The Rise of Deepfakes: A conceptual framework and research agenda for marketing

Lucas Whittakera*, Kate Letherenb, Rory Mulcahyc

a
School of Advertising, Marketing and Public Relations, Queensland University of

Technology, 2 George Street, Brisbane City, 4000, Queensland, Australia. Email:

ll.whittaker@qut.edu.au

b
School of Advertising, Marketing and Public Relations, Queensland University of

Technology, 2 George Street, Brisbane City, 4000, Queensland, Australia. Email:

kate.letheren@qut.edu.au

c
USC Business School, University of the Sunshine Coast, Sippy Downs Drive,

Maroochydore DC, 4558, Queensland, Australia. Email: rmulcahy@usc.edu.au

*
corresponding author

1
 The Rise of Deepfakes: A conceptual framework and research agenda for
marketing
Abstract

Deepfakes, digital content created via machine learning, a form of artificial intelligence

technology, are generating interest amongst marketers and the general population alike and

are often portrayed as a ‘phantom menace’ in the media. Despite relevance to marketing

theory and practice, deepfakes – and the opportunities for benefit or deviance they provide –

are little understood or discussed. This paper introduces deepfakes to the marketing literature

and proposes a typology, conceptual framework, and associated research agenda,

underpinned by theorizing based on balanced centricity, to guide the future investigation of

deepfakes in marketing scholarship. The paper makes an argument for balance (i.e. situations

where all stakeholders benefit) and it is hoped that this paper may provide a foundation for

future research and application of deepfakes as ‘a new hope’ for marketing.

Keywords: Artificial intelligence, deepfakes, generative adversarial networks, machine

learning, synthetic media, balanced centricity

1. Introduction

In 2017, there was a considerable debate regarding the digital resurrection of Carrie

Fisher (the actor who portrays Princess Leia) in the Star Wars: The Rise of Skywalker movie,

due to her passing before shooting (Lowry, 2017). To retain authenticity, Carrie’s daughter,

Billie Lourd, acted as a stand-in actor during shooting and Carrie Fisher’s facial and voice

features were digitally overlaid. Such digital manipulation of visual and audio is not just

limited to Hollywood blockbusters, and is becoming increasingly facilitated by artificial

intelligence (AI) technology. This phenomenon, known as the “deepfake”, is a new form of

AI-enabled content creation which is being implemented by marketers and customers. This

2
conceptual paper sets out to understand how AI has led to “The Rise of Deepfakes” and

discusses the relevance and need for future marketing scholarship in this area. This paper also

synthesizes literature from marketing, information science, and communication relating to

deepfakes and AI to develop a conceptual model and research agenda.

AI is a growing industry which is expected to reach $118 billion USD by 2025

(Tractica, 2019), and AI-enabled robots are already amongst us, in sectors such as retail,

foodservice, tourism, and health (Mende, Scott, van Doorn, Grewal, & Shanks, 2019). Yet it

is not only the speed of development that is unprecedented; it is also how the development is

occurring. Given machine learning technologies and collaboration opportunities over the

internet, development can sometimes be shared, or even contributed to by the AI itself if

autonomous machine learning functionalities have been programmed in by trained

information scientists. AI is a complex technology with broad applications across multiple

fields. For instance, AI can be used to enhance algorithms (for example, in AI-enabled

biometrics for recruiting; van Esch, Black, Franklin, & Harder, 2020), to provide the ‘mind’

inside embodied robots and virtual assistants, and to detect patterns and provide advice in

daily life – such as in share market trading. Synthetic media is also an application of AI and

describes automatically and artificially generated media that blur the lines between real and

fake (Whittaker, Kietzmann, Kietzmann, & Dabirian, 2020). Such applications are not

necessarily malicious. Indeed, customers already interact with AI such as chatbots and virtual

assistants. However, what remains to be seen is whether customers are accepting of deepfake

technologies, which may obscure the fact that seemingly ‘real’ content was created via AI.

There have been some initial discussions into the emergent AI-based deepfake

technology. Deepfakes are the product of AI and the ‘deep learning’ machine learning

technique which trains deep neural networks. These networks can be thought of as a brain

containing many interconnected neurons, with these artificial neurons being referred to as

3
‘units’. Though each of these units complete a simple computation, the multitude of units

within a deep neural network together can perform complex tasks (Kietzmann, McClure, &

Kriegeskorte, 2019). These deep neural networks can be trained to artificially generate and

manipulate video, images, and audio to create hyper-realistic fake content (Kietzmann, Lee,

McCarthy, & Kietzmann, 2020). This example demonstrates how deepfakes can be viewed as

a type of synthetic media as the realistic deepfake output is automatically and artificially

generated (Whittaker et al., 2020). With deepfakes, it is difficult to tell if the person in the

video is real or fake, or alternatively for customers to realize this is a question they should be

asking. There are numerous marketing opportunities for the application of deepfakes bringing

(a new) hope that deepfakes can be used for the benefit of both customers and business –

offering stronger engagement across cultural boundaries, enhanced messaging

personalization and effectiveness, and even opportunities for customer-led co-creation.

However, the potential of deepfakes as a ‘phantom menace’ should also not be ignored.

Indeed, consent and awareness are two central issues with deepfakes as the potential for

deviance is evident – consider the hijacking of political or marketing campaigns, criminal

activities like phishing or revenge pornography, or even the potential for psychological harm

stemming from the fluidity deepfakes seemingly grant to mortality. Despite deepfakes

possessing a light and dark side, marketing scholarship has not yet thoroughly theorized or

discussed these perspectives. Thus, this paper seeks to introduce deepfakes and theorize

directions for future research to explore the light and dark side of deepfakes.

The purpose of this paper is therefore threefold: first, to define and introduce

deepfakes to the marketing literature, second, to differentiate deepfakes from existing

marketing practices to highlight their unique contribution to the discipline; and third, to

provide a conceptual framework and research agenda to guide marketing scholars’

investigation of this new phenomenon. The introduction of deepfakes, alongside a typology

4
and conceptual framework to the marketing literature is important for two main reasons.

First, the creation of authentic, persuasive content is a key aim of marketing – deepfakes offer

a new way of providing such content, but as yet this technology is little understood outside of

information science and has not yet been integrated with marketing theory and differentiated

from existing marketing practice. Second, as with any technology that is not yet well

understood, there is potential for unexpected – and perhaps immoral – uses. Hence, marketers

need to know more about potential opportunities for deviance utilizing deepfakes, both on the

business-side and the customer-side, as well as understanding how deepfakes may affect the

overall marketing ecosystem. It is only through more research, shedding light on the

application of deepfakes and their relation to marketing theory and practice, that we can hope

to guide safe, sustainable practices for deepfakes in the marketing field.

This paper applies a Star Wars metaphor throughout to discuss both the light and the

dark side of deepfakes and how we may yet find balance by embracing the ‘force’ of

deepfakes for businesses and customers. The remainder of this paper is organized as follows.

First, we define and introduce deepfakes, providing a typology that differentiates deepfakes

from other marketing practices. Following this, we provide an overview of the benefits of AI

– with a particular focus on the deepfake applications of AI – for businesses and customers

through the theoretical lens of balanced centricity which sets the foundation for the proposed

conceptual framework. Finally, we present our conceptual framework and its propositions

which provide a research agenda for deepfakes.

2. What are Deepfakes? A New Hope or Phantom Menace for Marketing?

A deepfake can be defined as a product of AI technology and machine learning which

merges, combines, replaces or superimposes various types of content to produce a form of

synthetic media that obscures the distinction of authenticity (Maras & Alexandrou, 2019).

Deepfake media first rose to public awareness in 2017 and can take various forms within

5
digital media - whether audio, visual, or audio-visual – as summarized by Kietzmann and

colleagues (2020). For example, photographic deepfakes enable faces and bodies to be

swapped or blended into someone else’s seamlessly. As demonstrated in Figure 1, images of

three individuals can be integrated in different ways to create deepfake content. These two

images were generated using Artbreeder (www.artbreeder.com), an online tool which uses

deepfake-based technology to allow users to merge existing images and generate new ones.

The original three images (of two male co-authors and one female co-author) were uploaded

and merged together as shown in Figure 1, with characteristics of the female co-author being

more pronounced in the left image, and the male characteristics of the two co-authors being

more pronounced in the right image.

(a) (b)
Figure 1. Deepfakes generated by merging images of the authors – (a) female dominant
version, (b) male dominant version.
Deepfakes are not restricted to visual content and can also permit voice-swapping,

with audio being able to be synthesized via either changing or imitating the voice of someone

else (Kietzmann et al., 2020). Text can also be converted into the voice of a chosen individual

(Saito, Takamichi, & Saruwatari, 2017). Beyond photographic and audio deepfakes,

deepfakes can be used within video, where faces can be swapped or morphed (Kietzmann et

al., 2020). Beyond faces, deepfakes can even permit full-body puppetry, with the movement

6
of an individual within a video being manipulated via transposition from another person’s

movement (Chan, Ginosar, Zhou, & Efros, 2019). Perhaps the most sophisticated evolution

of deepfakes, audio-visual deepfakes allow for the manipulation of both the facial movements

and spoken words of an individual within video, convincingly making someone appear to say

things which they have never said (Whittaker et al., 2020).

Deepfakes are rapidly evolving in sophistication and will eventually be undetectable to

the untrained eye (Maras & Alexandrou, 2019). Indeed, the two main factors driving their

proliferation through social media includes their increasing accessibility and believability, as

deepfakes are becoming easier to produce with customer-grade apps such as Zao and

FakeApp, and also harder to distinguish from authentic media due to their increasing

sophistication (Kietzmann et al., 2020). In sum, deepfakes give both individuals and

organizations the power to create highly realistic, yet synthetic representations of whoever

they please.

While the deep neural networks which facilitate deepfakes can artificially generate and

manipulate audio-visual content, it is also noteworthy to briefly discuss that these networks

can generate entirely new, yet realistic content in the form of generative adversarial networks

(GANs). GANs comprise of a generator network and discriminator network, whereby the

generator network (acting the role of counterfeiter) generates content aiming to deceive the

discriminator network (acting the role of a counterfeit detective) (Goodfellow et al., 2014).

Over time, the generator network learns to improve its output to eventually deceive the

discriminator network – which when facilitating deepfake creation can result in highly

realistic synthesized content (Whittaker et al., 2020). That is, the AI that generates content

becomes proficient enough at doing so that the AI that judges the authenticity of the content

can no longer tell the difference. After this training process, the generator can create entirely

new content which has high similarity to the original source input, such as a person’s voice or

7
face (Whittaker et al., 2020). As GANs can generate novel output, they are slightly different

to deepfakes. However, as GANs utilize AI-driven machine learning to create synthesized

human content (like deepfakes), we refer to human-like content generated by GANs as

deepfakes within our paper. Next, we conceptualize how deepfakes differ from other

marketing practices which leverage human realism and/or technologically-facilitated

intelligence.

2.1 A Typology of Human Realism and Technologically-facilitated Intelligence within

Marketing Practice

This section presents a typology that highlights the placement of deepfakes amongst

other marketing practices, and hence, the unique contribution of deepfakes to the marketing

discipline. Our typology is developed to provide a more precise and nuanced understanding

of deepfake phenomena through differentiation from other marketing practices, hence

highlighting a conceptual contribution, in accordance with MacInnis (2011). Differentiation

creates scholarly contribution by distinguishing and classifying phenomena in accordance

with underlying dimensions of comparison to articulate important differences to add clarity

and reduce confusion between phenomena to create new insight (MacInnis, 2011). Our

proposed typology differentiates between deepfakes and related marketing practices to

provide new insight into the unique characteristics and advantages of deepfakes that can be

leveraged by marketers.

Two dimensions of comparison constitute the typology’s axes: human realism and

technologically-facilitated intelligence. Firstly, human realism refers to the ability of digital

content to be representative of humans in the real world, for example the use of humanistic

physical features, faces, and voices. This dimension goes beyond anthropomorphism (the

granting of human features or characteristics) and considers the perception of realism or

8
authenticity. Perceived realism is important because the sense of authenticity can lead to

enhanced trust (Sundar, 2008).

The second axis deals with levels of technologically-facilitated intelligence, and

describes the reliance upon, or utilization of, intelligent technologies to automate digital

content creation. We conceptualize that high-intelligence content relies more heavily on AI,

whereas low-intelligence content has reduced utilization, or absence of AI, relying more on

human input to facilitate content creation. Such a conceptualization is made in accordance

with the Passive Interactive Proactive (PIP) typology (Letheren, Russell-Bennett, Mulcahy, &

McAndrew, 2019), which proposes that as humans place greater reliance on technology, the

technology moves from adopting a passive role (requiring cognitive and behavioral human

input to operate), to becoming interactive (providing personalized output and making some

actions on behalf of the human), and finally proactive (autonomous decision-making and

actions made of behalf of the human, which may be overridden). Please refer to Figure 2,

which presents our typology.

High technologically-facilitated intelligence

Deepfakes
Synthetic video,
Chatbots
audio, and image
Text-based chatbots
content representing
humans
High human Low human
realism realism
Native advertising
Cartoon spokescharacters
Human-based/influe
Images or videos of
ncer endorsement
spokescharacters/cartoon
video and image
s
content

Low technologically-facilitated intelligence

Figure 2: Typology of human realism and technologically-facilitated intelligence.

9
 2.1.1 Deepfakes: High Human Realism and High Technologically-facilitated
Intelligence

Deepfakes are firstly positioned within the high human realism, high technologically-

facilitated level of intelligence quadrant. As deepfakes can be constructed to be highly

believable, albeit inauthentic, audio-visual content which manipulates the facial, vocal, and

bodily features of humans (Whittaker et al., 2020), we propose that deepfakes possess high

human realism. Through their utilization of AI to recognize a human’s facial features and

expression patterns, and via deep learning to optimize output by training on external data fed

into the system (Kietzmann et al., 2020), deepfakes are conceptualized to possess a high level

of cognitive intelligence rather than other advanced forms of intelligence such as emotional

or social intelligence (Kaplan & Haenlein, 2019). Deepfakes are therefore more proactive in

accordance with the Proactive-Interactive-Passive (PIP) framework of Letheren et al. (2019),

as while deepfakes require human input to create, they proactively use AI to act on the

human’s behalf to generate synthetic representations created from inputted data.

2.1.2 Positioning Existing Marketing Practices: Human Realism and

Technologically-facilitated Intelligence

We next compare deepfakes to existing marketing practices which fall within other

quadrants of our proposed typology but differ based upon their levels of realism and use of

technologically-facilitated intelligence. Chatbots attempt to emulate human characteristics,

for example by having humanized voices and visual animations, however such synthetic

representations of human characteristics by chatbots have been shown to produce negative

evaluations and the ‘uncanny valley’ effect (Ciechanowski, Przegalinska, Magnuski, &

Gloor, 2019), therefore they may possess low perceived human realism. However, chatbots

possess a high level of technology-facilitated intelligence as they can utilize AI and machine

learning to automate customer service and communications by employing language-

processing tools to understand customer requests and respond appropriately (Luo, Tong,

10
Fang, & Qu, 2019). Chatbots are therefore more interactive in accordance with the PIP

framework (Letheren et al., 2019), as they provide information and suggestions.

Native advertising possesses high human realism as it emulates authentic and realistic

content which makes it difficult for customers to recognize it is an advertisement (Campbell

& Evans, 2018). As native advertising is often produced by social media influencers (Kay,

Mulcahy, & Parkinson, 2020), relying on the human to generate content rather than utilizing

AI, it is conceptualized to possess low technologically-facilitated intelligence. Native

advertising is therefore passive in accordance within the PIP framework (Letheren et al.,

2019), as the technologies used to create and disseminate the endorsed content (e.g. cameras,

photo and video editing software, social media platforms) are static. Therefore, there is

greater reliance placed upon the human to generate the output.

Lastly, the animated nature of cartoon spokescharacters means that they possess low

human realism. Indeed, realism is generally not the aim of cartoon spokes-characters, which

are instead designed to act as symbols, and even evoke nostalgia for childhood (Hosany,

Prayag, Martin & Lee, 2013). Spokescharacters also possess a low level of technologically-

facilitated intelligence as they lack self-agency being entirely designed, animated, scripted,

and programmed by human visual designers.

The proposed typology does possess several caveats which will now be

acknowledged. Although deepfakes are beginning to become more sophisticated and harder

to distinguish from authentic content (Kietzmann et al., 2020), many deepfakes are

constructed with lower human realism, (particularly for parody purposes), with the alteration

being clearly visible. In addition, while deepfakes are conceptualized to possess high

intelligence, they are not automated and do require human intervention (e.g. to collect

optimal training data of the source and target and input it, in addition to converting and

11
exporting the video and conducting further visual editing if needed). In addition, recall that

deepfakes do not currently possess more advanced forms of AI, such as the simulated

emotional or social intelligence found within human-inspired or humanized AI (Kaplan &

Hainlein, 2019). Furthermore, if other marketing practices were to increase levels of realism

via technologically-facilitated intelligence such as chatbots and spokescharacters, we argue

this would signify these practices being or emulating being ‘deepfaked’. In summary, the

typology developed is not meant to be an exhaustive depiction of all deepfake, chatbot, native

advertising and cartoon spokescharacter content, as shifting can certainly occur between the

quadrants, and other examples may also be found which fit within these quadrants. The

typology instead aims to differentiate deepfakes from commonly deployed marketing

practices which use realism and technologically-facilitated intelligence to better understand

their unique contribution to the marketing discipline. Now that deepfakes have been

introduced, conceptualized and differentiated in comparison to other marketing practices, the

following sections outline their current and potential positive and negative impacts.

2.2. Deepfakes – A New Hope for Marketing?

From the perspective of ‘a new hope’, deepfakes can be utilized to create content

which can positively persuade or connect with customers. For example, deepfakes have been

used to reach voters speaking a different dialect such as in the video released by Indian

president Manoj Tiwari which depicted him speaking another dialect and reached

approximately 15 million people (Jee, 2019). In another instance, social marketing

applications of deepfakes are emerging, such as in the Malaria Must Die campaign. The

campaign uses symbolism and appeals that transcend cultural barriers, depicting former

English footballer David Beckham seemingly speaking nine different languages to promote a

petition to end malaria (Malaria Must Die, 2019).

12
 Customers can also create deepfakes, for instance, using personalized media creation

that uses face-swapping technology or even creating their own AI avatars. For example, users

can face-swap with actors in scenes from movies and TV shows by using Zao, a popular

Chinese app, and post their creations onto social media. Deepfakes therefore offer new

opportunities for businesses and organizations to personalize online experiences and enhance

customer engagement. Such an opportunity was realized when Doritos partnered with the

Sway app to create the first AI dance challenge to engage with customers. Users filmed

themselves striking a collection of dance poses, which the app transposed via deepfake over

the body of a professional dancer emulating the dance moves which Lil Nas X and Sam Elliot

performed in the Doritos Super Bowl commercial. Users were then encouraged to share their

professional dance moves using #CoolRanchDance via social media (Williams, 2020).

Deepfakes can also augment the personalization of online customer experiences such as

clothes shopping. With the recent development of AI-generated promotional models by

companies such as Artificial Talent, online clothes retailers could eventually allow customers

to generate their own personalized avatar and deepfake their facial characteristics onto it to

visualize prospective clothing purchases.

2.3. Deepfakes – a Phantom Menace for Marketing?

Alternatively, deepfakes can be argued to be a ‘phantom menace’, negatively

influencing both customers and businesses. Given deepfakes possess low barriers to creation,

have great potential to persuade the public and can be easily shared in this new digital era

(Kietzmann et al., 2020), these forged entities can be very dangerous indeed. Disconcertingly,

the growing simplicity involved in creating convincing deepfakes, combined with our

increasingly digitally-documented lives, will heighten the potential for deepfakes to be used

for malicious purposes such as disinformation or ‘fake news’, blackmail, intimidation,

13
sabotage, harassment, defamation, revenge pornography, identify theft, and bullying

(Chesney & Citron, 2019; Kietzmann et al., 2020).

Deepfakes present profoundly problematic implications for social media and political

marketers. Fake news proliferates widely on social media and studies have assessed how text-

based disinformation messages are accepted within a social media context (Pennycook &

Rand, 2019). However, as visual stimuli can evoke emotional and cognitive arousal more

effectively than text (Sherwin, Feigenson, & Spiesel, 2006), deepfakes are a more potent and

persuasive tool for disinformation purposes which may even result in the public losing trust

in factual information (Chesney & Citron, 2019). Indeed, Vaccari and Chadwick (2020)

found that political deepfakes do not need to deceive to do damage – instead, their presence

creates confusion, negatively influencing norms and civic culture in a way that could

potentially play a significant role in undermining democracy. In further evidence of the

potential for deepfakes to menace businesses and customers, Facebook recently removed over

900 counterfeit accounts which circulated pro-Trump messages as they utilized fake profile

pictures generated using deepfake technology (Nuñez, 2019). Given the emergence of these

practices, countries such as China and the United States are exploring legislative measures to

mitigate negative impacts of deepfakes. China for instance has declared it a criminal offense

to publish a deepfake without providing disclosure of its artificial nature (Woollacott, 2019).

Comparatively, the United States Senate recently passed the Deepfake Report Act. This Act

requires the Department of Homeland Security to publish an annual report on ‘digital content

forgery’ – digital content crafted using AI which is designed to mislead (Deepfake Report

Act, 2019).

In summary, these discussions highlight the potential for deepfakes to have a positive

and negative impact on customers and businesses. Understanding this juxtaposition, we next

14
argue how balance within the deepfake ‘force’ is needed by applying the theoretical lens of

balanced centricity.

3. The Need for Balance in the Deepfake Force: A Conceptual Model

In this paper, we argue for the need to balance the benefits between business and

customers to ensure all can benefit from the deepfake application of AI. Arguments regarding

the need for balance in marketing theorizing are not new, but they are yet to be extended into

the application of AI and deepfakes. This debate regarding the need for balance is evidenced

in discussions regarding the limitations of perspectives such as customer centricity, whereby

the focus of marketing on benefitting the customer (e.g. ‘customer is king’ or ‘the customer is

always right’; Shah, Rust, Parasuraman, Staelin, & Day, 2006). Instead, others such as

Gummesson (2008) urge marketing scholars to accept the complexity of marketing and

embrace and theorize towards a balanced centricity, whereby all stakeholders benefit, and

thus in turn create a more sustainable and harmonious economy. It is in line with the

theoretical perspective of balanced centricity that we suggest deepfakes can assist by

balancing the benefits for business and customers, which is what we demonstrate in the

propositions arising from our conceptual model.

Currently, in opposition to the central tenets of balanced centricity, it could be argued

that emergent empirical studies into deepfakes are predominantly customer-centric, with a

distinct focus on understanding how customers might interact with deepfake media in an

effort to begin to measure their negative impacts (Dobber et al., 2020; Vaccari & Chadwick,

2020). In comparison, there is only limited and emerging commentary discussion of the

business implications of deepfakes (see Kietzmann et al., 2020; Kwok & Koh, 2020;

Whittaker et al., 2020), with some reports suggesting they may result in deep losses for

businesses (Helms & Mutter, 2020). We argue that theorizing and studying deepfakes

through a balanced centricity lens will help businesses (and customers) in understanding how

15
all stakeholders, rather than one in isolation, can benefit or be protected from the negative

impacts of deepfakes. Balanced centricity therefore provides research motivation and

direction to ensure that the positive and negative impacts of deepfakes are understood within

both the customer and business sphere.

In line with this prior argument, the current paper proposes three main propositions,

which are broken into the business and customer sphere based upon the notion of balanced

centricity, as shown in Figure 3. The top of our model depicts our earlier discussion in

Section 2 which explained that AI powers deepfakes, and hence this is the first connection

drawn within our model. Next is the association between deepfakes and balanced centricity.

As argued in Section 3, there is a need for balanced centricity to ensure that both customers

and businesses can benefit, but also just as importantly, be protected from deepfakes. The

relationship drawn between deepfakes and balanced centricity provide the grounding for our

propositions and their variations which are broken up according to the business and customer

sphere. Next, the first proposition relating to customer service and self-service is discussed.

16
 Artificial Intelligence

Business Sphere Customer Sphere

Improved customer service Deepfakes Empowered self-service

P1

P1 b
a

Disclosure P2b P2a Disclosure

protection protection

P3
P3 a

b
Balanced Centricity Customer
Business Personalized
Smart marketing
advertising content

Figure 3. Conceptual model of deepfakes and their implications for businesses and customers.

17
3.1. Proposition 1 – Customer and Self-service via Deepfakes

3.1.1 Improved Customer Service in the Business Sphere

The broader AI literature provides support for deepfakes enhancing customer service.

Scholars argue that AI will begin – and continue to – replace repetitive service tasks and

undertake analytical tasks, whereas the proficiency of AI in empathetic tasks is limited

(Huang & Rust, 2018). For example, AI-generated advertisements are not as effective at

conveying emotional or hedonic appeals when compared to human-created advertisements,

and people may form negative evaluations of AI because of the inability of AI to experience

or feel (Bakpayev et al., in press).

As shown in recent studies of AI, customers have a negative perception of the use of

‘machines’ (Luo et al., 2019). Individuals are also known to possess fundamental biases

against algorithms which could impede the acceptance of AI within customer service

contexts, a phenomenon referred to as algorithm aversion (Dietvorst, Simmons, & Massey,

2015). Individuals prefer to rely on humans rather than algorithms even when this results in

an inferior overall outcome, as there is a greater intolerance of mistakes made by algorithms

when compared to humans, with a greater likelihood of abandoning an algorithmically based

decision-maker rather than a human alternative (Dietvorst et al., 2015). In addition,

individuals place less trust in algorithms when they are used to complete subjective tasks,

which require interpretation and intuition and emotional perception, as algorithm aversion is

primary driven by the perception that algorithms lack such human abilities (Castelo, Bos, &

Lehmann, 2019).

Whilst studies such as Huang and Rust (2018) suggest AI cannot be emotive (e.g.

express empathy), we suggest AI via deepfakes may bridge this gap. This aligns with

suggestions that AI can help interactions between machine and human, as well as human-to-

human interactions (Wilson & Daugherty, 2018). When algorithms are perceived to have

18
high affective human-likeness (e.g. creativity and emotional sensitivity) algorithm aversion

has been shown to have a reduced effect within more emotionally-orientated and intuitive

tasks (Castelo et al., 2019).

Deepfakes can potentially be used to enhance the naturalness of artificial agents and

improve their ability to build simulated empathy and emotional connection. Deep neural

networks, which facilitate the creation of deepfakes, can be trained to recognize and learn to

represent emotions based on inputted facial data (Lai & Lai, 2018). Deepfakes are already

being used to impart emotional experiences within customer contexts. The Dalí Museum in

Florida adopted deepfake technology to bring tourists face-to-face with Dalí, who appears as

a life-size incarnation and speaks with the visitor to better forge an emotional connection

between the visitor and his artworks (The Dali Museum, 2019). Deepfake technologies

therefore grant artificial agents the power to not only establish a form of emotional

connection but can imbue them with a simulation of emotional intelligence, enabling them to

adapt and better satisfy human desire for a more natural emotional experience.

Imagery can help to capture attention, increase memorability, influence attitudinal

changes, and evoke cognitive and emotional arousal to become more persuasive (Seo &

Dillard, 2019; Sherwin et al., 2006). Facial expressions serve a vital role in coordinating

social interactions, serving as a visual tool which helps the message recipient understand the

communicator’s emotions, beliefs and intentions (Keltner & Haidt, 1999). Positive emotions

such as enthusiasm, amusement, and attachment love facilitate greater acceptance of weak

persuasive messages (Griskevicius, Shiota, & Neufeld, 2010), therefore the addition of

emotional warmth to messaging may help to improve its persuasiveness. Using deepfakes to

visualize positive emotions such as facial expressions could therefore improve the

persuasiveness of communication over text-based alternatives and help to deliver the message

as it was intended.

19
 We therefore propose that deepfakes could be used to emotionally augment AI-

powered service agents (i.e. chatbots and virtual assistants), allowing them to possess greater

ability to simulate emotions, thereby enhancing perceived emotional intelligence and

adapting in real-time to human needs for empathy and understanding.

P1a: Deepfakes integrated into AI powered services (e.g. chatbots, virtual assistances)
will increase perceptions of their emotional intelligence

3.1.2 Empowered Self-Service Customer Sphere

Another benefit which may accrue from deepfakes and AI for customers is greater

empowerment of self-service within the marketplace. As André et al. (2018) point out, AI

will be able empower customers through lowering search, transaction, and decision-making

costs. Further, AI may also benefit vulnerable customers – individuals who experience a state

of powerlessness that hinders consumption goals and circumstances which negatively affect

perceptions of self (Baker, Gentry, & Rittenburg, 2005). Therefore, AI may help to alleviate

such powerlessness and allow vulnerable customers to take advantage of market

opportunities and be supported to deal with complex marketplaces and exchanges where they

may have limited experience or self-efficacy (Libai et al., in press).

It could therefore be posited that applications of deepfakes via AI could empower

vulnerable customers to use self-service technologies (SSTs) by providing them with

increased levels of self-efficacy, confidence and trust. Indeed, the concept of customer

empowerment has been extensively investigated in reference to self-service technology

(SSTs). Technological designs can increase levels of empowerment through self-efficacy,

confidence, and trust, which lead to greater use of digital services. For example, customers of

m-health services are known to use them more if they have a greater sense of empowerment

through heightened self-efficacy (Schuster, Proudfoot, & Drennan, 2015). Other studies also

show that if customers feel empowerment in digital and self-service settings through

20
increased levels self-confidence, or the sense of power or autonomy, they have an increased

likelihood toward adopting mobile banking (Chaouali, Souiden, & Ladhari, 2017). Perceived

empowerment has also been shown to lead to increased trust within virtual environments

(Füller, Mühlbacher, Matzler, & Jawecki, 2009).

We therefore extend these findings from prior technologies to suggest that deepfakes

may be able to replicate and improve such outcomes. For instance, individuals diagnosed

with amyotrophic lateral sclerosis (ALS) may rely on a text-to-speech interface to facilitate

verbal communication as the condition progressively inhibits voluntary muscle control and

can lead to total paralysis. Using generative AI to learn from voice samples provided by those

diagnosed with ALS, Project Revoice (www.projectrevoice.org) creates personalized

synthetic voices for vocally-paralyzed clients which allows them to communicate with others

using their own voice. Similarly, deepfakes can be used to empower the marginalized to

allow them to share their stories using authentic emotion without fear of identification.

Welcome to Chechnya, a documentary which explores the persecution of LGBTQ people in

Chechnya, uses deepfakes to anonymize the subjects of the documentary by transplanting the

faces of New York-based activists onto the faces of the Chechen documentary subjects. The

use of deepfakes to anonymize, instead of utilizing conventional anonymization tools within

documentaries such as pixilation, blurring, darkened rooms, and altered voices, allows the

documentary subjects to convey richer emotion (Richards, 2020). Deepfakes can therefore be

used to provide the disempowered with the confidence to communicate without fear of

identification while also preserving the emotional richness of their messages. Based upon the

prior discussion of prior studies and the practical application examples of deepfakes, we

propose the following:

P1b: Deepfakes integrated within self-service technologies will increase a customer’s

sense of empowerment (self-efficacy, confidence and trust) and in turn their usage.

21
3.2. Proposition 2 – Protecting Customers and Businesses from Deepfakes via Disclosure

The next proposition of our conceptual model suggests the need to protect the

customer and business sphere from the deviant use of deepfakes. Recall previously in Section

2.2 which discussed that deepfakes can potentially be a ‘phantom menace’ for marketers

(business) and customers. Further, to combat this threat, policy makers are currently

discussing regulatory measures such as disclosure of deepfake content creation to address this

issue. Disclosure has been of interest to marketers, particularly regarding sponsorship

(Campbell & Evans, 2018) and social media influencer marketing (Kay et al., 2020). Recent

developments such as Facebook banning AI-manipulated videos (deepfakes) as a result of

their potential influence within the 2020 US Presidential Election (Hern, 2020) have drawn

the attention of regulators and policy makers regarding the need for deepfake disclosure.

Whilst drawing this attention, these discussions of the integration of disclosure regulations

for deepfakes have yet to be empirically supported or understood. We therefore turn to the

broader disclosure literature to propose future research directions regarding the impact of

disclosure on deepfakes, which can contribute important theoretical and policy insights.

3.2.1 Disclosure of Deepfakes to Protect the Customer Sphere

In the literature, studies have evidenced that disclosure of advertising has assisted

customers in overcoming the persuasiveness of a message (Boerman, Williemsen, & Van Der

Aa, 2017; Campbell, Mohr, & Verlegh, 2013; De Jans, Cauberghe, & Hudders, 2019). For

example, Campbell and colleagues’ (2013) study shows that disclosure helps customers

overcome covert marketing efforts and correct persuasive knowledge. This most likely aligns

with the perspective of policy makers and regulators that introducing disclosure measures

will help protect the customer sphere. Thus, given the evidence of prior disclosure research,

we propose the following, relating to deepfakes and the customer sphere:

22
 P2a: Disclosure in deepfakes will protect customers from the persuasion of deepfakes.

3.2.2 Disclosure of Deepfakes to Protect the Business Sphere

Research also hints that the business sphere can benefit through disclosure. Kay and

colleagues’ (2020) study evidenced that when social media influencers disclosed sponsorship,

customers perceived products as more attractive and were more likely to purchase. Thus, in

line with these findings, it could be suggested that a balance could be struck for the business

and customer sphere by the introduction of disclosure measures for deepfakes. However,

disclosure is not only important from business-controlled messaging, but also protection from

the threat they may pose to their brands and products via fake news and fake posts using

deepfakes. Fake news can target brands, such as a malicious news article which intentionally

misquoted Pepsi’s CEO Indra Nooyi to say that Donald Trump’s supporters should not

purchase Pepsi, which led to calls for boycotts and potentially influenced the 3.75% drop in

Pepsi’s share price (Liffreing, 2016). In 2019, a company was defrauded of $243,000USD

when AI-enabled voice software was used to convince the CFO that the CEO was requesting

the money be transferred (Stupp, 2019). Deepfakes can therefore be constructed to add

artificial realism towards malicious agendas, meaning that businesses may need to begin to

formulate strategies to protect themselves from potential reputational damage, for example, in

the event of a malicious deepfake video targeting a senior executive going viral on social

media. Therefore, based upon the prior discussion, and in line with the findings of benefits of

disclosure for business outcomes from Kay and colleagues (2020), we propose:

P2b: Disclosure in deepfakes will enhance business outcomes and protect them from
the deviant/destructive use of deepfakes.

23
3.3 Proposition 3 – Smart and Personalized Advertising via Deepfakes

We propose that deepfakes will provide opportunities for the business sphere

regarding smart advertising, advertising which takes advantage of integrated customer

tracking solutions to create commercials and advertisements which are adapted to the

preferences of the public or individuals (Culic, Radovici, & Rasu, 2020). Deepfakes can

already be used to develop personalized videos to send to customers throughout their

customer journey. Personalized deepfake videos using an AI presenter can be created via

Synthesia (www.synthesia.io), where organizations can submit text to be converted into

personalized videos such as appointment reminders, welcome messages, and abandoned

shopping cart reminders, using the customer’s name and native language. Personalized

internet content has been shown to create positive evaluations (Kalyanaraman & Sundar,

2006), reduce information overload, and increase user satisfaction (Liang, Lai, & Ku, 2006).

In addition, anthropomorphized content has been shown to have a positive influence upon

customer behavior, where products possessing human-like features are evaluated more

positively (Aggarwal & McGill, 2007), and the presence of anthropomorphism within

branding can have positive influence on marketing outcomes such as brand love

(Rauschnabel & Ahuvia, 2014). As deepfakes are already being implemented to personalize

marketing communications and can be used anthropomorphize content via automatic

generation of human characteristics such as faces and voices, deepfakes will potentially

enhance the authenticity and effectiveness of advertising messages (Kietzmann et al., 2020).

3.3.1 Smart Advertising and Creating the Ideal Message Source for the Business
Sphere

The first benefit to business, smart advertising, refers to AI-based advertising which

allows “consumer-centered, data-driven and algorithm-mediated brand communication” (Li,

2019, p. 333). Research suggests AI is assisting in the curation and refinement of content

(Kietzmann, Paschen, & Treen, 2018), targeted and personalized advertising in real time, and

24
return on investment on advertisement expenditure (Davenport, Guha, Grewal, & Bressgott,

2020). Indeed, scholars such as Davenport and colleagues (2020, p. 35) pose the question

which encapsulates the benefits of smart advertising, “would [these] advertising dollars be

required in the future, wherein firms may be better predicting customers’ preferences, and

thus would not need to advertise as much?”.

Creating an ideal message source is central to the development of source credibility.

An individual’s tendency to accept information depends on the perceived credibility of the

communication source (Hovland, Janis, & Kelley, 1953). One such practice employed by

marketers to adopt an ideal message source is the utilization of celebrity endorsement for

product or brand promotion. Celebrities are considered as effective endorsers as they are

viewed to be trustworthy, believable, persuasive, and likeable (Freiden, 1984). The value of

celebrity endorsement is realized through the creation of an associative link between the

brand and celebrity, however this also creates inherent risk (Till, 1998). As endorser

credibility can directly influence brand credibility and indirectly influence customer-based

brand equity (Spry, Pappu, & Cornwell, 2011), negative publicity acquired by the celebrity

can be associated by customers with the endorsed brand (Till, 1998). To alleviate the risks

associated with celebrity endorsement, marketers can instead opt to utilize spokescharacters

for promotion. Through visual design or animation, marketers can control spokescharacters

and how they promote products or brands to instill trust towards the character and in turn

brand attitude (Garretson & Niedrich, 2004). Given that cartoon spokescharacters act not as

themselves, but are specifically animated to serve as a symbol and embodiment of brand

characteristics (Stafford, Stafford & Day, 2002), they are inherently less realistic and

authentic than a non-cartoon endorser who exists outside of their brand role. While less

human realism may be preferable in many instances, in some cases the level of authenticity

offered by a human endorser may be desirable. We propose that marketers can use deepfakes

25
to create a synthetic endorser, striking the balance between a human endorser and

controllable spokescharacter, to create an ideal message source.

Lil Miquela is one example of a synthetic endorser, who is constructed using

computer-generated imagery (CGI). Boasting 2.7 million followers on Instagram, Lil

Miquela engages in endorsement of brands such as Samsung, Calvin Klein, Dior, and Prada

(Powers, 2019). Synthetic endorsers such as Lil Miquela have distinct advantages for brands,

as they stick to the script, can be aesthetically perfected to suit the brand’s objective, have

their personal narrative and life drama designed by the brand, all without the uncontrollable

risks or issues which a human endorser can introduce (Koh & Wells, 2018; Powers, 2019).

Synthetic endorsers therefore retain the controllable aspects of conventional

spokescharacters, but with an enhanced ability to embody human qualities similar to those of

the target audience, and hence a greater sense of authenticity and relatability. For instance, a

personal narrative or life story, beliefs, personality, and physical attractiveness can all be

programmed into a synthetic endorser to create an ideal message source to fit the brand’s

audience. In turn, synthetic endorsers can feel real, as an individuals’ love or liking of a

fictional character can be attributed to greater perceived realness (Gardner & Knowles, 2008).

Beyond CGI, the use of deepfakes to produce an ideal message source is beginning to be

realized by companies such as Artificial Talent (https://artificialtalent.co), who use deepfake

technology to create AI-generated fashion models – giving brands the ability to generate

synthetic, region-specific promotional models who wear the clothing of the fashion brand

using customizable poses, all without incurring the costs associated with employing human

models, such as agency fees, photographers, and beautification services.

As noted in the previous sections, deepfakes offer unique opportunities to change the

appearance of individuals in marketing messages and further to reduce the ‘noise’ of

advertising by appropriately targeting messages. Not only will deepfakes leverage the

26
benefits of AI for targeting customers, deepfakes will also be able to refine messages and

message sources (e.g. promotional models and spokescharacters) to the preference of the

customer. Thus, given the previously noted advantages of AI and those combined with the

advantages of deepfakes, it is plausible to suggest that deepfake-powered advertising could

be superior to purely AI-based advertising messaging.

P3a: Deepfake-enhanced advertising messages will be more effective than purely AI-
based advertising messages.

3.3.2 Improved Advertising Authenticity and Personalization for the Customer Sphere

The broader AI literature outlines numerous advantages which technological advances

will provide customers, particularly regarding their customer experience, and we transfer this

thinking to their experience of advertising. AI-powered chatbots can be viewed as personal

assistants as they assist customers with requests and automate customer services (Luo et al.,

2019), which importantly demonstrates that AI can work for the customer and not just the

company, enhancing authenticity and believability. The integration of deepfakes can possibly

enhance the perception of authenticity and personalization of advertising messaging, all of

which are argued to be important for enhancing messaging effectiveness (Beverland,

Lindgreen, & Vink, 2008). For instance, artificial tools such as Google Lens and Amazon

Style Snap identify fashion items from videos or photos viewed by customers, providing

information about how they can be worn and where they can be purchased (Lui, 2019).

Deepfakes may also be increasingly deployed within the tourism industry to enable more

vivid visualization. For example, co-creation opportunities between companies and potential

tourists could occur if social media users were given super-resolution images of the tourist

destination and the ability to deepfake themselves into the destination image – allowing them

to better visualize a potential tourist experience (Kwok & Koh, 2020). Using the examples of

fashion and tourism, we propose that deepfakes can create more personalized advertising by

27
allowing for more realistic visualization of the customer within the consumption experience

and improving their ability to imagine how they will use products, which has been identified

as an important assistance for customers when understanding their purchases (Mulcahy,

Letheren, McAndrew, Glavas, & Russell-Bennett, 2019). We therefore propose based upon

the prior discussion:

P3b: Deepfake-enhanced advertisements will improve perceptions of advertising

authenticity and personalization, as well as enhance a customer’s ability to imagine
use of products.

4. Agenda for the Future

By design, our review of extant work in deepfakes has taken a predominately

marketing focus to ensure the conceptual model and propositions are actionable, but also to

uncover the gaps in the marketing literature which provide new and exciting topics for

exploration. From this perspective, we provide the below specific research questions using

the aforementioned areas of smart advertising and personalization, customer and self-service,

and disclosure to provide a research agenda for marketing scholars’ investigation of

deepfakes. Please see Table 1.

28
Table 1.
Overview of marketing practice, propositions and research questions for deepfakes.
Area of Marketing Practice Proposition(s) Research Questions

Customer and Self-Service (P1) P1a: Deepfakes integrated into AI Q1a: Can deepfakes of digital
powered services (e.g. chatbots, service employees enhance
virtual assistances) will increase customer perceptions of a service
perceptions of their emotional organization’s customer service?
intelligence.
Q1b: Do customers perceive
P1 : Deepfakes integrated within
b
deepfake-enhanced digital
self-service technologies will assistants as superior to other
increase a customer’s sense of digital assistants (e.g. chatbots?).
empowerment (self-efficacy,
confidence and trust) and in turn Q1c: How can deepfakes be used
their usage. to empower vulnerable customers
in consumption settings?

Disclosure Protection for P2a: Disclosure in deepfakes will Q2a: What impact does disclosure
Customers and Businesses (P2) protect customers from the of deepfakes have upon customer
persuasion of deepfakes. reactions and business outcomes?
P2b: Disclosure in deepfakes will Q2b: Do certain types of disclosure
enhance business outcomes and of deepfakes work more
protect them from the effectively than others?
deviant/destructive use of deepfakes.
Q2c: Does the timing and
placement of disclosure for
deepfakes impact their
effectiveness?

Smart and Personalized P3a: Deepfake-enhanced advertising Q3a: To what extent can business
Advertising (P3) messages will be more effective than advertising benefit from
purely AI-based advertising deepfakes?
messages.
Q3b: What are the opportunities
P3 : Deepfake-enhanced
b
for personalized deepfake
advertisements will improve advertising, and how do customers
perceptions of advertising respond to such content?
authenticity and personalization, as
well as enhance a customer’s ability
to imagine use of products.

4.1. Customer and Self-service

A significant body of research has, and is, continuing to be conducted in the use of AI

technology for customer service and self-service. These studies to date have often been

focused on machine (robot and chatbot) or human (self-service) input. Deepfakes provide an

opportunity to begin to blur these foci, which provides opportunities for marketing scholars to

gain insights into the issues and opportunities of blending human and machine. For example,

29
recall that in our proposed typology we positioned chatbots within the low human realism,

high technologically-facilitated intelligence quadrant – it is possible that with the inclusion of

deepfake enhancement that human realism and technologically-facilitated intelligence may be

improved. To guide the identification of these insights we propose the following research

questions:

Q1a: Can deepfakes of digital service employees enhance customer perceptions of a

service organization’s customer service?
Q1b: Do customers perceive deepfake-enhanced digital assistants as superior to other
digital assistants (e.g. chatbots)?
Q1c: How can deepfakes be used to empower vulnerable customers in consumption
settings?

4.2. Disclosure

In evaluating deepfakes and their potential to be a ‘phantom menace’, understanding

strategies such as the impact of disclosure on their level of persuasion is critical. These have

important implications for the development of policy and regulations for deepfakes –

specifically the impact which disclosure has on deepfakes attempting to persuade customers.

Further, insights could be drawn as to when, how, and where disclosure of a deepfake must

be made to mitigate their level of persuasion. Marketing scholars can therefore seek to

address the following research questions relating to deepfakes and disclosure:

Q2a: What impact does disclosure of deepfakes have upon customer reactions and
business outcomes?
Q2b: Do certain types of disclosure of deepfakes work more effectively than others?
Q2c: Does the timing and placement of disclosure for deepfakes impact their
effectiveness?

4.3. Smart Advertising and Personalization

It is widely acknowledged that AI will change the way marketers create content and

advertise, and it is hopeful that this will benefit both organizations and customers. Finding the

30
balance whereby AI enhances marketing and customer outcomes will be challenging. Of

importance is to understand how and if deepfakes can potentially be used achieve this

balance. To direct marketing scholarship to investigate whether deepfakes may achieve this

balance, the following research questions are proposed:

Q3a: To what extent can business advertising benefit from deepfakes?

Q3b: What are the opportunities for personalized deepfake advertising, and how do
customers respond to such content?

5. Conclusion

Our key aims in this paper were to define and introduce deepfakes to the marketing

literature, differentiate deepfakes from existing marketing practices, and provide a conceptual

model which sets an agenda for future research. We have defined deepfakes in a broad sense,

providing a series of recent examples to demonstrate the contemporary importance of this

topic to marketing scholarship. We proposed a typology which demonstrates that deepfakes

provide a unique contribution to the marketing discipline due to their high human realism and

technologically-facilitated intelligence. Lastly, we developed and proposed a conceptual

framework that highlights the need for balanced centricity, whereby businesses and

customers can benefit from deepfakes but also be protected from them. The unresolved

questions we provide have been organized in accordance to three identified areas of

importance to “The Rise of Deepfakes” and set an agenda to direct future empirical

investigations in this domain.

Acknowledgements: This research did not receive any specific grant from funding

agencies in the public, commercial, or not-for-profit sectors.

31
References

Aggarwal, P., & McGill, A. L. (2007). Is that car smiling at me? Schema congruity as a basis
for evaluating anthropomorphized products. Journal of Consumer Research, 34(4),
468-479.
André, Q., Carmon, Z., Wertenbroch, K., Crum, A., Frank, D., Goldstein, W., ... & Yang, H.
(2018). Consumer choice and autonomy in the age of artificial intelligence and big
data. Customer Needs and Solutions, 5(1-2), 28-37.
Baker, S. M., Gentry, J. W., & Rittenburg, T. L. (2005). Building understanding of the
domain of consumer vulnerability. Journal of Macromarketing, 25(2), 128-139.
Bakpayev, M., Baek, T. H., van Esch, P., & Yoon, S. (2020). Programmatic creative: AI can
think but it cannot feel. Australasian Marketing Journal. In press.
Beverland, M. B., Lindgreen, A., & Vink, M. W. (2008). Projecting authenticity through
advertising: Consumer judgments of advertisers' claims. Journal of Advertising, 37(1),
5-15.
Boerman, S. C., Willemsen, L. M., & Van Der Aa, E. P. (2017). “This post is sponsored”:
Effects of sponsorship disclosure on persuasion knowledge and electronic word of
mouth in the context of Facebook. Journal of Interactive Marketing, 38, 82-92.
Campbell, C., & Evans, N. J. (2018). The role of a companion banner and sponsorship
transparency in recognizing and evaluating article-style native advertising. Journal of
Interactive Marketing, 43, 17-32.
Campbell, M. C., Mohr, G. S., & Verlegh, P. W. (2013). Can disclosures lead consumers to
resist covert persuasion? The important roles of disclosure timing and type of
response. Journal of Consumer Psychology, 23(4), 483-495.
Castelo, N., Bos, M. W., & Lehmann, D. R. (2019). Task-dependent algorithm aversion.
Journal of Marketing Research, 56(5), 809-825.
Chan, C., Ginosar, S., Zhou, T., & Efros, A. A. (2019). Everybody dance now. Proceedings
of the IEEE International Conference on Computer Vision. 5933-5942.
Chaouali, W., Souiden, N., & Ladhari, R. (2017). Explaining adoption of mobile banking
with the theory of trying, general self-confidence, and cynicism. Journal of Retailing
and Consumer Services, 35, 57-67.
Chesney, R., & Citron, D. (2019). Deepfakes and the new disinformation war: The coming
age of post-truth geopolitics. Foreign Affairs, 98(1), 147-155.
Ciechanowski, L., Przegalinska, A., Magnuski, M., & Gloor, P. (2019). In the shades of the
uncanny valley: An experimental study of human–chatbot interaction. Future
Generation Computer Systems, 92, 539-548.
Culic, I., Radovici, A., & Rusu, C. (2020). Smart advertising system. In Commercial and
Industrial Internet of Things Applications with the Raspberry Pi (pp. 125-162).
Berkeley, CA: Apress.
Davenport, T., Guha, A., Grewal, D., & Bressgott, T. (2020). How artificial intelligence will
change the future of marketing. Journal of the Academy of Marketing Science, 48(1),
24-42.

32
De Jans, S., Cauberghe, V., & Hudders, L. (2019). How an advertising disclosure alerts
young adolescents to sponsored vlogs: The moderating role of a peer-based advertising
literacy intervention through an informational vlog. Journal of Advertising, 47(4), 309-
325.
Deepfake Report Act of 2019, S. 2065, 116th Cong., 1st Sess. (2019).
Dietvorst, B. J., Simmons, J. P., & Massey, C. (2015). Algorithm aversion: People
erroneously avoid algorithms after seeing them err. Journal of Experimental
Psychology: General, 144(1), 114-126.
Dobber, T., Metoui, N., Trilling, D., Helberger, N., & de Vreese, C. (2019). Do
(microtargeted) deepfakes have real effects on political attitudes?. The International
Journal of Press/Politics, 1-23.
Freiden, J. B. (1984). Advertising spokesperson effects: an examination of endorser type and
gender on two audiences. Journal of Advertising Research, 24(5), 33-41.
Füller, J., Mühlbacher, H., Matzler, K., & Jawecki, G. (2009). Consumer empowerment
through internet-based co-creation. Journal of Management Information Systems, 26(3),
71-102.
Gardner, W. L., & Knowles, M. L. (2008). Love makes you real: Favorite television
characters are perceived as “real” in a social facilitation paradigm. Social Cognition,
26(2), 156-168.
Garretson, J. A., & Niedrich, R. W. (2004). Spokes-characters: Creating character trust and
positive brand attitudes. Journal of Advertising, 33(2), 25-36.
Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville,
A., & Bengio, Y. (2014). Generative adversarial nets. Proceedings of the Neural
Information Processing Systems Conference. 2672-2680.
Griskevicius, V., Shiota, M. N., & Neufeld, S. L. (2010). Influence of different positive
emotions on persuasion processing: A functional evolutionary approach. Emotion,
10(2), 190-206.
Gummesson, E. (2008). Extending the service-dominant logic: from customer centricity to
balanced centricity. Journal of the Academy of Marketing Science, 36(1), 15-17.
Helms, K. J., & Mutter, D. (2020). Deepfakes could mean deep losses for businesses if not
prepared. Retrieved from:
http://www.prweb.com/releases/deepfakes_could_mean_deep_losses_for_businesses_if
_not_prepared/prweb17431805.htm
Hern, A. (2020, January, 07). Facebook bans 'deepfake' videos in run-up to US election. The
Guardian. Retrieved from:
https://www.theguardian.com/technology/2020/jan/07/facebook-bans-deepfake-videos-
in-run-up-to-us-election
Hosany, S., Prayag, G., Martin, D., & Lee, W. Y. (2013). Theory and strategies of
anthropomorphic brand characters from Peter Rabbit, Mickey Mouse, and Ronald
McDonald, to Hello Kitty. Journal of Marketing Management, 29(1-2), 48-68.
Hovland, C. I., Janis, I. L., & Kelley, H. H. (1953). Communication and persuasion, New
Haven, CT: Yale University Press.
Huang, M. H., & Rust, R. T. (2018). Artificial intelligence in service. Journal of Service
Research, 21(2), 155-172.
33
Jee, C. (2020). An Indian politician is using deepfake technology to win new voters.
Retrieved from: https://www.technologyreview.com/f/615247/an-indian-politician-is-
using-deepfakes-to-try-and-win-voters/
Kalyanaraman, S., & Sundar, S. S. (2006). The psychological appeal of personalized content
in web portals: Does customization affect attitudes and behavior?. Journal of
Communication, 56(1), 110-132.
Kaplan, A., & Haenlein, M. (2019). Siri, Siri, in my hand: Who’s the fairest in the land? On
the interpretations, illustrations, and implications of artificial intelligence. Business
Horizons, 62(1), 15-25.
Kay, S., Mulcahy, R., & Parkinson, J. (2020). When less is more: the impact of macro and
micro social media influencers’ disclosure. Journal of Marketing Management, 36(3-4),
248-278.
Keltner, D., & Haidt, J. (1999). Social functions of emotions at four levels of analysis.
Cognition & Emotion, 13(5), 505-521.
Kietzmann, J., Lee, L. W., McCarthy, I. P., & Kietzmann, T. C. (2020). Deepfakes: Trick or
treat?. Business Horizons, 63(2), 135-146.
Kietzmann, J., Paschen, J., & Treen, E. (2018). Artificial intelligence in advertising: How
marketers can leverage artificial intelligence along the consumer journey. Journal of
Advertising Research, 58(3), 263-267.
Kietzmann, T. C., McClure, P., & Kriegeskorte, N. (2019). Deep neural networks in
computational neuroscience. In Oxford Research Encyclopedia of Neuroscience.
Oxford, UK: Oxford University Press.
Koh, Y., & Wells, G. (2018). The making of a computer-generated influencer. The Wall
Street Journal. Retrieved from: https://www.wsj.com/articles/the-making-of-a-
computer-generated-influencer-11544702401
Kwok, A. O., & Koh, S. G. (2020). Deepfake: a social construction of technology
perspective. Current Issues in Tourism, 1-5.
Lai, Y. H., & Lai, S. H. (2018). Emotion-preserving representation learning via generative
adversarial network for multi-view facial expression recognition. 13th IEEE
International Conference on Automatic Face & Gesture Recognition (pp. 263-270),
Xi’an, China: Conference Publishing Services.
Letheren, K., Russell-Bennett, R., Mulcahy, R., & McAndrew, R. (2019). Rules of
(household) engagement: technology as manager, assistant and intern. European
Journal of Marketing, 53(9), 1934-1961.
Li, H. (2019). Special section introduction: Artificial intelligence and advertising. Journal of
Advertising, 48(4), 333-337.
Liang, T. P., Lai, H. J., & Ku, Y. C. (2006). Personalized content recommendation and user
satisfaction: Theoretical synthesis and empirical findings. Journal of Management
Information Systems, 23(3), 45-70.
Libai, B., Bart, Y., Gensler, S., Hofacker, C. F., Kaplan, A., Kötterheinrich, K., & Kroll, E.
B. (in press). Brave new world? On AI and the management of customer
relationships. Journal of Interactive Marketing.

34
Liffreing, I. (2016). Fake news sites spark Pepsi boycott, with other brands in crosshairs.
Retrieved from: https://www.campaignlive.com/article/fake-news-sites-spark-pepsi-
boycott-brands-crosshairs/1416027
Lowry, B. (2017). Carrie Fisher's death renews questions about digital re-creation. Retrieved
from: https://edition.cnn.com/2017/01/16/entertainment/carrie-fisher-digital-recreation/
index.html
Lui, M. (2019). The robot wears Prada: what happens when AI starts giving out fashion tips?
Retrieved from: https://theconversation.com/the-robot-wears-prada-what-happens-
when-ai-starts-giving-out-fashion-tips-125140
Luo, X., Tong, S., Fang, Z., & Qu, Z. (2019). Frontiers: Machines vs. humans: The impact of
artificial intelligence chatbot disclosure on customer purchases. Marketing Science,
38(6), 937-947.
MacInnis, D. J. (2011). A framework for conceptual contributions in marketing. Journal of
Marketing, 75(4), 136-154.
Malaria Must Die. (2019, April 08). David Beckham speaks nine languages to launch malaria
must die voice petition. [Video file]. https://www.youtube.com/watch?v=QiiSAvKJIHo
Maras, M. H., & Alexandrou, A. (2019). Determining authenticity of video evidence in the
age of artificial intelligence and in the wake of Deepfake videos. The International
Journal of Evidence & Proof, 23(3), 255-262.
Mende, M., Scott, M. L., van Doorn, J., Grewal, D., & Shanks, I. (2019). Service robots
rising: How humanoid robots influence service experiences and elicit compensatory
consumer responses. Journal of Marketing Research, 56(4), 535-556.
Mulcahy, R., Letheren, K., McAndrew, R., Glavas, C., & Russell-Bennett, R. (2019). Are
households ready to engage with smart home technology?. Journal of Marketing
Management, 35(15-16), 1370-1400.
Nuñez, M. (2019). Facebook removes hundreds of fake pro-Trump accounts using AI-
generated profile photos. Retrieved from:
https://www.forbes.com/sites/mnunez/2019/12/20/facebook-removes-hundreds-of-fake-
pro-trump-accounts-using-ai-generated-profile-photos/#170794956175
Pennycook, G., & Rand, D. G. (2019). Lazy, not biased: Susceptibility to partisan fake news
is better explained by lack of reasoning than by motivated reasoning. Cognition, 188,
39-50.
Powers, K. (2019). Virtual influencers are becoming more real—here’s why brands should be
cautious. Retrieved from: https://www.ama.org/marketing-news/virtual-influencers-are-
becoming-more-real-heres-why-brands-should-be-cautious/
Rauschnabel, P. A., & Ahuvia, A. C. (2014). You’re so lovable: Anthropomorphism and
brand love. Journal of Brand Management, 21(5), 372-395.
Richards, R. (2020). Deepfake technology unlocks real stories of LGBTQ persecution in
Welcome to Chechnya. https://theconversation.com/deepfake-technology-unlocks-real-
stories-of-lgbtq-persecution-in-welcome-to-chechnya-144053
Saito, Y., Takamichi, S., & Saruwatari, H. (2017). Statistical parametric speech synthesis
incorporating generative adversarial networks. IEEE/ACM Transactions on Audio,
Speech, and Language Processing, 26(1), 84-96.

35
Schuster, L., Proudfoot, J., & Drennan, J. (2015). Understanding consumer loyalty to
technology-based self-services with credence qualities. Journal of Services Marketing,
29(6/7), 522-532.
Seo, K., & Dillard, J. (2019). The persuasive effects of two stylistic elements: Framing and
imagery. Communication Research, 46(7), 891-907.
Shah, D., Rust, R. T., Parasuraman, A., Staelin, R., & Day, G. S. (2006). The path to
customer centricity. Journal of Service Research, 9(2), 113-124.
Sherwin, R. K., Feigenson, N., & Spiesel, C. (2006). Law in the digital age: How visual
communication technologies are transforming the practice, theory, and teaching of law.
Boston University Journal of Science & Technology Law, 12(2), 227-270.
Spry, A., Pappu, R., & Cornwell, B. T. (2011). Celebrity endorsement, brand credibility and
brand equity. European Journal of Marketing, 45(6), 882-909.
Stafford, M. R., Stafford, T. F., & Day, E. (2002). A contingency approach: The effects of
spokesperson type and service type on service advertising perceptions. Journal of
Advertising, 31(2), 17-35.
Stupp, C. (2019, August 30). Fraudsters used AI to mimic CEO’s voice in unusual
cybercrime case. The Wall Street Journal. Retrieved from:
https://www.wsj.com/articles/fraudsters-use-ai-to-mimic-ceos-voice-in-unusual-
cybercrime-case-11567157402
Sundar, S. S. (2008). The MAIN model: A heuristic approach to understanding technology
effects on credibility. In M. J. Metzger & A. J. Flanagin (Eds.), Digital media, youth,
and credibility (pp. 73–100). The MIT Press.
The Dali Museum. (2019, May 08). Behind the scenes: Dali lives. [Video file].
https://www.youtube.com/watch?v=BIDaxl4xqJ4
Till, B. D. (1998). Using celebrity endorsers effectively: Lessons from associative learning.
Journal of Product and Brand Management, 7(5), 400-407.
Tractica. (2019). Artificial intelligence software market to reach $118.6 billion in annual
worldwide revenue by 2025. Retrieved from:
https://tractica.omdia.com/newsroom/press-releases/artificial-intelligence-software-
market-to-reach-118-6-billion-in-annual-worldwide-revenue-by-2025/
Vaccari, C., & Chadwick, A. (2020). Deepfakes and disinformation: Exploring the impact of
synthetic political video on deception, uncertainty, and trust in news. Social Media +
Society, 6(1), 1-13.
van Esch, P., Black, J. S., Franklin, D., & Harder, M. (2020). AI-enabled biometrics in
recruiting: Insights from marketers for managers. Australasian Marketing Journal. In
press.
Whittaker, L. W., Kietzmann, T. C., Kietzmann, J. & Dabirian, A. (2020). “All around me are
synthetic faces”: the mad world of AI-generated media. IT Professional, 22(5), 90-99.
Williams, R. (2020). Doritos lets mobile fans create Super Bowl dance videos with AI app.
https://www.mobilemarketer.com/news/doritos-lets-mobile-fans-create-super-bowl-
dance-videos-with-ai-app/571361/
Wilson, H. J., & Daugherty, P. R. (2018). Collaborative intelligence: humans and AI are
joining forces. Harvard Business Review, 96(4), 114-123.

36
Woollacott, E. (2019). China bans deepfakes in new content crackdown. Retrieved from:
https://www.forbes.com/sites/emmawoollacott/2019/11/30/china-bans-deepfakes-in-
new-content-crackdown

37