Unlocking Adaptive User Experience with Generative AI

Yutan Huang

, Tanjila Kanij

, Anuradha Madugalla

, Shruti Mahajan, Chetan Arora

and John Grundy

Department of Software Systems and Cybersecurity, Monash University, Clayton, Melbourne, Australia

Keywords:

Adaptive UI/UX, User-Centered Designs, Generative AI, ChatGPT, Persona.

Abstract:

Developing user-centred applications that address diverse user needs requires rigorous user research. This is

time, effort and cost-consuming. With the recent rise of generative AI techniques based on Large Language

Models (LLMs), there is a possibility that these powerful tools can be used to develop adaptive interfaces.

This paper presents a novel approach to develop user personas and adaptive interface candidates for a speciﬁc

domain using ChatGPT. We develop user personas and adaptive interfaces using both ChatGPT and a tradi-

tional manual process and compare these outcomes. To obtain data for the personas we collected data from

37 survey participants and 4 interviews in collaboration with a not-for-proﬁt organisation. The comparison of

ChatGPT generated content and manual content indicates promising results that encourage using LLMs in the

adaptive interfaces design process.

1 INTRODUCTION

Designing and developing user-friendly web inter-

faces is both an art and science. User Interface (UI)

and User Experience (UX) design requires striking a

balance between aesthetic appeal and meeting func-

tional and non-functional requirements. The require-

ments will be based on a deep understanding of user

behaviour, expectations and special ‘human-centric’

needs. These needs are often diverse and individu-

alised (Benyon, 2019). The traditional design pro-

cesses, while structured and methodical, are charac-

terised by their resource-intensive nature. They often

require input from multidisciplinary teams (domain

experts, software engineers and UI/UX designers) and

provides solutions that may not effectively address the

individual requirements of diverse users (Lewis and

Sauro, 2021). In addition, there is a growing need for

swift time-to-market of web applications (web apps),

which has led to a pressing demand for innovative,

efﬁcient and more adaptable UI/UX design method-

ologies in the industry (Main and Grierson, 2020).

The recent advancements in artiﬁcial intelligence

https://orcid.org/0000-0002-6239-9665

https://orcid.org/0000-0002-5293-1718

https://orcid.org/0000-0002-3813-8254

https://orcid.org/0000-0003-1466-7386

https://orcid.org/0000-0003-4928-7076

(AI) techniques, particularly generative AI and large

language models (LLMs) like ChatGPT, mark the be-

ginning of potential new innovations in several areas,

including UI/UX designs. These technologies offer a

promising shift toward more personalised and adapt-

able design processes, paving the way for innovative

solutions to longstanding issues. (Wang et al., 2023;

Nguyen-Duc et al., 2023).

In this new ideas paper we introduce a new ap-

proach to develop user personas and UI/UX for web

applications by leveraging the generative capabilities

of ChatGPT. We explore generating an adaptive UI

with LLMs by providing user personas as an input to

the design process. We then compare these results

with another set of adaptive UIs that were manually

developed based on rigorous user research. We com-

pare ChatGPT-generated outcomes with those crafted

through conventional user research and design prac-

tices. We show that the trialled approach offers a pos-

sibility of a ‘true’ human-centred design of web apps

that could streamline design processes and tailor user

experiences more closely to individual needs. The re-

sults reﬂects an anticipating stance, inviting further

research and discussion from the community and aims

to set stage for future empirical research to extend and

validate our preliminary ﬁndings. This paper is organ-

ised as follows: Section 2 reviews existing literature,

Section 3 outlines our research design and methodol-

760

Huang, Y., Kanij, T., Madugalla, A., Mahajan, S., Arora, C. and Grundy, J.

Unlocking Adaptive User Experience with Generative AI.

DOI: 10.5220/0012741000003687

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 19th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE 2024), pages 760-768

ISBN: 978-989-758-696-5; ISSN: 2184-4895

ogy, Section 4 describes our preliminary results and

presents a comparative analysis of the results. Sec-

tion 5 presents a SWOT analysis, reﬂecting on the

strengths, weaknesses, opportunities, and threats of

using LLMs in the design process based on our out-

comes. Section 7 presents our conclusions and fu-

ture research direction ideas for using generative AI

in adaptive UI/UX design.

2 RELATED WORK

2.1 Generative AI

ChatGPT. Generative Pre-trained Transformer

(GPT) is an autoregressive large language model

that produces human-like text through deep learning

(Atlas, 2023). The latest version released in 2023,

GPT 3.5 has been shown to achieve a signiﬁcant

improvement compared to its previous GPT models

(Talebi et al., 2023). ChatGPT platform developed by

OpenAI uses GPT 3.5 and is considered one of the

best LLMs (Magar and Schwartz, 2022). As such we

selected ChatGPT 3.5 as the LLM for this study.

Prompt Engineering with ChatGPT. Prompt engi-

neering refers to the set of instructions provided to

LLMs to receive responses and serves as an essential

process in leveraging ChatGPT to generate meaning-

ful and contextually relevant outputs for UI/UX de-

sign and other SE activities (Liu et al., 2023). The

OpenAI Playground, a popular tool for experiment-

ing with prompts and examples, provides signiﬁcant

value to users. However, there is currently little guid-

ance on systematically crafting prompts in a rigorous

manner (Arora et al., 2023). Consequently, we have

incorporated prompting techniques based on rigorous

evaluation by the researchers, our methodology in-

volved a deliberate and iterative approach to crafting

prompts that would lead to the persona generation and

web page designs that aligns closely with user needs.

For prompt engineering with ChatGPT, we used the

output customisation and error identiﬁcation category

from the catalogue suggested by White et al. (White

et al., 2023a). The crafting of prompts for Chat-

GPT ensures creating personas and web pages that

are comparable, if not superior, to those developed

through traditional methods.

2.2 Persona

A persona is a ﬁctional character that reﬂects the char-

acteristics of clusters of real end users, they are of-

ten utilised for analysing end-user requirements and

ensuring that the proposed software products ulti-

mately meet these standards (Karolita et al., 2023).

There are no universally accepted methods for cre-

ating personas, and researchers have proposed vari-

ous approaches to develop personas for user groups.

These approaches fall into three categories: qualita-

tive, quantitative, and mixed (Tu et al., 2010). Qual-

itative methods for persona creation depend on ex-

ploratory research with a medium-sized sample of

users. This method involves users at all stages of de-

sign, focuses on understanding and analysing user be-

havior, and iterates the creation of personas (Hosono

et al., 2009). On the other hand, quantitative methods

aim to leverage user data from diverse sources to con-

struct personas, thereby improving understanding of

users. Mixed methods combine elements from both

quantitative and qualitative approaches, often target-

ing speciﬁc user demographics, such as groups de-

ﬁned by age (Cooper et al., 2007).

2.3 Nielsen’s Persona Method

We employ a generalised mixed method for creat-

ing personas, with Nielsen’s ten-step approach as

our manual method (Nielsen, 2004). We selected

Nielsen’s approach as our information does not align

perfectly with either the qualitative or quantitative ap-

proaches. By following Nielsen’s structured process,

we ensure that our personas are well-founded, accu-

rately reﬂecting the complex needs, attitudes, and be-

haviors of our target user groups.

Nielsen’s ten-step approach to creating personas

encompasses three crucial domains: data collection,

engagement with persona descriptions, and organi-

sational buy-in. Initially, it involves gathering data

from various sources to understand target users, in-

cluding the methods for data collection and storage.

After collecting data, an initial hypothesis about the

users is formed to guide the persona creation pro-

cess. This hypothesis is then tested by verifying the

data. As patterns within user groups emerge, these

groups are categorised, leading to the construction

of personas equipped with detailed backgrounds and

personal traits. It’s essential to place these personas

in speciﬁc scenarios and validate them with real-life

users to start creating scenarios that are relevant to the

personas. Once validated, these personas are shared

with organisations to assist developers. The ﬁnal step

involves crafting a narrative for the persona, detailing

scenarios that highlight the persona’s goals.

Unlocking Adaptive User Experience with Generative AI

761

3 RESEARCH DESIGN

3.1 Methodology Overview

We want to answer the following key research ques-

tions (RQs):

1. RQ1 - How do personas developed by LLMs per-

form against the manually created personas?

2. RQ2 - What is the quality of web pages developed

with the use of LLMs?

3. RQ3 - What is the effectiveness of the customized

UI/UX in web pages developed by LLMs based

on speciﬁc personas?

To address these RQs, we adopted a mixed-

method research design, integrating quantitative data

from surveys with qualitative insights from inter-

views. This blended approach facilitated a compre-

hensive comparison between the innovative use of

ChatGPT in generating personas and web pages, and

the traditional manual methodologies employed in

UI/UX design. The research was designed in a phased

approach, with distinct stages aimed at directly ad-

dressing each research question: Persona develop-

ment phase - This examines the depth, accuracy and

applicability of ChatGPT-generated personas in com-

parison to those crafted manually, which addresses

RQ1. Webpage design phase -This addresses RQ2

by evaluating the quality of ChatGPT-developed web

pages with user representatives, focusing on their de-

sign principles, aesthetic appeal, as well as user en-

gagement metrics. UI/UX customization phase -

This addresses RQ3 by exploring the effectiveness

of ChatGPT in tailoring web pages to speciﬁc user

personas, assessing ChatGPT’s adaptability and pre-

cision in meeting user-speciﬁc design requirements.

3.2 Survey and Interview

We partnered with an NFP organisation and con-

ducted a survey (with 37 participants) and interviews

(with four participants) to obtain the user preferences

and needs for their website. The NFP is a commu-

nity support organisation providing health and well-

ness support to users, including organising events and

workshops that they advertise through the website (in

question) and wanted people to register through the

website as well. Participants were recruited by con-

venience sampling with the help of the partnered or-

ganisation. The survey setup with Google Forms, col-

lected data on demographic information, the daily us-

age of websites and web applications, and their feed-

back on the website of the partner organisation. In-

terviews were conducted through the Zoom platform

where participants were asked to explore the organi-

sation’s website and provide suggestions to improve

the website design. They were then instructed to

write the features and information they desired on

empty stickers on the Miro board, an online collab-

orative whiteboard platform. The insights gathered

inform the criteria for evaluating the effectiveness of

ChatGPT-generated outputs against manual efforts,

they were instrumental in guiding the development of

user personas, ensuring the outputs of our study to be

grounded in real-world environments.

4 RESULTS

4.1 RQ1 Results

4.1.1 Persona Development with ChatGPT

For persona development, we implemented a compre-

hensive prompt engineering strategy that involved the

integration of user scenarios and characteristics de-

rived from preliminary user research (Zhang et al.,

2023). Our design process recognises the importance

of speciﬁcity in prompts to obtain high-quality out-

puts; we utilised the “output customisation and er-

ror identiﬁcation” pattern as suggested by White et

al. (White et al., 2023a) to reﬁne our prompts based

on initial feedback loops with our model. An ex-

ample prompt for persona creation was structured as

follows: Given a not-for-proﬁt organisation aiming

to increase engagement among its diverse user base,

create a persona representing a typical user. Con-

sider age, interests, digital literacy, and potential en-

gagement barriers. The persona should reﬂect the or-

ganisation’s focus on health and wellness support, in-

cluding attending workshops and events. This prompt

was designed to encapsulate the diverse user base and

the organisation’s mission, ensuring the personas gen-

erated were diverse and represented real-world users.

We then used iterative reﬁnement to adjust the level of

detail and scope of the personas, ensuring they served

as foundations for further design decisions. A persona

generated by ChatGPT using our prompt is given on

the left-hand side in Figure 1.

4.1.2 Persona Development from Survey

We applied Nielsen’s approach for persona develop-

ment from the survey data (Nielsen, 2004). Follow-

ing the approach, we found three main clusters of user

facets from the survey results. Cluster 1: Consists of

end-users from the age group 56-65, who are mostly

retired individuals with various cultural backgrounds.

ENASE 2024 - 19th International Conference on Evaluation of Novel Approaches to Software Engineering

762

Figure 1: Persona examples (left) generated by ChatGPT and (right) by following Nielsen’s approach manually.

They experience difﬁculties navigating websites and

prefer bigger texts and simple layouts; they browse

websites for news media and register for community

activities on the organisation’s website. Cluster 2:

Comprises users aged 26-35; most come from Chi-

nese cultural backgrounds and various occupations.

They experience little difﬁculty navigating websites

and, like content-rich websites with a range of vi-

brant colours, they often register for workshops and

social activities on the organisation’s website. Clus-

ter 3: Consists of end-users from the age group of

18-26, who are primarily students from Asian cul-

tural backgrounds. They browse websites with no

difﬁculties and prefer a monochromatic colour for

websites; they register and attend stress management

workshops speciﬁcally on the organisation’s website.

Three personas were developed from each cluster;

named Jack, Linda and James. Linda persona is pre-

sented on the right-hand side in Figure 1.

4.1.3 Comparative Analysis

We summarise our comparison of manually devel-

oped personas vs ChatGPT generated personas in Ta-

ble 1. The demographics displayed signiﬁcant simi-

larities, albeit with variations in titles. The manually-

created personas included visual avatars and notes on

cultural distinctions, enhancing their relatability. In

contrast, the ChatGPT personas offered an in-depth

exploration of the goals and needs speciﬁc to each

persona, attached with a dedicated section titles ’How

Organisation X’s Website Helps,’ which mapped the

website’s functions to the persona’s daily life.

The manual approach reﬂected a broad classiﬁca-

Table 1: Persona features developed manually and with

ChatGPT (*MP - Manual Persona, *CP- ChatGPT Per-

sona).

Category Feature MP* CP*

Demographics

Age ✓ ✓

Gender ✓ ✓

Cultural Afﬁliation ✓ ✗

Work ✓ ✓

Location ✓ ✓

Family ✓ ✓

Background ✓ ✓

Motivations

Goals ✓ ✓

Challenges ✗ ✓

Frustrations ✓ ✗

User Needs and

Preferences

Website preferences ✓ ✗

Colour preferences ✓ ✗

User Environment ✓ ✓

Organisation X’s role ✗ ✓

tion suitable for larger user groups and the ChatGPT

personas suggested a more personalised UI/UX de-

sign approach. This suggests that generative AI can

complement user research methods by providing a

deeper, data-driven understanding of user segments

with highlighted unique user preferences and engage-

ment barriers. It also highlights the need for a com-

bination of AI efﬁciency with human-centric insights.

We aim to explore this new research area more in our

future studies.

4.2 RQ2 Results

4.2.1 Webpage Development with ChatGPT

To develop webpages for our partner NFP organisa-

tion, we followed a two-step prompt engineering pro-

cess. In the ﬁrst step, we furnished ChatGPT with

Unlocking Adaptive User Experience with Generative AI

763

Figure 2: Webpage created by ChatGPT.

essential background information regarding the or-

ganisation, website structure, and details about each

webpage. For example, we provided the details of

the purpose and functionality of the NFP organisa-

tion’s website, how many web pages they have, and

the description and purpose of each web page, struc-

turing prompts that described the desired features and

content of each page clearly. In the second step, we

provided a clear and speciﬁc prompt for the desired

output, i.e., the generation of HTML scripts and CSS

features for the website designs. In these prompts

we provided the following information to ChatGPT.

Background Information -“Create a website for a

not-for-proﬁt organisation that provides workshops

and events to enrich community health and wellness.

The website aims to provide support and care for

the community along with in-person services.” Web-

site Information - “The website should have six web

pages: A home page, an About page, a Courses page,

a Services page, a Resources page, and a Donate page

with a consistent colour scheme.” Home Page - “For

the home page: A “Menu” option should be available

on either the top or the side of the page; a consistent

color scheme should be chosen for the website; a hero

feature, which is a unique feature to the home page,

should be included.”

We provided detailed speciﬁcations in addition to

the prompt template to complete the webpage gener-

ation process. We were able to guide ChatGPT to-

wards generating webpage designs that were not only

appealing but also functional and aligned with the or-

ganisation’s goals by iterating on the prompts based

on initial outputs. A snapshot of the homepage gener-

ated by ChatGPT is presented in Figure 2.

4.2.2 Website Development from Interviews

Participants were requested to visit organisation’s

website to make the interview questions consistent

with the prompt engineering for ChatGPT. The in-

terview questions were crafted based on the analy-

sis of survey responses. The participants were in-

quired about their preferences regarding the website’s

landing page, its level of dynamism, customisability,

ENASE 2024 - 19th International Conference on Evaluation of Novel Approaches to Software Engineering

764

and how information is represented. Based on their

preference, the review of the website’s content and

UI/UX, they were asked to share their design insights

on Miroboard. One example design created on the

MiroBoard is given in Figure 3.

Figure 3: Miroboard webpage design.

4.2.3 Comparative Analysis

We compared key differences and similarities be-

tween web pages manually designed by our intervie-

wees, visualised through Miroboard, and those Chat-

GPT generated. One signiﬁcant observation is Chat-

GPT’s consistent inclusion of explicit title speciﬁca-

tions, a detail frequently overlooked by out intervie-

wees. This aspect highlights ChatGPT’s capability to

maintain essential web design elements from prompts

even when they are not directly mentioned, which ad-

heres to structured content creation.

Our analysis revealed limitations in ChatGPT’s

ability to meet speciﬁc user expectations, notably the

absence of anticipated elements e.g a video on ‘about

page’ and contact details on the ‘donation page’. This

show a challenge in incorporating implicit user pref-

erences. Additionally, aligning ChatGPT’s designs

with detailed user requirements, like clear naviga-

tion labels, reduced scrolling, and informational pop-

ups, proved difﬁcult. However, ChatGPT’s designs

demonstrated greater structure and consistency, bene-

ﬁting from user-centered design principles. Features

like interactive buttons that change color upon hover-

ing enhanced the user experience.

4.3 RQ3 Results

In addressing RQ3, we provided ChatGPT with the

three personas we developed manually based on our

survey. This step aimed to understand how well Chat-

GPT can customise the UI/UX based on the user

groups described by real-life personas. We provided

one persona at a time and found ChatGPT developed

web pages with the same content but differing colour

themes and styles. Figure 4 presents a snapshot of

three web pages ChatGPT developed for Jack, Linda

and James, respectively. Web pages for Linda and

James used colour themes consistent with their colour

preferences. However, Jack’s persona had hyperlinks

instead of interactive buttons with no colour theme.

This can be because Jack belongs to an elderly adult

group, and web pages a few decades ago were mainly

designed with hyperlinks.

Furthermore, we gathered all manually developed

personas and asked ChatGPT to develop a webpage

for them as a whole. This resulted in ChatGPT keep-

ing the structural design features like interactive but-

tons and menu options shown in Figure 4. However, it

integrated the colour preferences of all personas and

developed a light green-yellow colour for interactive

buttons and menu options, with the text colour chang-

ing to blue when hovering.

These results indicate that ChatGPT is capable of

cuztomizing web pages to to reﬂect distinct user pro-

ﬁles, adapting to design elements like color themes

and navigation features, which highlights the signif-

icance of LLMs in enabling user-sentric design ap-

proaches.

5 DISCUSSION

LLM based Generative AI tools such as ChatGPT, of-

fer features that can be used in many development

tasks, such as developing adaptive UI/UX. However,

these opportunities are yet to be explored in detail.

Bartao and Joo found that UI/UX developers do not

widely use AI tools (Bert

ao and Joo, 2021). The re-

cent research which incorporated ChatGPT in proto-

type designing found some beneﬁts of the approach;

but received mixed responses from developers (Ek-

vall and Winnberg, 2023). Another research with an

older version of GPT found that incorporating LLMs

in earlier prototyping stages can save effort and cost

(Bilgram and Laarmann, 2023). However, none of

the research systematically compared ChatGPT’s out-

come at each step based on speciﬁc prompt engi-

neering with a traditional manual process. Our pre-

liminary attempt to do a comparative analysis shows

several promising opportunities. The persona devel-

oped by ChatGPT based on domain speciﬁcation (in

this case - the NFP organisation) was detailed enough

and even contained a section exploring websites in-

tegration to Persona user’s life, which the manually

developed persona didn’t have. The differences in

cultural background and other preferences presented

with three manually developed personas were absent

in one persona developed by ChatGPT. We think this

Unlocking Adaptive User Experience with Generative AI

765

Table 2: Features Comparison.

Categories Website Features Website developed

by ChatGPT

Miroboard Design 1 Miroboard Design 2 Miroboard Design 3

Home Page

Title ✓ ✓ ✗ ✓

Menu options ✓ ✓ ✓ ✓

Content Highlights ✗ ✓ ✓ ✓

Courses & Events description ✓ ✗ ✗ ✓

Simple Words for navigation ✗ ✓ ✗ ✓

About Page

Title ✓ ✗ ✗ ✓

Vision ✓ ✓ ✗ ✗

Motivation ✓ ✓ ✗ ✓

Videos ✗ ✓ ✓ ✓

Courses Page

Title ✓ ✓ ✗ ✓

Courses & Events details ✓ ✗ ✓ ✓

Registration ✓ ✓ ✓ ✓

Less Screen Scrolling ✗ ✓ ✓ ✗

Sub menu for course separations ✗ ✓ ✓ ✓

Resources Page

Title ✓ ✗ ✗ ✗

Resource categorization ✗ ✓ ✓ ✓

Details of resources ✓ ✗ ✓ ✓

Services Page

Title ✓ ✓ ✗ ✗

Pop-ups for expert information ✗ ✓ ✗ ✗

More appealing titles ✗ ✓ ✗ ✓

Service categorization ✗ ✓ ✓ ✓

Details of services ✓ ✗ ✗ ✓

Donate Page

Title ✓ ✗ ✗ ✗

Contact details ✗ ✓ ✓ ✓

Donation Methods ✓ ✗ ✓ ✓

Figure 4: Webpages developed by ChatGPT with Personas.

can be alleviated with speciﬁc instructions to develop

more than one persona.

Regarding the webpages, ChatGPT developed

consistent designs with different colour themes for

different personas. The style was also adapted to

classical style for the elderly persona. These ﬁnd-

ings indicate the opportunities ChatGPT presents to

customize UI/UX based on speciﬁc user needs and

preferences. Prompts can be speciﬁcally tailored to

achieve these outcomes. However, tailoring prompts

can be challenging and the heavy dependence of the

outcome on prompt engineering poses a threat to the

approach.

Our initial ﬁndings indicate promising results for

using ChatGPT to develop user persona and UI/UX

customized towards the persona. This can help avoid

extensive user research to understand user needs and

preferences and to develop UI/UX customized for the

users. Based on the ﬁndings, we developed a SWOT

matrix for developing UI/UX based on customized

user needs and preferences with the help of ChatGPT,

shown in Figure 5.

Prompt engineering is critical for ChatGPT, as the

quality of prompts directly affects the relevance and

quality of responses generated by ChatGPT (Yu et al.,

2023). This is particularly evident in the persona de-

velopment process, where we found that minor de-

viations in prompt choices can lead to substantial

deviations in persona contexts (White et al., 2023b;

Ubani et al., 2023). We also learned that prompts

need to be provided within speciﬁc structural frames

to achieve desired responses, without which ChatGPT

would only produce the simplest forms of responses.

Therefore, rigorous prompts are necessary for devel-

oping personas and web pages using ChatGPT.

6 THREATS TO VALIDITY

A potential threat to our ﬁndings are in the evalua-

tion method as there may be a possibility of subjec-

tive bias in qualitative assessments within our study.

ENASE 2024 - 19th International Conference on Evaluation of Novel Approaches to Software Engineering

766

Figure 5: SWOT analysis of ChatGPT.

To mitigate this threat, we plan to reﬁne our evalu-

ation process by involving domain experts and user

representatives. Additionally, we will employ a blind

review process among the evaluators to ensure they

are unaware of whether the webpages and personas

were generated by ChatGPT or developed manually.

Another threat is that our results may be limited by

the speciﬁc context of our study—a health and well-

ness NFP organization. It is necessary for us to con-

duct further research across diverse domains to vali-

date the broad applicability of our ﬁndings.

7 CONCLUSION AND FUTURE

PLANS

We investigated whether ChatGPT can develop adap-

tive UI/UX. From our experiments, persona and web-

site development with LLMs can be more efﬁcient

with tailored prompts being used. LLMs can gener-

ate desired outputs for developers in a short time, also

providing more details and insights for outputs. The

traditional approach of using quantitative and qualita-

tive user study is time consuming, but effective for de-

veloping lightweight personas and websites. We plan

to broaden our research to encompass additional as-

pects of generative AI in UI/UX design. We will work

towards creating a robust framework to automate per-

sona and website development, aiming to capture crit-

ical user-centric content that aids designers. Such a

framework will not only streamline the design pro-

cess but also serve to ﬁne-tine and personalise many

interactivity elements in UI/UX designs.

ACKNOWLEDGEMENTS

Kanij, Madugalla and Grundy are supported by ARC

Laureate Fellowship FL190100035.

REFERENCES

Arora, C., Grundy, J., and Abdelrazek, M. (2023). Ad-

vancing requirements engineering through genera-

tive ai: Assessing the role of llms. arXiv preprint

arXiv:2310.13976.

Atlas, S. (2023). Chatgpt for higher education and profes-

sional development: A guide to conversational ai.

Benyon, D. (2019). Designing user experience. Pearson

UK.

Bert

ao, R. A. and Joo, J. (2021). Artiﬁcial intelligence in

ux/ui design: a survey on current adoption and [future]

practices. Safe Harbors for Design Research, pages

1–10.

Bilgram, V. and Laarmann, F. (2023). Accelerating inno-

vation with generative ai: Ai-augmented digital pro-

totyping and innovation methods. IEEE Engineering

Management Review.

Cooper, A., Reimann, R., and Cronin, D. (2007). About face

3: the essentials of interaction design. John Wiley &

Sons.

Ekvall, H. and Winnberg, P. (2023). Integrating chatgpt into

the ux design process: Ideation and prototyping with

llms.

Hosono, S., Hasegawa, M., Hara, T., Shimomura, Y., and

Arai, T. (2009). A methodology of persona-centric

service design. In Proceedings of the 19th CIRP De-

sign Conference–Competitive Design.

Karolita, D., Grundy, J., Kanij, T., Obie, H., and McIntosh,

J. (2023). What’s in a persona? a preliminary taxon-

omy from persona use in requirements engineering. In

International Conference on Evaluation of Novel Ap-

proaches to Software Engineering 2023, pages 39–51.

Lewis, J. R. and Sauro, J. (2021). USABILITY AND USER

EXPERIENCE: DESIGN AND EVALUATION, chap-

ter 38, pages 972–1015. John Wiley & Sons, Ltd.

Liu, P., Yuan, W., Fu, J., Jiang, Z., Hayashi, H., and Neubig,

G. (2023). Pre-train, prompt, and predict: A system-

atic survey of prompting methods in natural language

processing. ACM Computing Surveys, 55(9):1–35.

Magar, I. and Schwartz, R. (2022). Data contamination:

From memorization to exploitation. arXiv preprint

arXiv:2203.08242.

Main, A. and Grierson, M. (2020). Guru, partner, or pencil

sharpener? understanding designers’ attitudes towards

intelligent creativity support tools. arXiv preprint

arXiv:2007.04848.

Nguyen-Duc, A., Cabrero-Daniel, B., Przybylek, A., Arora,

C., Khanna, D., Herda, T., Raﬁq, U., Melegati, J.,

Guerra, E., Kemell, K.-K., et al. (2023). Generative

artiﬁcial intelligence for software engineering–a re-

search agenda. arXiv preprint arXiv:2310.18648.

Nielsen, L. (2004). Engaging personas and narrative sce-

narios.

Talebi, S., Tong, E., and Mofrad, M. R. (2023). Beyond

the hype: Assessing the performance, trustworthiness,

and clinical suitability of gpt3. 5. arXiv preprint

arXiv:2306.15887.

Tu, N., He, Q., Zhang, T., Zhang, H., Li, Y., Xu, H., and Xi-

ang, Y. (2010). Combine qualitative and quantitative

Unlocking Adaptive User Experience with Generative AI

767

methods to create persona. In 2010 3rd International

Conference on Information Management, Innovation

Management and Industrial Engineering, volume 3,

pages 597–603.

Ubani, S., Polat, S. O., and Nielsen, R. (2023). Zeroshot-

dataaug: Generating and augmenting training data

with chatgpt. arXiv preprint arXiv:2304.14334.

Wang, W., Khalajzadeh, H., Grundy, J., Madugalla, A.,

McIntosh, J., and Obie, H. O. (2023). Adaptive user

interfaces in systems targeting chronic disease: a sys-

tematic literature review. User Modeling and User-

Adapted Interaction, pages 1–68.

White, J., Fu, Q., Hays, S., Sandborn, M., Olea, C., Gilbert,

H., Elnashar, A., Spencer-Smith, J., and Schmidt,

D. C. (2023a). A prompt pattern catalog to enhance

prompt engineering with chatgpt. arXiv preprint

arXiv:2302.11382.

White, J., Hays, S., Fu, Q., Spencer-Smith, J., and Schmidt,

D. C. (2023b). Chatgpt prompt patterns for improving

code quality, refactoring, requirements elicitation, and

software design. arXiv preprint arXiv:2303.07839.

Yu, F., Quartey, L., and Schilder, F. (2023). Exploring the

effectiveness of prompt engineering for legal reason-

ing tasks. In Findings of the Association for Compu-

tational Linguistics: ACL 2023, pages 13582–13596.

Zhang, X., Liu, L., Wang, Y., Liu, X., Wang, H., Ren, A.,

and Arora, C. (2023). Personagen: A tool for gen-

erating personas from user feedback. In 2023 IEEE

31st International Requirements Engineering Confer-

ence (RE), pages 353–354.

ENASE 2024 - 19th International Conference on Evaluation of Novel Approaches to Software Engineering

768