 (1.41) .83 5.30 (1.64) .83 3.35 (1.75) .69
Unfaithful–Faithful 5.51 (1.40) .78 5.34 (1.25) .76 4.44 (1.30) .70
Not natural–Natural 5.00 (1.83) .71 5.30 (1.41) .80 4.29 (1.47) .67

Expected Product Quality - α = .85 α = .96

This sweater would be reliable. - - 5.69 (1.02) .86 4.67 (1.28) .94
This sweater would be dependable. - - 5.58 (1.08) .67 4.62 (1.27) .94
This sweater would be durable. - - 5.60 (1.05) .79 4.64 (1.25) .91
The workmanship on this sweater would be good. - - 5.66 (1.13) .90 4.41 (1.35) .91

Product Attitude α = .94 α = .94 α = .97

Dislikable–Likable1 5.95 (1.25) .89 - - - -
Bad–Good123 5.92 (1.31) .89 5.30 (1.46) .91 4.41 (1.74) .95
Negative–Positive123 5.93 (1.33) .96 5.24 (1.48) .92 4.38 (1.72) .93
Unfavorable–Favorable23 - - 5.17 (1.56) .88 4.26 (1.77) .95
Undesirable–Desirable23 - - 5.10 (1.55) .88 4.04 (1.72) .92

Purchase Intention α = .95 α = .95 α = .97

The likelihood of purchasing this sweater is high.1 5.22 (1.66) .93 - - - -
The probability that I would consider buying the sweater is high.1 5.44 (1.65) .91 - - - -
My willingness to buy the sweater is high.1 5.37 (1.62) .94 - - - -
I would consider buying the sweater.123 5.60 (1.48) .88 5.31 (1.64) .90 3.52 (1.94) .91
It is likely that I will purchase this sweater.23 - - 5.07 (1.73) .93 3.01 (1.80) .98
It is probable that I will purchase this sweater.23 - - 5.09 (1.67) .95 3.08 (1.86) .99

Brand Attitude - α = .92 α = .97

Bad–Good - - 5.12 (1.58) .92 4.34 (1.49) .96
Dislike very much–Like very much* - - 2.89 (1.57) .38* 4.45 (1.46) .94
Unpleasant–Pleasant - - 5.06 (1.67) .92 4.27 (1.49) .97

Note. 1Study 1; 2Study 2; 3Study 3; *reversed and deleted in Study 2 due to low factor loading.
All factor loadings were significant at 0.001 level.

4
G. Lee and H.-Y. Kim Journal of Retailing and Consumer Services 77 (2024) 103690

and completed the questionnaire. The questionnaire including the replicates.

shopping scenarios in an online repository and a diagrammatic repre­ MANCOVA was performed with design entity and brand replicates as
sentation of study procedure are available in Appendix D. independent variables and perceived authenticity, product attitude, and
Shopping scenarios manipulated the design entity (AI vs. human). In purchase intention as dependent variables. Both brand familiarity and
addition, two different brand types were used in the scenarios as brand brand attitude showed significant correlations with the dependent var­
replicates to see if the effect of design entity holds across the brands. iables, showing suitability as potential covariates. Next, while brand
Based on the previous literature on brand type (Mitchell and Balabanis, familiarity did not significantly affect dependent variables (p = .076),
2021), Louis Vuitton and H&M were used to represent different brand brand attitude showed a significant effect (p < .001) and thus was
types. Design entity and brand replicates were manipulated using a entered as a covariate. The statistical assumptions were checked, and the
brand sweater scenario. A sweater was used in the scenario to rule out results showed the appropriateness of the methods used (see Appendix
potential confounding effects, as past literature indicates that the F). The main effect of design entity on the dependent variables was
conceptualization of a sweater is well-defined, consistent among various significant (Pillai’s Λ = 0.16, Wilks’ λ = 0.84, p < .001, partial η2 =
people, and stable over time (DeLong et al., 1986). 0.16) whereas brand replicates did not show a significant effect (p =
.151). The design entity × brand replicate interaction effect was not
4.1.2. Measurements significant (p = .415). The results remained the same sans the covariate
The multi-item scales were drawn from the existing literature and in that nonsignificant and significant effects hold with and without the
modified to fit the context. The measurement items for the main vari­ covariate, which supports the robustness of the findings (see Appendix G
ables, item wise descriptive, and Cronbach’s alpha scores are presented for the results without covariates). The subsequent ANCOVA showed
in Table 1. Participants’ responses to multi-item scales for a construct participants perceived higher authenticity from human-designed (vs. AI-
were averaged to calculate an index score. Perceived authenticity was designed) sweater (MAI = 5.06 vs. Mhuman = 5.97; F = 18.81, p < .001,
measured using five items derived from Beverland and Farrelly’s (2010) η2 = 0.14). Participants in the human condition showed more favorable
study. Product attitude was measured using three items from Chae and attitudes (MAI = 5.72 vs. Mhuman = 6.17; F = 5.18, p = .025, η2 = 0.04)
Hoegg’s (2013) study and purchase intention was measured with four and marginally higher purchase intention (MAI = 5.17 vs. Mhuman =
items developed by Dodds et al. (1991). Brand luxuriousness (main­ 5.67; F = 3.67, p = .058, η2 = 0.03). Thus, H1 was supported.
stream–luxury; Moreau et al., 2020) and reputation (bad reputation–good
reputation; Chaudhuri, 2002) were measured on 7-point semantic scale 4.2.3. Mediation
items from the literature to ensure whether the two brands represent PROCESS macro version 4.2 in SPSS was used for mediation analysis.
different types. Harman’s one-factor test was used to test common The evidence of a significant indirect effect is provided if the 95%
method bias. The total variance extracted by one factor using the 32 bootstrap confidence intervals do not contain zero. Model 4 (5000
items was 38.68%. As the threshold indicating common method bias is bootstrapped samples) was performed to test the mediation of perceived
50% (Podsakoff and Organ, 1986), common method bias was not a authenticity between design entity and the dependent variables while
major concern. controlling brand attitude. As hypothesized, the indirect effects of
design entity on product attitude (b = 0.57, 95% bootstrap CI = [0.2483,
4.1.3. Covariates 0.9334]) and purchase intention (b = 0.81, CI = [0.4382, 1.2041])
Potential covariates were selected a priori based on the literature through perceived authenticity were significant. The direct effect of
review and the study design. Given that real brands were used in Study design entity on product attitude (p = .448) and purchase intention (p =
1, brand attitude (negative–positive) and familiarity (unfamiliar–familiar) .139) became insignificant when perceived authenticity was entered as a
were measured on 7-point semantic scale items as potential covariates. mediator, suggesting the full mediation of perceived authenticity on the
This was because the literature suggested neutralizing the confounding effect of design entity and the two dependent variables. Thus, H2 was
effects of existing brands by controlling for familiarity or previous supported.
experience with and general attitude toward existing brands (Geuens
and De Pelsmacker, 2017; Ha et al., 2019). 4.3. Discussion

4.2. Results Study 1 confirmed that consumers more favorably respond to and are
slightly more willing to purchase clothing designed by humans than by
4.2.1. Measurement testing AI (H1). Such an effect was significant across the brand replicates in
The validity and reliability of the measurement were tested by Study 1. Also, Study 1 provided initial evidence for the mediating effect
confirmatory factor analysis (CFA) with maximum-likelihood estimation of perceived authenticity (H2). Compared to human-designed clothing,
using SPSS AMOS. The measurement model fit was adequate (χ2 = AI-designed clothing was perceived to be less authentic, which led to
78.86, df = 49; χ2/df = 1.61; CFI = 0.98; NNFI = 0.97; RMSEA = 0.07; lowered product attitude and purchase intention. As a scenario-based
SRMR = 0.03). Construct validity was confirmed as all standardized study using real brands was conducted in Study 1, Study 2 utilizes
factor loadings were high and significant at a p-value of .001 (see product image stimuli and a fictitious brand.
Table 1). Discriminant validity was achieved as all average variance
extracted (AVE) of the constructs were larger than the squared corre­ 5. Study 2
lations. Cronbach’s alpha (α>0.91) and composite reliability (CR>0.92)
were larger than the general threshold of 0.70, supporting reliability. 5.1. Methods
The results of measurement model testing are in Appendix E.
5.1.1. Study design and procedure
4.2.2. Effect of design entity A single-factor (design entity: AI vs. human) between-subject online
SPSS version 29.0 was used for statistical analysis throughout the experiment was conducted. One hundred and eighty U.S. participants
research. A p-value at the level of 0.05 was considered to be significant. were recruited from Amazon Mturk. After deleting those who failed the
Participants perceived Louis Vuitton as more luxurious (M = 6.52 vs. multiple attention check questions, 161 participants remained for
4.34; t = 7.07, p < .001) and more reputational (M = 6.23 vs. 5.61; t = further analysis (Mage = 36.88, SDage = 10.69; 67.7% male; see Ap­
2.95, p = .004) than H&M. No differences were found in brand attitude pendix B and C). Similar to Study 1, shopping scenarios manipulated the
(p = .126) and brand familiarity (p = .575) between the two brands. design entity. Participants were randomly given one of the two shopping
Thus, the two real brands were appropriate to be used for brand scenarios, viewed a product image, and completed the questionnaire.

5
G. Lee and H.-Y. Kim Journal of Retailing and Consumer Services 77 (2024) 103690

The scenario asked them to imagine that they found a sweater designed 5.2. Results
by either Artificial Intelligence (AI) or fashion designers online. Again, a
sweater was used in the study based on the suggestions of the past 5.2.1. Measurement testing
literature and the results of pretest (DeLong et al., 1986; see 4.1.1 The measurement model fit was adequate (χ2 = 212.61, df = 125; χ2/
above). The questionnaire and a diagrammatic representation of study df = 1.701; CFI = 0.97; NNFI = 0.96; RMSEA = 0.07; SRMR = 0.05; see
procedure are available in Appendix D. Appendix E). Construct validity was confirmed as the factor loadings
were high and significant at a p-value of .001 (see Table 1). Discriminant
5.1.2. Stimuli validity was achieved as all AVE values of the constructs were larger
An identical product image was given to participants across the study than the squared correlations. Cronbach’s alpha (α>0.85) and com­
conditions to ensure testing of the causal effect of design entity. A pretest posite reliability (CR>0.88) supported measurement reliability.
(n = 72; Mage = 33.75, SDage = 10.38; 65.3% male; see Appendix C) was
conducted on Amazon Mturk to select appropriate product image stimuli 5.2.2. Effect of design entity
and fictitious brand name. The pretest questionnaire is available in MANCOVA was performed with brand familiarity, perceived threats
Appendix D. The pretest consisted of two parts. Participants were told from AI, and product involvement as covariates. All three covariates
they would see multiple brand names and products and give their showed significant correlations with at least one dependent variable and
opinions about them. First, participants were given a list of six fictitious showed a significant effect (brand familiarity, product involvement: p <
brand names which were created using an online fake brand name .001; perceived threats: p = .040). The statistical assumptions of the
generator. For each brand name, brand attitude and brand familiarity methods used were met (see Appendix F). The main effect of design
were measured using single-item scales: “To what extent do you feel entity was significant (Pillai’s Λ = 0.09, Wilks’ λ = 0.90, p = .012, η2 =
positive or negative about these fashion brand names?” (1 = Extremely 0.09). The results were generally sustained sans covariates, showing
negative–7 = Extremely positive) and “To what extent are you familiar minimal changes in statistics values1 (see Appendix G). Participants
with the following fashion brand names?” (1 = Extremely unfamiliar–7 = perceived higher authenticity from the human-designed (vs. AI-
Extremely familiar). Next, participants viewed four clothing images in designed) sweater (MAI = 5.19 vs. Mhuman = 5.60; F = 8.80, p = .003,
random order: pique shirt, sweater, shirt, and jacket. For each product η2 = 0.06), consistent with Study 1 results. Compared to the AI-designed
image, product attitude was measured using three-item semantic dif­ sweater, the human-designed sweater was perceived to be of marginally
ferential scales: negative–positive, dislikable–likable, and bad–good significantly higher quality (MAI = 5.52 vs. Mhuman = 5.75; F = 3.49, p =
(0.85<α<0.90). .064, η2 = 0.02). Participants in the human-designed condition formed
A fictitious brand name ‘Camource’ were chosen because partici­ more favorable attitudes toward the sweater (MAI = 5.02 vs. Mhuman =
pants showed relatively neutral attitudes toward the brand name (M = 5.39; F = 4.94, p = .028, η2 = 0.03) and the brand (MAI = 4.93 vs.
4.56, SD = 1.45) and low brand familiarity (M = 4.26, SD = 1.88), Mhuman = 5.27; F = 4.20, p = .042, η2 = 0.03). No significant difference
compared to other brand names. Next, a sweater image was selected as was found in purchase intention between the human-designed and the
final because participants showed neutral attitudes toward the sweater AI-designed conditions (p = .837). Thus, H1 was partially supported for
compared to other products (M = 4.92, SD = 1.47). product attitude and brand attitude, but not for purchase intention.

5.1.3. Measurements 5.2.3. Mediation

The multi-item scales were drawn from the literature and modified PROCESS Model 4 was performed with brand familiarity, perceived
(see Table 1). Perceived authenticity was measured using the same items threats from AI, and product involvement as covariates. The approach of
used in Study 1. Expected product quality was measured using Sweeney performing model 4 first was to test the mediation of perceived
et al.’s (1999) items which represent participants’ evaluation of quality authenticity and examine H2 thoroughly in a different study context.
value of the product. Product attitude (Rajagopal and Montgomery, Consistent with Study 1, the indirect effects of design entity through
2012) and purchase intention (Singh and Cole, 1993) were measured perceived authenticity were significant on product attitude (b = 0.28, CI
using other sets of well-established scales. Because Study 2 used ficti­ = [0.0853, 0.5131]), purchase intention (b = 0.31, CI = [0.0984,
tious brands, brand attitude was measured using the two items (Mitch­ 0.5508]), and brand attitude (b = 0.19, CI = [0.0481, 0.3909]). Thus,
ell, 1986) and used as a dependent variable. Harman’s one-factor test H2 was again supported.
confirmed the absence of common method bias (total variance extrac­ Next, Model 6 tested the serial mediation of perceived authenticity
ted: 39.76%). and expected quality simultaneously to answer H3. Fig. 3 and Table 2
shows the results of the serial mediation analysis on the effect of design
5.1.4. Covariates entity. The indirect effects of design entity serially through perceived
Brand attitude was not considered a covariate in Study 2 because authenticity and expected product quality were significant on product
Study 2 used a fictitious brand name, and the pre-existing attitude was attitude (b = 0.09, CI = [0.0242, 0.1898]), purchase intention (b = 0.13,
not a major concern (Schneider and Cornwell, 2005). Rather, brand CI = [0.0326, 0.2758]), and brand attitude (b = 0.10, CI = [0.0283,
attitude was entered as a dependent variable to examine if design entity 0.2080]). The sole mediation of perceived authenticity remained sig­
affects brand attitude. Brand familiarity (Batra et al., 2000; α = 0.98) nificant for product attitude and purchase intention, indicating that both
was still measured as a potential covariate to control potential influence perceived authenticity and the authenticity–product quality chain
of the feeling of knowing the brand. Next, perceived threats from AI mediate the effect of design entity on the two. Thus, H3 was supported.
(White et al., 2012; α = 0.96) were measured because the literature
suggested the feeling of being threatened by novel technology can make 5.3. Discussion
people negatively respond to outcomes produced by the technology
(Khasawneh, 2018). Lastly, product involvement (Zaichkowsky, 1985; Study 2 aimed to replicate Study 1 using a fictitious brand and a
α = 0.96) was measured as a potential covariate. This was because a product image. Study 2 showed consistent results with Study 1 such that
sweater image was used as a stimulus, and one’s interest levels in
sweaters might provoke confounding effects on the evaluation of the
products (Richins and Bloch, 1986). 1
The differences in perceived quality between AI-designed versus human-
designed sweaters were marginally significant when including covariates (p
= .064) but significant excluding covariates (p = .044). The authors interpret
the results controlling for covariates to rule out the effects of covariates.

6
G. Lee and H.-Y. Kim Journal of Retailing and Consumer Services 77 (2024) 103690

Fig. 3. Study 2 serial mediation analysis

Note. Dashed line shows non-significant path. *<0.05, **<0.01, ***<0.001.

6. Study 3
Table 2
Study 2 indirect effects.
6.1. Methods
Paths b 95% bootstrap CI

Product Attitude 6.1.1. Study design and procedure

Direct effect 0.08 (p = -0.1862 0.3541 A single-factor (AI design process: AI vs. AI customization) between-
.540)
subject online experiment was conducted to test the effect of providing
Design entity→authenticity→product attitude 0.19* 0.0549 0.3749
Design entity→product quality→product 0.00 -0.0817 0.0808
real-time customization options in the AI design process. One hundred
attitude and sixty U.S. participants fluent in English were recruited from Prolific,
Design entity→authenticity→product 0.09* 0.0242 0.1898 and 156 participants remained for data analysis (Mage = 37.50, SDage =
quality→product attitude 13.26; 38.5% male; see Appendix B and C). The AI design process was
Purchase Intention
manipulated through shopping scenarios and customization tasks. Par­
Direct effect -0.28 (p = -0.5878 0.0253
.072) ticipants were randomly assigned to one of the two conditions. Those in
Design entity→authenticity→purchase 0.18* 0.0379 0.3776 the AI condition read a scenario and viewed a product image. The sce­
intention nario told them AI designed a sweater. Participants in the AI custom­
Design entity→product quality→purchase 0.01 -0.1076 0.1252 ization condition read a scenario telling them AI would design a sweater
intention
Design entity→authenticity→product 0.13* 0.0326 0.2758
based on their input. They were given options to select the sweater’s
quality→purchase intention color and pattern, and a product image based on their input was then
Brand Attitude provided to them. This approach was adopted to ensure internal validity
Direct effect 0.15 (p = -0.1483 0.4454 by minimizing the potential differences between diverse stimuli caused
.324)
by user-typed prompts. The questionnaire and a diagrammatic repre­
Design entity→authenticity→brand attitude 0.08 -0.0040 0.2058
Design entity→product quality→brand 0.01 -0.0868 0.0975 sentation of study procedure are available in Appendix D.
attitude
Design entity→authenticity→product 0.10* 0.0283 0.2080 6.1.2. Stimuli
quality→brand attitude The sweater image used in Study 2 was used in Study 3. Sixteen
Note. Asterisk (*) indicates significant effect. variations of the sweater image were developed to reflect the design
options in the AI customization condition: four colors (light gray, gray,
participants formed more favorable attitudes toward human-designed light blue, and blue) and four patterns (no pattern; plain, argyle, stipe,
(vs. AI-designed) clothing because of perceived authenticity (H1, H2). and dot). To avoid potential confounding effects of sweater design be­
The effect was also significant on brand attitude, but not on purchase tween study conditions and increase internal validity, the same set of
intention or expected product quality. In contrast to the scenario-based sweater images available in the AI customization condition was used in
experiment in Study 1, the insignificant differences in purchase inten­ the AI condition to show a randomly selected image.
tion between the study conditions may be due to the product image
stimuli used in Study 2. Additionally, Study 2 confirmed the serial 6.1.3. Measurements
mediation of perceived authenticity and expected product quality be­ The scales used in Study 2 were used to measure perceived authen­
tween design entity (AI vs. human) and the three consumer responses ticity, expected product quality, product attitude, purchase intention,
(H3). Although the perceived authenticity–product quality chain and brand attitude (see Table 1). Additionally, perceived automation
explained the effect of design entity, the mediation of perceived and customization in design were measured to check similar perceived
authenticity still remained significant. Next, Study 3 examines the role automation and different perceived customization between the study
of AI customization in alleviating consumers’ negative responses to AI- conditions. Participants indicated perceived automation in design by
designed clothing by testing the effect of AI design process. specifying the extent to which the design process seemed more like (1 =
mostly by computer—10 = mostly by humans). Perceived customization
was measured by employing the items from Srinivasan et al.’s (2002)
study on a 7-point Likert scale (α = 0.93): “This sweater design was
customized to my needs,” “I believe that this sweater design was
customized to my characteristics,” and “This sweater was designed
based on my input.” Harman’s one-factor test confirmed the absence of

7
G. Lee and H.-Y. Kim Journal of Retailing and Consumer Services 77 (2024) 103690

common method bias (total variance extracted: 34.70%). indicating that the effects of AI design process are partially mediated by
perceived authenticity and the authenticity–product quality chain.
6.1.4. Covariates Thus, H3 is supported in the context of AI design process.
Similar to Study 2, brand familiarity (α = 0.98), perceived threats
from AI (α = 0.92), and product involvement (α = 0.95) were measured 6.3. Discussion
as the potential covariates using the same measurements.
Study 3 confirmed the role of AI customization in mitigating the
6.2. Results negative effect of AI as design entity by comparing AI versus AI cus­
tomization (H4). Participants’ negative attitudes and purchase in­
6.2.1. Measurement testing tentions toward AI-designed clothing were mitigated when they
The measurement model fit was adequate (χ2 = 264.01, df = 141; χ2/ provided their input to customize the product design. While expected
df = 1.872; CFI = 0.97; NNFI = 0.96; RMSEA = 0.07; SRMR = 0.05; see product quality did not differ, Study 3 confirmed the mediation of
Appendix E). Construct validity was confirmed as the factor loadings perceived authenticity (H2) and the serial mediation of perceived
were high and significant at a p-value of .001 (see Table 1). Discriminant authenticity and expected product quality (H3) when comparing AI
validity was achieved as all AVE values of the constructs were larger versus AI customization. In line with Study 2, the mediation of perceived
than the squared correlations. Cronbach’s alpha (α>0.88) and com­ authenticity remained significant when the perceived authenticity–­
posite reliability (CR>0.88) supported measurement reliability. product quality chain was entered. Unexpectedly, the direct effect of AI
design process on the responses was found to be significant, suggesting
6.2.2. Effect of AI design process that a third factor may explain such a relationship. The following sec­
There were no differences in perceived automation between the two tions present a general discussion of the study findings along with
study conditions (p = .487). Compared to those in the AI condition, theoretical and practical implications.
participants in the AI customization condition believed more that the
sweater was designed with customization (M = 2.62 vs. 5.13; t = -11.31, 7. General discussion
p < .001). Thus, the manipulation of AI design process was successful.
MANCOVA was performed with brand familiarity, perceived threats The huge progress in generative AI, an algorithm-based computer
from AI, and product involvement as covariates. The three covariates system that enables new content creation, is clearly bringing revolutions
showed significant correlations with at least one dependent variable and to retail and consumer environment. Companies adopt AI-assisted
showed a significant effect (perceived threats, product involvement: p < product design processes to optimize overall operations and improve
.001; brand familiarity: p = .003). The statistical assumptions were met efficiency while better meeting market demand. To provide a theory-
(see Appendix F). The main effect of the AI design process was signifi­ based understanding of AI-designed clothing from the consumer psy­
cant (Pillai’s Λ = 0.18, Wilks’ λ = 0.82, p < .001, η2 = 0.18). The results chology perspective, this study investigated how consumers respond to
were generally sustained sans covariates, showing minimal changes in AI-designed fashion products, focusing on clothing.
statistics values2 (see Appendix G). Participants in the AI customization The findings indicate that consumers generally form negative re­
(vs. AI) condition perceived higher authenticity (MAI = 3.91 vs. MAI­ sponses to AI-designed (vs. human-designed) clothing. This pattern is
2 attributed to low perceived authenticity from AI, which is based on the
custom = 4.32; F = 4.81, p = .030, η = 0.03). Also, participants in the AI
customization (vs. AI) condition formed higher product attitude (MAI = large discrepancy between consumers’ preexisting fashion design
3.84 vs. MAIcustom = 4.70; F = 12.65, p < .001, η2 = 0.08), purchase schema and AI. Further, the authenticity–product quality chain also
intention (MAI = 2.52 vs. MAIcustom = 3.87; F = 29.79, p < .001, η2 = explains consumer responses to fashion products designed by AI versus
0.17), and brand attitude (MAI = 4.05 vs. MAIcustom = 4.65; F = 9.18, p = humans. AI versus humans as design entity does not always lead to
.003, η2 = 0.06). Expected product quality between the two conditions significantly different expected product quality. Still, they serially
did not differ (p = .353). Thus, H4 was supported for all variables except explain the effect on consumer responses because perceived authenticity
for expected quality. positively affects expected product quality. Finally, the findings confirm
the role of AI customization. Although consumers form negative bias
6.2.3. Mediation regarding the authenticity of AI-designed clothing, such bias against AI-
PROCESS Model 4 was performed with brand familiarity, perceived assisted design processes can be attenuated through the option to
threats from AI, and product involvement as covariates. The indirect customize their own product design. Consequently, consumers’ attitudes
effects of the AI design process through perceived authenticity were and purchase intentions toward clothing designed through AI custom­
significant on product attitude (b = 0.32, CI = [0.0362, 0.6244]), pur­ ization become more positive compared to non-AI-customization.
chase intention (b = 0.28, CI = [0.0334, 0.5444]), and brand attitude (b Perceived authenticity increased through AI customization in turn im­
= 0.28, CI = [0.0261, 0.5465]). Thus, the mediation of perceived proves expected product quality, making them serially explain the effect
authenticity was confirmed in the context of AI design process. of AI customization.
Model 6 was performed to test the serial mediation. Fig. 4 and
Table 3 shows the results of the serial mediation analysis. The indirect 7.1. Implications
effects of AI design process through perceived authenticity and expected
quality were significant on product attitude (b = 0.06, CI = [0.0015, This study contributes to the literature by providing theoretical im­
0.1445]), purchase intention (b = 0.06, CI = [0.0032, 0.1488]), and plications on consumer responses to fashion products designed through
brand attitude (b = 0.07, CI = [0.0067, 0.1639]). The sole mediation of the AI-assisted process. First, this study helps advance the knowledge of
perceived authenticity still remained large and significant for all three AI-assisted design processes in the consumer environment by demon­
dependent variables. In addition, the direct effects of AI design process strating consumers’ negative biases against to AI-designed (human-
on the responses were significant after specifying the serial mediations, designed) clothing. These findings align with the studies that human-
generated (vs. AI-generated) fashion design is more favorably evalu­
ated (Lee, 2022; Xu and Mehta, 2022). Furthermore, extending the
2
The differences in perceived authenticity of the sweater designed by AI findings to marketing outcomes, the current study found that the
versus AI customization were significant when including covariates (p = .030) negative responses to AI-designed clothing may or may not impact sales
but marginal excluding covariates (p = .073). The authors interpret the results revenue by lowering purchase intention. These findings have the po­
controlling for covariates to rule out the effects of covariates. tential to describe the discrepancy in the literature that consumers form

8
G. Lee and H.-Y. Kim Journal of Retailing and Consumer Services 77 (2024) 103690

Fig. 4. Study 3 serial mediation analysis

Note. Dashed line shows non-significant path. *<0.05, **<0.01, ***<0.001.

authenticity and expected product quality also explain the benefits of AI

Table 3
customization.
Study 3 indirect effects.
Third, this study contributes to the question of how negative bias
Paths b 95% bootstrap CI against AI-designed clothing can be attenuated. The findings empirically
Product Attitude reveal that AI customization, the process where AI generates designs
Direct effect 0.54* (p 0.1668 0.9128 based on consumers’ real-time personalization of their product design, is
= .005)
effective in raising perceived authenticity and thus drawing more pos­
AI design process→authenticity→product 0.27* 0.0236 0.5272
attitude
itive consumer attitudes and intentions. These findings extend the pre­
AI design process→product quality→product -0.01 -0.1005 0.0876 vious research on the benefits of mass customization or customer
attitude involvement (Moreau and Herd, 2010; Nishikawa et al., 2013; Stevens
AI design process→authenticity→product 0.06* 0.0015 0.1445 et al., 2017) to AI design process. Notably, the relationship between
quality→product attitude
personalization and perceived authenticity is proven in the context of AI
Purchase Intention
Direct effect 1.09* (p 0.6650 1.5054 customization.
< .001) The findings have important implications for practitioners who
AI design process→authenticity→purchase 0.22* 0.0278 0.4582 adopt generative AI in the product design process. First, practitioners
intention can have insights into how and why consumers would negatively eval­
AI design process→product quality→purchase -0.01 -0.1076 0.0939
intention
uate AI-designed clothing compared to human-designed clothing.
AI design process→authenticity→product 0.06* 0.0032 0.1488 Considering that fashion design schema is established based on previ­
quality→purchase intention ously accumulated information revolving around fashion design, it may
Brand Attitude be beneficial to first target the young generations who are deemed
Direct effect 0.34* (p 0.0469 0.4454
digital natives and more familiar with advanced technology when pro­
= .023)
AI design process→authenticity→brand 0.21* 0.0267 0.3992 moting AI-assisted design. Rather than pursuing a short-term increase in
attitude sales through AI-assisted design processes, the long-term business goal
AI design process→product quality→brand -0.01 -0.1196 0.1055 of helping consumers become more familiar with AI as a design entity
attitude will be needed. More importantly, as perceived authenticity and ex­
AI design process→authenticity→product 0.07* 0.0067 0.1639
quality→brand attitude
pected product quality explain the negative evaluations, practitioners
can focus on enhancing and emphasizing the uniqueness of fashion
Note. Asterisk (*) indicates significant effect. design generated through AI-assisted design processes.
Second, this study offers practitioners valuable insights regarding the
negative responses but may be willing to pay more due to its novelty for benefits of AI customization for making consumers more favorably
AI-designed products. While it is beyond the current study’s scope, it respond to fashion products designed with AI-assisted processes. AI
will be interesting if future research explores which design aspects of AI customization is suggested as a customer-centric way to better align
versus human-designed clothing differentially affects purchase intention with the consumers’ ever-growing and changing needs while sustaining
and willingness to pay more. the fashion design innovation driven by generative AI. Practitioners can
Second, this study unveils the mechanisms under the negative re­ maintain computer-aided customization systems that are available to
sponses to AI-designed clothing building on the schema theory. Because individual consumers to compensate for the decreased perceived
consumers’ schemata about fashion design are mainly associated with authenticity and mitigate negative responses to AI-designed clothing.
human work and traits, clothing designed by AI that uses existing data is For example, brands can produce designated product lines exploiting a
perceived as less genuine or authentic. Consequently, perceived range of design options that consumers can use for their AI custom­
authenticity formed by the information about design entity shapes ex­ ization and increase the perceived authenticity of the AI-customized
pected product quality and thus explains negative attitudes and pur­ product. Nevertheless, further efforts will be inevitable to overcome
chase intentions. Extending the previous findings on schema, the challenges in adapting the AI customization as the production dif­
authenticity, and product quality perception to AI-assisted design pro­ ficulty increases with customization (Aggarwal et al., 2020).
cesses (e.g., Bem, 1981; Cinelli and LeBoeuf, 2020; Halkias and Kokki­ Some implications for policymakers regarding AI design can be
naki, 2014; Meyers-Levy and Tybout, 1989; Rao and Monroe, 1989), the considered based on the findings and the topic of this research, although
current study shows that fashion design schemata explain how and why the study findings provide meaningful implications mainly for re­
consumers evaluate AI-designed clothing. Furthermore, perceived searchers and practitioners. First, a guideline that companies utilizing AI

9
G. Lee and H.-Y. Kim Journal of Retailing and Consumer Services 77 (2024) 103690

design clearly disclose the use of AI is deemed to be important. Since CRC Press, pp. 129–141. https://doi-org.ezp2.lib.umn.edu/10.1201/97810030
32410.
consumers do not see AI-designed clothing as sincere and authentic as
Alimamy, S., Kuhail, M.A., 2023. I will be with you Alexa! The impact of intelligent
human-designed clothing, not sharing information about AI usage in the virtual assistant’s authenticity and personalization on user reusage intentions.
product design process can lead to a potential discussion about con­ Comput. Hum. Behav. 143, 107711 https://doi.org/10.1016/j.chb.2023.107711.
sumer deception. Second, as AI customization options become more Appel, G., Neelbauer, J., Schweidel, D.A., 2023. Generative AI has an intellectual
property problem. Harvard Business Review. https://hbr.org/2023/04/generative-
accessible to consumers, a policy to monitor consumers’ unethical AI ai-has-an-intellectual-property-problem.
usage for product design can be discussed. Nevertheless, future research Audrezet, A., De Kerviler, G., Moulard, J.G., 2020. Authenticity under threat: when social
is needed to dig into the policy implications as the objective of the media influencers need to go beyond self-presentation. J. Bus. Res. 117, 557–569.
https://doi.org/10.1016/j.jbusres.2018.07.008.
current study is not to provide a full, detailed picture for policy Bang, Y., Cahyawijaya, S., Lee, N., Dai, W., Su, D., Wilie, B., et al., 2023. A multitask,
implications. multilingual, multimodal evaluation of ChatGPT on reasoning, hallucination, and
interactivity. arXiv preprint arXiv:2302.04023. https://doi.org/10.48550/
arXiv.2302.04023.
7.2. Limitations and future research Batra, R., Ramaswamy, V., Alden, D.L., Steenkamp, J.B.E., Ramachander, S., 2000.
Effects of brand local and nonlocal origin on consumer attitudes in developing
countries. J. Consum. Psychol. 9 (2), 83–95. https://doi.org/10.1207/
The current study is among the first to conduct a theory-based study
S15327663JCP0902_3.
exploring consumer responses to AI-designed fashion products. Still, Bem, S.L., 1981. Gender schema theory: a cognitive account of sex typing. Psychol. Rev.
several study limitations remain, which can be considered in future 88 (4), 354–364. https://doi.org/10.1037/0033-295X.88.4.354.
research. First, AI customization design process can be operationalized Beverland, M.B., Farrelly, F.J., 2010. The quest for authenticity in consumption:
consumers’ purposive choice of authentic cues to shape experienced outcomes.
in different ways other than customizing colors and patterns for J. Consum. Res. 36 (5), 838–856. https://doi.org/10.1086/615047.
sweaters. Arising platforms for product customization using generative Boush, D.M., Loken, B., 1991. A process-tracing study of brand extension evaluation.
AI provide diverse customization options to incorporate customers’ J. Market. Res. 28 (1), 16–28. https://doi.org/10.2307/3172723.
Campagna, C.L., Donthu, N., Yoo, B., 2023. Brand authenticity: literature review,
input, including selecting or typing in not only certain design compo­ comprehensive definition, and an amalgamated scale. J. Market. Theor. Pract. 31
nents (e.g., colors or patterns) but also materials (e.g., cotton or linen), (2), 129–145. https://doi.org/10.1080/10696679.2021.2018937.
trims (e.g., buttons), features (e.g., pockets), or adjectives (e.g., fash­ Carsana, L., Jolibert, A., 2018. Influence of iconic, indexical cues, and brand schematicity
on perceived authenticity dimensions of private-label brands. J. Retailing Consum.
ionable). Further, this study used a sweater image to control product Serv. 40, 213–220. https://doi.org/10.1016/j.jretconser.2017.10.006.
category effects. It will be interesting to test how perceptions of AI- Chae, B., Hoegg, J., 2013. The future looks “right”: effects of the horizontal location of
designed clothing vary depending on product categories (e.g., coat, advertising images on product attitude. J. Consum. Res. 40 (2), 223–238. https://
doi.org/10.1086/669476.
dress, leggings). Chamberlain, R., Mullin, C., Scheerlinck, B., Wagemans, J., 2018. Putting the art in
Second, the current study developed the hypotheses building upon artificial: aesthetic responses to computer-generated art. Psychology of Aesthetics,
schema theory, focusing on the influence of schemata on product per­ Creativity, and the Arts 12 (2), 177–192. https://doi.org/10.1037/aca0000136.
Chaudhuri, A., 2002. How brand reputation affects the advertising-brand equity link.
ceptions (i.e., perceived authenticity and product quality). Given that
J. Advert. Res. 42 (3), 33–43. https://doi.org/10.2501/JAR-42-3-33-43.
this study did not directly examine how AI is perceived as a design entity Chu, S.C., Deng, T., Mundel, J., 2022. The impact of personalization on viral behavior
from schema perspectives because it was out of scope, future research intentions on TikTok: the role of perceived creativity, authenticity, and need for
can explore factors resulting from one’s schemata that may affect per­ uniqueness. J. Market. Commun. 1–20. https://doi.org/10.1080/
13527266.2022.2098364.
ceptions of AI itself working for clothing design. It will be valuable to Cinelli, M.D., LeBoeuf, R.A., 2020. Keeping it real: how perceived brand authenticity
test whether consumers’ particular AI schemata affect how they respond affects product perceptions. J. Consum. Psychol. 30 (1), 40–59. https://doi.org/
to AI as a design entity. 10.1002/jcpy.1123.
Davvetas, V., Diamantopoulos, A., 2016. How product category shapes preferences
Lastly, Study 3 yielded interesting findings that the effects of AI toward global and local brands: a schema theory perspective. J. Int. Market. 24 (4),
design process are partially mediated by perceived authenticity and the 61–81. https://doi.org/10.1509/jim.15.0110.
authenticity–product quality chain and that the direct effect of AI design DeLong, M.R., Minshall, B., Larntz, K., 1986. Use of schema for evaluating consumer
response to an apparel product. Cloth. Text. Res. J. 5 (1), 17–26. https://doi.org/
process on consumer responses still remains. While this study does not 10.1177/0887302X8600500103.
entirely unveil which factor explains different responses to clothing Dodds, W.B., Monroe, K.B., Grewal, D., 1991. Effects of price, brand, and store
designed by AI versus AI customization, these findings call for future information on buyers’ product evaluations. J. Market. Res. 28 (3), 307–319.
https://doi.org/10.2307/3172866.
studies employing diverse perspectives. Figoli, F.A., Rampino, L., Mattioli, F., 2022. AI in design idea development: a workshop
on creativity and human-AI collaboration. In: Proceedings of the DRS2022
Funding Conference. https://doi.org/10.21606/drs.2022.414.
Fiske, S.T., 1982. Schema-triggered affect: applications to social perception. In: Affect
and Cognition: 17th Annual Carnegie Mellon Symposium on Cognition. Lawrence
This work was supported by College of Design, University of Erlbaum, Hillsdale, pp. 55–78.
Minnesota. Geuens, M., De Pelsmacker, P., 2017. Planning and conducting experimental advertising
research and questionnaire design. J. Advert. 46 (1), 83–100. https://doi.org/
10.1080/00913367.2016.1225233.
Declaration of competing interest Gilmore, J.H., Pine, B.J., 2007. Authenticity: what Consumers Really Want. Harvard
Business School Press, Cambridge, MA.
Guo, Z., Zhu, Z., Li, Y., Cao, S., Chen, H., Wang, G., 2023. AI Assisted Fashion Design: A
None. Review. IEEE Access. https://10.1109/ACCESS.2023.3306235.
Ha, S., Huang, R., Park, J.S., 2019. Persuasive brand messages in social media: a mental
Data availability imagery processing perspective. J. Retailing Consum. Serv. 48, 41–49. https://doi.
org/10.1016/j.jretconser.2019.01.006.
Halkias, G., Kokkinaki, F., 2014. The degree of ad–brand incongruity and the distinction
Data will be made available on request. between schema-driven and stimulus-driven attitudes. J. Advert. 43 (4), 397–409.
https://doi.org/10.1080/00913367.2014.891087.
Harper, S., 2023. ChatGPT Threatens Authenticity of DEI Communications from Leaders.
Appendix A. Supplementary data Forbes. https://www.forbes.com/sites/shaunharper/2023/02/19/chatgpt-will-wea
ken-authenticity-of-executive-communication-about-dei/?sh=2a06f8974d85.
Supplementary data to this article can be found online at https://doi. Harreis, H., Koullias, T., Roberts, R., Te, K., 2023. Generative AI: Unlocking the Future of
Fashion. McKinsey & Company. https://www.mckinsey.com/industries/retail/our
org/10.1016/j.jretconser.2023.103690. -insights/generative-ai-unlocking-the-future-of-fashion.
Hong, J.W., Curran, N.M., 2019. Artificial intelligence, artists, and art: attitudes toward
References artwork produced by humans vs. artificial intelligence. ACM Trans. Multimed
Comput. Commun. Appl 15 (2s), 1–16. https://doi.org/10.1145/3326337.
Jago, A.S., 2019. Algorithms and authenticity. Acad. Manag. Discov. 5 (1), 38–56.
Aggarwal, S., Bhardwaj, P., Arora, J., 2020. AI in fashion: present and future
https://doi.org/10.5465/amd.2017.0002.
applications. In: Transforming Management Using Artificial Intelligence Techniques.

10
G. Lee and H.-Y. Kim Journal of Retailing and Consumer Services 77 (2024) 103690

Khasawneh, O.Y., 2018. Technophobia: examining its hidden factors and defining it. Rajagopal, P., Montgomery, N., 2012. Remembering better or remembering worse: age
Technol. Soc. 54, 93–100. https://doi.org/10.1016/j.techsoc.2018.03.008. effects on false memory. In: Gürhan-Canli, Zeynep, Otnes, Cele, Zhu, Rui (Juliet)
Kim, J., Kim, J.H., Kim, C., Park, J., 2023. Decisions with ChatGPT: reexamining choice (Eds.), Advances in Consumer Research, 40. Association for Consumer Research,
overload in ChatGPT recommendations. J. Retailing Consum. Serv. 75, 103494 Duluth, MN, 928-928.
https://doi.org/10.1016/j.jretconser.2023.103494. Rao, A.R., Monroe, K.B., 1989. The effect of price, brand name, and store name on
Lamb, J.M., Kallal, M.J., 1992. A conceptual framework for apparel design. Cloth. Text. buyers’ perceptions of product quality: an integrative review. J. Market. Res. 26 (3),
Res. J. 10 (2), 42–47. https://doi.org/10.1177/0887302X9201000207. 351–357. https://doi.org/10.2307/3172907.
Lee, J., Kim, H., 2022. How to survive in advertisement flooding: the effects of Richins, M.L., Bloch, P.H., 1986. After the new wears off: the temporal context of product
schema–product congruity and attribute relevance on advertisement attitude. involvement. J. Consum. Res. 13 (2), 280–285. https://doi.org/10.1086/209067.
J. Consum. Behav. 21 (2), 214–230. https://doi.org/10.1002/cb.1991. Schneider, L.P., Cornwell, T.B., 2005. Cashing in on crashes via brand placement in
Lee, Y.K., 2022. How complex systems get engaged in fashion design creation: using computer games: the effects of experience and flow on memory. Int. J. Advert. 24
artificial intelligence. Think. Skills Creativ. 46, 101137 https://doi.org/10.1016/j. (3), 321–343. https://doi.org/10.1080/02650487.2005.11072928.
tsc.2022.101137. Singh, S.N., Cole, C.A., 1993. The effects of length, content, and repetition on television
Li, H., Lei, Y., Zhou, Q., Yuan, H., 2023. Can you sense without being human? Comparing commercial effectiveness. J. Market. Res. 30 (1), 91–104. https://doi.org/10.2307/
virtual and human influencers endorsement effectiveness. J. Retailing Consum. Serv. 3172516.
75, 103456 https://doi.org/10.1016/j.jretconser.2023.103456. Sohn, K., Sung, C.E., Koo, G., Kwon, O., 2020. Artificial intelligence in the fashion
Littrell, M., Anderson, L.F., Brown, P., 1993. What makes a craft souvenir authentic? industry: consumer responses to generative adversarial network (GAN) technology.
Ann. Tourism Res. 20 (1), 197–215. https://doi.org/10.1016/0160-7383(93)90118- Int. J. Retail Distrib. Manag. 49 (1), 61–80. https://doi.org/10.1108/IJRDM-03-
M. 2020-0091.
Mandler, G., 1981. The Structure of Value: Accounting for Taste. Center for Human Srinivasan, S.S., Anderson, R., Ponnavolu, K., 2002. Customer loyalty in e-commerce: an
Information Processing, Department of Psychology, University of California, San exploration of its antecedents and consequences. J. Retailing 78 (1), 41–50. https://
Diego, pp. 3–36. doi.org/10.1016/S0022-4359(01)00065-3.
Meyers-Levy, J., Tybout, A.M., 1989. Schema congruity as a basis for product evaluation. Stevens, J., Esmark, C.L., Noble, S.M., Lee, N.Y., 2017. Co-producing with consumers:
J. Consum. Res. 16 (1), 39–54. https://doi.org/10.1086/209192. how varying levels of control and co-production impact affect. Market. Lett. 28,
Mitchell, A.A., 1986. The effect of verbal and visual components of advertisements on 171–187. https://doi.org/10.1007/s11002-016-9413-2.
brand attitudes and attitude toward the advertisement. J. Consum. Res. 13 (1), Sujan, M., Bettman, J.R., 1989. The effects of brand positioning strategies on consumers’
12–24. https://doi.org/10.1086/209044. brand and category perceptions: some insights from schema research. J. Market. Res.
Mitchell, V.W., Balabanis, G., 2021. The role of brand strength, type, image and product- 26 (4), 454–467. https://doi.org/10.2307/3172765.
category fit in retail brand collaborations. J. Retailing Consum. Serv. 60, 102445 Sweeney, J.C., Soutar, G.N., Johnson, L.W., 1999. The role of perceived risk in the
https://doi.org/10.1016/j.jretconser.2021.102445. quality-value relationship: a study in a retail environment. J. Retailing 75 (1),
Moreau, C.P., Herd, K.B., 2010. To each his own? How comparisons with others 77–105. https://doi.org/10.1016/S0022-4359(99)80005-0.
influence consumers’ evaluations of their self-designed products. J. Consum. Res. 36 White, K., Argo, J.J., Sengupta, J., 2012. Dissociative versus associative responses to
(5), 806–819. https://doi.org/10.1086/644612. social identity threat: the role of consumer self-construal. J. Consum. Res. 39 (4),
Moreau, C.P., Prandelli, E., Schreier, M., Hieke, S., 2020. Customization in luxury 704–719. https://doi.org/10.1086/664977.
brands: can Valentino get personal? J. Market. Res. 57 (5), 937–947. https://doi. Xian, W., Sangkloy, P., Agrawal, V., Raj, A., Lu, J., Fang, C., et al., 2018. Texturegan:
org/10.1177/0022243720943191. controlling deep image synthesis with texture patches. In: Proceedings of the IEEE
Moulard, J.G., Raggio, R.D., Folse, J.A.G., 2016. Brand authenticity: testing the Conference on Computer Vision and Pattern Recognition, pp. 8456–8465. https://
antecedents and outcomes of brand management’s passion for its products. Psychol. doi.org/10.48550/arXiv.1706.02823.
Market. 33 (6), 421–436. https://doi.org/10.1002/mar.20888. Xu, L., Mehta, R., 2022. Technology devalues luxury? Exploring consumer responses to
Moulard, J.G., Rice, D.H., Garrity, C.P., Mangus, S.M., 2014. Artist authenticity: how AI-designed luxury products. J. Acad. Market. Sci. 50, 1135–1152. https://doi.org/
artists’ passion and commitment shape consumers’ perceptions and behavioral 10.1007/s11747-022-00854-x.
intentions across genders. Psychol. Market. 31 (8), 576–590. https://doi.org/ Yan, H., Zhang, H., Liu, L., Zhou, D., Xu, X., Zhang, Z., Yan, S., 2022. Toward intelligent
10.1002/mar.20719. design: an AI-based fashion designer using generative adversarial networks aided by
Murphy, K.P., 2022. Probabilistic Machine Learning: an Introduction. MIT press. sketch and rendering generators. IEEE Trans. Multimed. https://doi.org/10.1109/
Napoli, J., Dickinson, S.J., Beverland, M.B., Farrelly, F., 2014. Measuring consumer- TMM.2022.3146010.
based brand authenticity. J. Bus. Res. 67 (6), 1090–1098. https://doi.org/10.1016/j. Yu, J., Dickinger, A., So, K.K.F., Egger, R., 2023. Artificial intelligence-generated virtual
jbusres.2013.06.001. influencer: examining the effects of emotional display on user engagement.
Newman, G.E., Dhar, R., 2014. Authenticity is contagious: brand essence and the original J. Retailing Consum. Serv. 76, 103560 https://doi.org/10.1016/j.
source of production. J. Market. Res. 51 (3), 371–386. https://doi.org/10.1509/ jretconser.2023.103560.
jmr.11.0022. Zaichkowsky, J.L., 1985. Measuring the involvement construct. J. Consum. Res. 12 (3),
Nishikawa, H., Schreier, M., Ogawa, S., 2013. User-generated versus designer-generated 341–352. https://doi.org/10.1086/208520.
products: a performance assessment at Muji. Int. J. Res. Market. 30 (2), 160–167. Zhang, H., Bai, X., Ma, Z., 2022. Consumer reactions to AI design: exploring consumer
https://doi.org/10.1016/j.ijresmar.2012.09.002. willingness to pay for AI-designed products. Psychol. Market. 39 (11), 2171–2183.
Pieconka, J., 2023. Creative Machines-Consumers’ Perception Of AI Designed Products https://doi.org/10.1002/mar.21721.
(Doctoral Dissertation, Leopold-Franzens-Universität Innsbruck). Zhu, Y., You, J., Alard, R., Schönsleben, P., 2009. Design quality: a key to improve
Podsakoff, P.M., Organ, D.W., 1986. Self-reports in organizational research: problems product quality in international production network. Prod. Plann. Control 20 (2),
and prospects. J. Manag. 12 (4), 531–544. https://doi.org/10.1177/ 168–177. https://doi.org/10.1080/09537280802705062.
014920638601200408.

11