INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 521

www.rsisinternational.org

A Comparative Genre Analysis of Human-Written and Ai-Generated

Research Abstracts

Puteri Zarina Megat Khalid, Haddi@Junaidi Kussin, Nurul Farehah Mohamad Uri, Aireen Aina

Bahari, Khazaila Zaini, Nur Aliaa Amirah Kasuahdi

Department of English Language and Literature, Faculty of Languages and Communication Universiti

Pendidikan Sultan Idris Perak, Malaysia

*Corresponding Author

DOI: https://dx.doi.org/10.47772/IJRISS.2025.910000044

Received: 02 October 2025; Accepted: 08 October 2025; Published: 03 November 2025

ABSTRACT

This comparative study explores the distinctive generic features between abstracts written by human authors

and those generated by artificial intelligence tools through the genre analysis methods. Over-reliance on such

external mechanisms and plagiarism are perennial issues that affect the global academia particularly in regards

to the widespread dependence on artificial intelligence. The study compares ten research abstracts written by

postgraduate master students specialising in English Language and Literature from a Malaysian public

university to AI-generated abstracts produced using Chat Generative Pre-Trained Transformer 3, also known

as ChatGPT. The study looks into the frequency and quality of key elements or moves such as statements of

objectives, methods, results, and contextualisation to determine their recurrence patterns. Findings indicate that

human-inscribed abstracts reveal a more stable and thorough presentation, highlighting contextualisation and

inclusive results, while AI-generated abstracts possess clarity in statements of objectives with minimal

coverage on results and contextual details. The findings in this research thus recommend for the development

of an innovative method of detecting AI-generated content written by students using the genre analysis

approach. It also emphasises the necessity for specialised teacher training and rigorous evaluation criteria to

preserve academic integrity and overcome the limitations of using AI in academic writing.

Keywords: Genre analysis; AI-generated writings; comparative analysis; human-writers; research abstracts

INTRODUCTION

The plagiarism detection analysis of students’ academic texts can be performed using various software and

applications. One such example of software is the Turnitin which is an established platform to detect

plagiarism committed by students in their scholarly texts by comparing the texts to the vast online database of

academic sources. Correspondingly, this research adopts a genre analysis approach to compare and contrast the

characteristics of both human-inscribed texts and those generated by tools of Artificial Intelligence (AI). Just

as the plagiarism indicators in a text such as linguistic and semantic resemblances are identified by the Turnitin,

the genre analysis method that this study employs scrutinises the generic structures and technical nuances of

abstracts in final year project papers. Thus, it is crucial to distinguish the structural and linguistic patterns of

texts organically produced by humans and those generated by AI.

The goal is to distinguish between structural and linguistic patterns in texts written by humans and those

created by AI. Genre analysis can be used to study the intricacies of human communication, like written texts.

This method is crucial to examining textual structures and conventions as it provides a systematic way in

exploring how genres work in different socio-cultural settings. The current technological advancement has

engendered the debate regarding the variations between human and AI-generated academic texts. Genre

analysis is important in literacy education because it helps people understand discourse, which is a basic

learning tool. It thus avails applied linguists a socially validated theory of language and an educational

methodology obtained from research into texts and their contexts (Rakrak, 2025). Recent genre analysis

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 522

www.rsisinternational.org

studies revealed that the levels of coherence and diversity in genres could be analysed by studying their

structure and understanding the social processes related to them (Thane, 2024; Tan, 2024). These studies

highlight the necessity to identify lexico-grammatical patterns and rhetorical strategies employed by students

in texts. Alqudah (2024) notes that this is crucial for adopting the genre analysis approach to identify AI-

generated academic texts. AI can mimic the superficial aspects of texts created by humans but not the complex

rhetorical and contextual choices behind human writing. One of the ways to assess the differences between

various types of academic writing is the genre analysis method. This way, there is no need to determine

whether the students employ AI assistance and focus on what is really important. It is important for any author

to learn them in order to be proficient in language and critical literacy techniques. The present study aims to

broaden previous investigations by analysing the genre features of research abstracts authored by humans and

AI. While the findings from our focused, small-scale dataset offer valuable initial insights, we frame this work

as a pilot study that highlights the need for larger-scale validation and provides a transparent methodological

framework for future research in this domain.

Theoretical underpinnings

Understanding how academic writing functions more broadly is crucial for genre analysis. The method was

first proposed by Swales (1990). He demonstrated that the forms of written and spoken language, which exist

to fulfil certain functions of communication within a society, are intentional. This highlights the importance of

genre in academic writing. The rhetorical building blocks used to label texts for genre categorisation are called

“moves” and “steps” (Swales, 1990). In the introductions of research articles, authors generally use several

writing steps such as establishing a research territory, identifying a niche, and occupying the niche. To have an

academic discussion, you must be aware of the discourse structure. Swales’ method is a helpful way to teach

genre-specific writing. Additionally, the method provides a solid basis for identifying AI-generated texts that

imitate human writing structure but lack its rhetorical and social components (Nanola et al., 2025). This

highlights the need to integrate genre-based pedagogy into the language curriculum as a way of promoting

critical literacy (Zhou 2025).

Bhatia (1993) extended genre analysis to professional and academic genres, such as research abstracts and

articles, pointing out that disciplinary practices and technologies of communication are dynamic, thereby

continuously evolving. His work enhanced understanding that academic writing is both structured and flexible.

The underpinning theories are reflected in the Figure 1 below. Genre analysis has been extensively used for the

evaluation of research abstracts. Hyland (2000) showed that writing an abstract follows a move-based model:

purpose, method, result, and conclusion. This format makes abstracts short but still useful, so readers can

quickly understand the main point of a study.

Figure 1: Foundational theories of Genre Analysis in this study

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 523

www.rsisinternational.org

REVIEW OF RELATED LITERATURE

Moves and Steps in Different Types of Scholarly Writing

Swales (1990) defined "moves" as the communicative actions utilised by writers to achieve the goals of a

genre. In the introductions of research articles, common strategies include defining the research area,

assessing prior studies, and identifying a gap. Bhatia (1993) further clarified this concept by introducing

"steps," which are smaller components within moves that support the overall communicative goal. In abstracts,

typical moves involve providing background information, stating the research objective, outlining the

methodology, reporting findings, and drawing conclusions. Recurring move structures are evident across

academic genres. A widely recognised example is the IMRaD format (Introduction, Methods, Results,

Discussion), which organises many research articles. Each section serves a specific role: the Introduction

presents the problem, Methods describe the procedures, Results report findings, and Discussion interprets

those findings and their implications (Alqudah, 2024). The genre analysis via its move identification processes

offers a rigorous investigation of how scholarly texts are structured to serve the intended their communicative

purposes. With a clear understanding of these structures, scholars and researchers can thus be better equipped

to distinguish the texts produced by humans from those generated by AI.

Comparison between human-written texts and AI-generated ones

Several studies have been conducted on distinguishing the writings produced by humans and AI tools.

Frangieh and Abdallah (2024) identified the essays generated by ChatGPT displayed more sophisticated

vocabulary, more succinct sentences, and a professional, objective writing style. They also found that texts

written by humans were far richer in terms of emotions, authenticity, and personal references. There was also

more profound diversity in ideas and stylistics in the students’ writing, whereas the texts generated by

ChatGPT exhibited uniformity with formulaic expressions. These findings are corroborated by Eğin et al.

(2025) who performed a comparative analysis of human-written materials and texts generated by ChatGPT,

BingAI, and Gemini. Their study revealed that human authors produced longer texts with more complex

sentence structures and richer vocabulary. They also found that the AI outputs, on the contrary, depended on

fillers, repetitive stereotypical expressions, and structured but reduced linguistic subtlety. Verbosity and

repetitions were inherent in the texts produced by Gemini which assisted in convenience of detection.

ChatGPT’s outputs, on the other hand, bore close resemblance to human-written texts, posing acute difficulties

to distinguish.

Studies by Hakam et al. (2024) and Muñoz-Ortiz et al. (2024) examine the differences between texts generated

using the AI tools and those authored by humans from varying perspectives. Hakam et al. highlight that AI

outputs are normally lacking in depth, critical analysis, and originality despite the tools being able to generate

texts which coherent and structured. The AI materials were also found to rely more on surface-level

generalisations than practical and substantial arguments. On the contrary, Muñoz-Ortiz et al. adopt a

quantitative linguistic approach to the textual analysis via comparison of human-written news articles to the

texts from multiple large language models. Their findings indicate that the vocabulary and sentence length in

the texts produced by human authors are more varied with varying grammatical patterns, and contain greater

emotional elements. AI-generated texts, on the other hand, are more objective, patented by heavier use of

figures and symbols. Despite their apparent diverge views, both research converge on the perspective that AI-

generated writing is generally devoid of complexity, variability, and analytical nuances of human authors’

expressions. This reinforces the notion that AI merely serves as a complement and not a replacement for

human writers.

All in all, these studies have highlighted the fact that human authorship has not lost its originality and

sophistication despite the clarity of the outputs by the AI tools. This underscores the need for the development

of effective detection tools in order to sustain academic integrity and fidelity.

The Drawbacks of Using AI for Academic Genre Creations

Numerous significant developments have been accomplished using AI in natural language processing and text

creation. One such tool is GPT-3, built by OpenAI, which is able to generate interdisciplinary texts with

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 524

www.rsisinternational.org

coherence and contextual relevance (Barbosa, 2024). Nevertheless, the incorporation of AI in academic text

creations poses numerous setbacks, specifically with regard to regulating academic integrity. In addition,

challenges are also inherent in ensuring the authenticity of academic genres.

Among the major drawbacks is the problems in differentiating human-written texts from AI-generated ones.

Federiv et al. (2024) highlighted that AI-developed texts frequently display particular features, namely

redundant phrasing and insufficient contextual depth. The generic features are formed by the algorithms

employed in the training of AI models. Coherence and clarity are given more weight in the features than

comprehension and originality. The texts generated by ChatGPT-3 are grounded on patterns acquired from a

wide array of training data. Although such a tool can generate texts that replicate human writing, it fails to

engage in great depth with the content or even incorporate previous research meaningfully (Amirjalili et al.,

2024; Maurya & Maurya, 2024). Because AI tools often provide limited contextualisation and lack critical

engagement with the literature, producing rigorous academic content using AI remains a challenge. Another

long-standing concern is the issue of authorship and originality as human writers increasingly rely on machines

(Mazzi, 2024), the boundary between human and AI-written texts becomes blurred. This shift raises ethical

questions about the credibility of AI-generated academic work and its implications for the trustworthiness of

scholarly publications.

It is even harder to use AI in research writing because we need reliable ways for discovering information.

Ghiurău and Popescu (2024) underscore the necessity of improving instruments capable of detecting AI-

generated content by concentrating on recurring structures, superficial analysis, and various linguistic

indicators. A better technique would be to combine automatic text assessments and expert ratings

(Verhulsdonck et al., 2024). Algorithms use machine learning and NLP to rapidly analyse large amounts of

text data. The machines detect regular patterns, style problems, and repetitions that reflect AI-written content

(Arshed et al., 2024). Nevertheless, those machines encounter challenges in comprehending comprehend

rhetorical strategies and genre conventions that are critical for academic writing coherence (Ramazani et al.,

2025). Such an issue is best addressed by human raters as they have the capacity to detect subtleties and other

inconspicuous elements in academic writings, particularly those produced by human authors. This shows the

necessity to incorporate genre analysis and critical literacy into language learning. Such an approach will

ensure that academic writing is not only precise but also contextually and culturally appropriate.

The present research aims to address the issue of over-reliance on AI gadgets for writing by analysing ways of

ensuring that the AI-generated content meets the rigorous academic writing standards. The algorithmic setup

employ machine learning and natural language processing (NLP) for swift identification of large text

collections. They are able to detect regularities in patterns, style problems, and repetitions that may indicate

AI-formulated materials (Arshed et al., 2024). For efficient and effective algorithmic functions that capitalise

on machine learning and NLP, human writers should be trained to master the academic rules and regulations,

genre rules and ways to improve their writing beyond basic issues pertaining to texts. Academic writers often

have specific knowledge of rhetorical methods, reasons for selections, and the appropriateness of the texts

within the relevant disciplines (Zhang, 2023). This understanding is helpful in ascertaining if a text merely

replicates academic writing or whether it genuinely adds value to the subject matter. Detection is made easier

when embedded with human experience. With the inclusion of human knowledge, the AI content is identified

not only through variations of styles but through better content engagement and clear composition skills

(Mobilio, 2024).

The combined execution of text analysis methods and human judgment is a blended approach for achieving

academic integrity (Gupta, 2024). Computerised systems are installed with the scaled capacity for speedy

operations to scrutinise vast quantities of text and provide annotated markers that flag potentially AI-generated

materials. This function facilitates human writers to perform a thorough examination of the marked texts in

making informed conclusions. This process if further guided by the author’s own knowledge and mastery of

the particular academic genres and the related rhetorical strategies.

Human feedback now has the ability to improvise automated models with greater accuracy as the detection of

AI-generated texts is enabled through this blended method. This leads to an enhanced AI-powered text

generation where investment in both automated and human authors yields a more accurate and reliable

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 525

www.rsisinternational.org

detection process which ensures the authenticity and integrity of scholarly writings (Aburass & Abu Rumman,

2024).

Genre analysis is a method to help scholars understand how academic writing is pieced together through strict

adherence to the corresponding rules. This shows the importance of identifying moves and steps in the writing

of scholarly themes. Nevertheless, several setbacks have also been discovered in the utilisation of AI in

scholarly writing. Certain pertinent threats like originality, creativity, and effective detection of AI intervention

must be clearly addressed by the users. For a more efficient and effective handling of these issues, a

comprehensive strategy encompassing enhanced detection methods and a better understanding of the various

scholarly genres is thus crucial. This paper thus attempts to analyse both AI and human texts and suggest

solutions to deal with the possible problems with AI-generated materials.

METHODOLOGY

This study adopted a comparative genre analysis approach, utilizing the foundational models of Swales (1990)

and Bhatia (1993). Given the exploratory nature of this inquiry and the detailed, qualitative analysis required, a

pilot study was conducted with a focused dataset. The primary dataset consisted of ten research abstracts

written by postgraduate master's students specializing in English Language and Literature from a Malaysian

public university. These abstracts, covering diverse topics in language, literature, and education, provided a

rich, authentic corpus of human-authored academic writing.

For the AI-generated corpus, Chat Generative Pre-Trained Transformer 3 (ChatGPT), specifically the gpt-3.5-

turbo version (May 2024 snapshot), was utilised. To ensure a direct comparison, the AI was prompted to

generate abstracts based on the same thematic areas as the human-written abstracts. The specific prompt used

was: "Write a research abstract for a master's thesis in [Topic Area, e.g., 'applied linguistics']." This prompt

was chosen for its simplicity and to mimic a likely student input, allowing the AI's default structuring

tendencies to emerge without heavy steering.

The analytical framework for the first phase of the study was Melliti's (2016) Create a Research Letter (CARL)

Introduction Model. This model was selected for its structured approach to identifying moves and steps, which

align well with the succinct nature of research abstracts. A detailed codebook (see Appendix A) was developed

based on the CARL model, defining each move (e.g., Purpose of Study, Methodology, Results) with clear

linguistic criteria and examples.

Coding Procedure and Reliability

Two trained coders, both applied linguistics researchers, independently analysed all twenty abstracts (10

human, 10 AI). The coding process involved segmenting each abstract into sentences or T-units and

classifying them according to the moves defined in the codebook. Following the initial independent coding, an

inter-rater reliability analysis was performed using Cohen's Kappa, which yielded a score of κ = 0.85,

indicating a high level of agreement. All discrepancies were discussed and resolved through consensus,

ensuring the final frequency counts were consistent and reliable.

Data collection procedure

In this study, ten mini-memoir abstracts that were written between December 2024 and April 2025 as part of

an in-class activity were selected for analysis. The texts were written by postgraduate master students

majoring in English Language and Literature at a public university in Malysia covering a wide range of topics

on language, literature, and discourse. Four abstracts focused on language, three on literature and three on

discourse. An advanced tool. ChatGPT3, was used, for the creation of another set of mini-memoir abstracts for

similar thematic areas, while maintaining a thorough comparison between the two sets. The researcher typed

the themes of the mini-memoirs into the ChatGPT chat bar and requested it to produce an abstract for each.

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 526

www.rsisinternational.org

Analysis of data

Next, each abstract was carefully read, whether it was authored by a person or an AI. Due to the complexities

associated with genre analysis, this method was preferred over automated ones. Human analysts, on the other

hand, possess the cognitive ability to distinguish the subtle meanings underlying each statement (shown by the

letter S in Table 1 below). The breaking down of the components of the abstracts was done with great detail for

the purpose of determining the underlying generic elements, employing Melliti’s (2016) framework for

Creating a Research Letter (CARL) Introduction Model.

Table 1 Melliti’s Create a Research Letter Model (CARL Model) (2016)

Introduction

Phase (IP): 3 S

Background Information (BI)* 1 S

Previous Research (PR)** 1 S

Previous Research/Background Information (PR/BI)* 1 S

Contextualisation

Phase (CP): 5 S

Identification of Gap (IG)* 1 S

Purpose of Study (PS)* 1S

Rationale for Study (RS)** 1 S

Methodology (ME)** 1S

Previous Research/Identification of Gap (PR/IG)** 1 S

Findings Phase (FP): 4

Results (R)* 2 S

Conclusion (C)* 1 S

Results/Conclusion (R/C)** 1 S

*note: The letter ‘S’ refers to ‘Sentence'

The choice of Melliti’s (2016) Research Letter Introduction Model for this study was intentional and

strategically aligned with the attributes of the mini-memoir genre. The justification for this selection is based

on the intrinsic similarities between the research letter and the mini-memoir, as both function as succinct

representations of their corresponding extended versions within academia. The research letter is a shorter

version of a regular research article that includes all the important parts of a scholarly inquiry in a small space.

In the same way, the mini-memoir, which is a short version of the master thesis, captures the main points of

the research project in a shorter form while retaining its academic nuances. Both types of writing are short,

which is a good way to share academic ideas without losing the essence of scholarly research. The study used

a model customised for research letters, which is in line with established academic norms to ensure the

consistency and comparability of the methods to other scholarly frameworks.

RESULTS AND DISCUSSION

The table provided below identifies the recurrence of keys in the AI generated vs. human written abstracts.

Table 2 below illustrates that in the AI-generated set of abstracts, "Purpose of Study" (PS) is the predominant

text function, comprising almost fifty percent of the references (24.1%). This elevated percentage signifies a

robust focus on explicitly describing the research's objective. The presence of PS in these abstracts indicates

that the principal purpose is to ensure that readers promptly comprehend the research's aims. In contrast, only

11.8% of the human-generated abstracts in the collection contain PS. It implies that asserting the purpose is

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 527

www.rsisinternational.org

essential but it is not the only thing that matters. The fact that there is less focus on PS in the human-written

set could suggest that the writer is taking a more balanced approach to writing abstracts, giving more weight to

components like methodology, results, and previous research. The "Methodology" (ME) section remains

constant in both datasets. It appears more frequently in the human-written dataset (20%) than in the AI-

generated set (17.2 %). The consistency shows how important it is to clearly explain the methodological

approach in both AI-generated and human-written abstracts. The small increase in the human-written set may

mean that more attention is being paid to the research process and the methods used, which could indicate that

the approach is thorough or particularly methodical.

Table 2 Frequency count of key recurrence in Human-written vs. AI-developed abstracts

Text Function

Examples of AI generated

sentences

Count

Percentage

Human

Count

Human

Percentage

Background

Information

(BI)

"The historical trajectory of English

in Asia, stemming from diverse

colonial encounters and evolving

through complex processes of

indigenization, adaptation, and

multilingual contact, necessitates a

comprehensive diachronic analysis

to understand its contemporary

forms and functions across the

region."

10%

11.8%

Research (PR)

“Studies indicate that women utilise

more reduced speech acts than

men.”

10%

15.7%

Identification of

Gap (IG)

“There is a dearth of research on

how young learners master

grammatical competence in

bilingual settings.”

20.7%

11.8%

Purpose of

Study (PS)

“This study aims to investigate how

gender representation is expressed

in local novels.”

24.1%

11.8%

Methodology

(ME)

“Employing a qualitative approach,

this study utilised semi-structured

interviews to explore student’s

experiences through thematic

analysis.”

17.2%

20%

Results (R)

“Allegory is a useful tool to

expound on abstract themes in

literary works.”

17.6%

Conclusion (C)

“This study shows a positive

relationship between linguistic

competency and linguistic

performance.”

17.2%

11.8%

Results/

Conclusion

“The findings indicate that

discourse markers improve

3.7%

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 528

www.rsisinternational.org

(R/C)

narrative cohesion, thereby

affirming their significant role in

enhancing reader comprehension.”

Hypothesis (H)

“There is a relationship between

linguistic competence and linguistic

performance.”

5.6%

Earlier studies show that there is an immense distinction between the two sets. There are 15.7% of instances of

PR in the human-written collection, but only 10% in the AI-generated set. This substantial variation

demonstrates that the human-written set emphasises on placing the current study in the context of other

research. The contextualisation is essential for exhibiting how relevant and original the research is. Its higher

frequency in the human-written dataset could imply that the literature review elements are more fully

integrated. To assess the statistical significance of the observed frequency differences between human and AI-

generated abstracts, a Fisher's Exact Test was performed due to the small sample size. The test revealed a

statistically significant difference in the distribution of moves between the two groups (p < .05). Specifically,

the over-representation of "Purpose of Study" (PS) in AI abstracts and the under-representation of "Results" (R)

were key drivers of this significance. This statistical test strengthens the confidence in our findings, suggesting

that these divergent patterns are not due to random chance but reflect fundamental differences in compositional

focus.

There are about the same number of times that both sets mention "Background Information" (BI) and

"Identification of Gap" (IG). In the dataset made by AI, BI and IG both display 10% and 20.7%, but in the

dataset written by humans, BI and IG appeared 11.8% each. This means that both groups still need to provide

context and highlight the research gap. The small increase in BI in the human-written set may mean that the

subject matter was presented in more detail. The IG levels, on the other hand, suggest that the group was

focused on finding and fixing gaps in their current understanding. The higher proportion of “Identifying Gaps”

(IG) in AI-generated abstracts suggests that the system is designed to make research gaps stand out more

clearly. Such a clarity is advantageous to novice writers as articulating gaps in the literature is often one of the

most commonly challenging aspects of abstract writing. Generally, in this context, AI devices may be

considered a helpful assistant to users for identification of research gaps. Therefore, these machines have the

potential to be an invaluable resource for research activities. However, along the same breadth, the AI tools

pose a significant drawback to novice users. The users’ over-reliance on AI may impact their own writing

abilities. especially in areas where the beginner users have the least capability.

On the contrary, the abstracts written by human authors in this research contain a higher percentage of

coverage for Results (17.6%) and Conclusions (11.8%), meanwhile the AI-written abstracts lack a specific

section on Results and cover only 17.2% of Conclusions. This suggests that the human-written abstracts

dedicate a section to discussing the findings of the study with explicit emphasis on the outcomes of the

research. Effective statements of significance and contributions of a study necessitate an accuracy in the

composition of the research. In the dataset produced by human authors, a blended section comprising

Results/Conclusions (3.7%) and Hypotheses (5.6%) was found, but these characteristics did not exist in AI-

written data. This indicates that there is a wider array of rhetorical functions in the abstracts produced by

human authors. Additionally, the abstracts too are more nuanced and diverse. This practice is termed as “a

tendency to innovate” by Bhatia (1993). Thus, it is not uncommon in academic discourse communities that

when reporting research results scholars try out different ways of doing things, like stating hypotheses

explicitly or combining results and conclusions. The presence of these features indicates that human writers

employ various techniques for enhancing the communication aspects of their study. The analysis of patterns in

abstracts written by both humans and AI shows that they follow the rules of writing academic abstracts. The

genre analysis reveals that both groups generally follow the expected structure of research articles (background

information, purpose, gap, method, results), although they may emphasise different aspects. This is in line with

the conclusion that academic abstracts are a very conventional genre targeted at effective and precise

transmission of critical information to the academic community.

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 529

www.rsisinternational.org

A significant variation lies in the way PSs are treated. AI abstracts tend to focus more on the study’s purpose,

whereas human abstracts distribute emphasis across various aspects. The AI appears to place more emphasis

on the clarity of an article’s goals, while humans balance that aspect with the context, method, and results. AI-

written abstracts focus more on “what the study is about,” whereas human-written abstracts include both “what

the study is” and “what it found.” It can be observed that both human and AI-generated abstracts share a

similar structure but place emphasis on different sections of the rhetorical aspect. Genre analysis helps identify

these distinct features, enabling educators and researchers to understand that AI-generated texts, while rule-

based, lack the flexibility and nuance inherent in human writing. These differences not only demonstrate the

equality of academic rigor but also highlight the distinct approaches through which human and AI critics

comprehend the same genre. The findings have important implications for identifying AI-generated writing in

students’ work. Educators and detection devices will better differentiate between human and AI authors’

structural and rhetorical strategies to find plagiarised texts. According to this study, the signs that are

considered the most trustworthy are:

• Increased Frequency of Statements of Purpose: an uncommon prevalence of "Purpose of Study" statements

may indicate AI-generated content, as AI often emphasizes clear and stated aims.

 Incomplete Results and Conclusions: AI-generated articles fall short of complete results and conclusions

that elicit the purpose and methodology of the research rather than the research itself.

 Insufficient Contextualisation: AI-generated information can suffer from a lack of complete

contextualisation that is often found in human-generated abstracts, particularly with the incorporation of

existing research.

Discourse-Level Analysis of the Abstracts

This research contrasts students' abstracts and AI-generated abstracts on the basis of two significant

dimensions: the degree of complexity in language and how framed the discourse is.

Language Complexity

We measured the complexity of the language in terms of the quantity of detail and technicality in every

abstract (Godwin-Jones, 2025). This involved how specialised words, discipline terminology, and sentence

structure were used. An examination was also carried on identifying the functionality of technical terms in the

addition of depth to the content without making it challenging for the intended audience to understand. This

method showed how word and sentence-level choices could make an abstract more accurate and

understandable. A chi-square test of independence confirmed that the use of discipline-specific terminology

was significantly higher in human-written abstracts (χ²(1, N = 20) = 6.67, p < .01), reinforcing the qualitative

observation of greater lexical complexity.

Organising Discourse

Analysis of discourse organisation focused on the organisation and presentation of content within abstracts

(Botchwey & Owusu, 2024). The following characteristics were analysed: move structures, ideational ordering

of ideas, and theoretical presentation-empirical data-methodological description relationships. An observation

was also performed to determine whether or not the abstracts referred to practical application, significant

findings, or recommendations. Organisational systematicity between abstracts was realised through

application of the similar set of analytical criteria.

Table 3 AI-generated vs Human-written abstracts

Set

Aspect

AI generated

Human written

Language

complexity

Clearer and more general descriptions of

methods and objectives

Employs discipline-specific

terminologies

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 530

www.rsisinternational.org

Discourse

organisation

Emphasises objectives, methodologies,

and implications, with reduced focus on

theoretical frameworks

Methodology and outcomes are

presented in details with specific

references provided

Language

Complexity

The approach is straightforward and

emphasises tangible consequences

alongside empirical studies

Comprehensive definitions and their

implications; intricate sentence

structures employed

Discourse

organisation

Structured around empirical studies and

practical outcomes with reduced focus on

definitions

Organised into definitions,

methodologies, results, and

implications

Language

Complexity

Language is generally formal and

structured with consistency in grammar

and clarity in expressions

Language laced with informality and

fragments alongside irregularities in

grammar and overall structures

Discourse

organisation

Structured with distinct sections

addressing research methods, findings,

and implications.

Disorganised discourse structures with

sentence fragments; lacks clarity in

focus and organisation

Language

Complexity

Concise and direct, yet theoretical in

nature; emphasises practical implications

Intricate and theoretical with thorough

discussions of various factors

Discourse

organisation

Emphasis given to empirical findings and

implications; structured around particular

case study and context

Thorough investigation of hypotheses

and variables, encompassing a range

of research techniques and

ramifications

Language

Complexity

Clarity of expressions and focus on

empirical elements with less attention

given to technical particulars

Comprehensive and specialised

terminologies encompassing precise

definitions and theoretical elucidations

Discourse

organisation

Emphasis given to empirical strategies

and results which were structured around

case studies and implications

Thorough structuring includes an in-

depth examination of theoretical

frameworks and methodologies

Language

Complexity

Concise and pragmatic; emphasises

execution and practical application of the

subject matter

In-depth examination of the subject;

intricate syntactic constructions

Discourse

organisation

Structured around empirical findings and

recommendations with less focus on

theoretical underpinnings

Organised around theoretical

background, methodology, and

analysis

Language

Complexity

Concise and practical; emphasises a

particular case study and empirical

evidence

Emphasised theories with thorough

references and intricate explanations

Discourse

organisation

Organised around practical research and

findings related to specific case studies

Comprehensive analysis of the study

preceded by theoretical framework

Language

Complexity

Theoretical yet orientated towards

practical applications; succinct and lucid

reporting

Intricate and abstract; comprehensive

examination of the subject

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 531

www.rsisinternational.org

Discourse

organisation

Organised around empirical analysis of

particular case study and theoretical

ramifications

Comprehensive examination

incorporating theoretical and narrative

components.

Language

Complexity

Expressions are more direct with

emphasis on particular drawbacks and

recommendations

Thorough details of limitations and

hypotheses with diverse complexity

Discourse

organisation

Structured around particular findings and

recommendations with clarity in

organisation

Comprehensive investigation of

drawbacks with diverse organizational

structure

S10

Language

Complexity

Concentrated on the pragmatic

dimensions of emerging technology using

precise, empirical terminology.

In-depth examination of emerging

technology; incorporates particular

citations and intricate language

structures.

Discourse

organisation

Organised around case studies and real-

world examples, with little emphasis on

theoretical foundations.

A complete analysis that covers both

theory and practice

Comparisons between the two datasets

When the two datasets were compared, it became evident that they were very different in terms of language

and discourse organisation. The analysis indicated that human authors and the AI machines employed different

strategies for information and expressions of thoughts. Several notable differences were detected in linguistic

sophistication and discourse structures between AI-generated abstracts and human-written ones. The findings

offered insights into the varying practice of academic writing by both human authors and AI tools. A thorough

examination into the degree of complexity of technicality of the research abstracts involved the analysis of

how the technical, discipline-specific words and sentence structures were used (Godwin-Jones, 2025). The

evaluation of the use of technical terms by the authors was performed to establish the depth of the abstracts

without compromising its clarity. The technique demonstrated how the author’s choice of word and sentence

structure impacted the accuracy and clarity of the intended meaning.

The structure and presentation of the content in an abstract can be studied by examining its discourse

organisation. The examination was performed by analysing the move structure and flow as well as the balance

between theory framing, empirical argumentation, and methodological explication (Botchwey & Owusu, 2024).

It was also important to determine if the abstracts mentioned the implications, key findings, and

recommendations. The systematic analysis of content organisation in the abstracts was guaranteed by applying

the same criteria. The analysis focused on the strategies adopted by human writers and AI tools in the writing

of research abstracts and how linguistic choices and structural organisation affect clarity and coherence. It was

found that the differences between the two datasets in terms of linguistic complexity and discourse structure

were apparent.

Language Intricacy

Academic abstracts written by human writers tend to display a higher quality of language command. As

illustrated in Table 3, the examples taken from the human authors indicate that they use the exact wording

together with technical terms and phrases as well as discipline-specific jargons. The use of such structures

allows the authors to provide accurate explanations with greater degree of engagement with the ideas presented

in the abstract. This system, however, presents privileges for human authors with advanced knowledge but it

poses challenges for those with little understanding. These findings support Sardinha’s (2024) claims that AI

systems face challenges in understanding or interpreting complex content.

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 532

www.rsisinternational.org

AI-generated ones use simple words and sentences to generally describe the methods and goals. While that

makes things easier, what they produce is not on the same level as human work. AI also favours clarity and

conciseness, which make reading about complex topics easier but less informative (Zheng, 2024). One of the

major criticisms is that AI-generated texts are general and vague. Academic texts that are excessively simple

and do not apply any scientific terminology might have been generated by software.

Organisation of Discourse

Human abstracts exhibit a precise and highly organized discourse structure. They talk about things like

methodology, findings, implications and theoretical discussions. This well-organised study makes it possible to

look at the subject matter from different angles and present qualitative conclusions in a balanced manner. AI

abstract is more focused on the research goals, methods, and practical results than the human one. AI abstracts

are easier to understand because they predominantly discuss empirical study results and applications. However,

they may not exhibit the same level of analytical depth and theoretical rigor as human-generated academic

papers. The structure and organization of the AI-generated text are very basic, which affects the overall clarity

and coherence of the paper. This indicates that the discourse structure could be used as a provenance marker.

AI-generated content is quite basic and practical without immersion into theoretical and methodological issues.

Towards Automated Detection: Operationalizing Genre Features

The distinct genre features identified in this study have practical implications for developing AI-detection

systems. The move-based patterns can be translated into feature sets for machine learning classifiers. For

instance:

 Lexical Features: A lower frequency of discipline-specific jargon and technical terms could serve as an

indicator.

 Syntactic Features: Lower sentence complexity scores (e.g., measured via parse tree depth) may flag

AI-generated t

 ext.

 Move-Structure Features: A classifier could be trained to identify an imbalance in move frequencies,

such as a high PS-to-R ratio or a complete absence of a Results section, which was a hallmark of the

AI-generated abstracts in this study.

While current automated detectors often rely on perplexity and burstiness metrics, integrating explicit genre

and rhetorical features could enhance their accuracy, especially against models fine-tuned to evade statistical

detection. Our findings suggest that a hybrid system, combining statistical analysis with rule-based checks for

rhetorical structure completeness, could be a promising avenue for future development.

CONCLUSION

The study employed genre analysis to identify distinctive features of research abstracts generated by humans

and AI. While the pilot-scale dataset offers valuable insights, we acknowledge that the small, single-site

sample limits the generalisability of the findings. The exclusive use of ChatGPT-3.5 also means the results

may not be fully representative of other, more advanced LLMs. Future studies should employ larger, multi-

institutional datasets and include a wider array of LLMs (e.g., GPT-4, Gemini, Claude) to validate and extend

these findings.

This study underscores the necessity of developing specialised methods for detecting AI-generated writing.

The move-analysis approach, complemented by statistical validation, provides a transparent and pedagogically

useful framework. For educators, this translates into the need for targeted training to recognize not just stylistic

quirks, but structural and rhetorical shortcomings in student submissions. For the research community, it

highlights the potential of integrating genre-based features into the next generation of AI-detection tools.

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 533

www.rsisinternational.org

Ethical considerations

 The study was conducted with the informed consent from the participants. The anonymity of the

students who participated in this study was guaranteed as their identities and personal details were not

disclosed in this paper. In addition, the confidentiality of their writing outcomes was also guaranteed to

only be shared in this paper. The students were also allowed to withdraw from the study should they

feel like doing so prior to the start of or during the research.

 This study was conducted after obtaining the research ethics approval from Universiti Pendidikan

Sultan Idris [UPSIPPPI/PYK/ETIKA(M)/JId.19(97)].

 We declare that there are no potential conflicts of interest with respect to this research, authorship, or

publication of this article.

Data availability

The datasets generated and analysed during this pilot study are not publicly available to protect the

confidentiality and academic integrity of the student participants, as the human-written materials are original

coursework that could reveal identities and writing styles. The AI-generated data are also withheld to prevent

their potential misuse as templates for academic work. However, to ensure the transparency and reproducibility

of our analysis, anonymised exemplars of one human-written and one AI-generated abstract, along with the

full analytical codebook (Appendix A), are provided in the supplementary materials. The complete dataset is

securely stored in the institutional repository and may be made available to qualified researchers upon

reasonable request, subject to ethical approval from Universiti Pendidikan Sultan Idris.

REFERENCES

1. Aburass, S., & Abu Rumman, M. (2024). Authenticity in authorship: the Writer’s Integrity framework

for verifying human-generated text. Ethics and Information Technology, 26(3), 1-18.

https://arxiv.org/pdf/2404.10781

2. Alqudah, A. R. (2024). A Comparative Genre Analysis Study of Scientific Articles Abstracts and AI-

Generated Abstracts. [Doctoral dissertation University of Saskatchewan, Saskatoon, Canada].

3. Amirjalili, F., Neysani, M., and Nikbakht, A. (2024) Exploring the boundaries of authorship: a

comparative analysis of AI-generated text and human academic writing in English literature. Front.

Educ., 9(1347421), 1-11. doi: 10.3389/feduc.2024.1347421

4. Arshed, M. A., Gherghina, Ș. C., Dewi, C., Iqbal, A., & Mumtaz, S. (2024). Unveiling AI-generated

financial text: A computational approach using natural language processing and generative artificial

intelligence. Computation, 12(5), 1-24. https://doi.org/10.3390/computation12050101

5. Barbosa, B. K. D. S. (2024). LLMs as tools for evaluating textual coherence: a comparative analysis.

[Master dissertation, Universidade Federal de Campina Grande, Paraiba Brazil].

https://dspace.sti.ufcg.edu.br/jspui/handle/riufcg/40791

6. Bhatia, V. K. (1993). Analyzing genre: Language use in professional settings. Routledge.

7. Botchwey, E., & Owusu, E. (2024). A Genre Analysis of Minutes of Academic Meetings: A Case Study

of a Technical University in Ghana. Linguistics Initiative, 4(1), 62-75.

https://doi.org/10.53696/27753719.41126

8. Eğin, F., A. Onan, and H. Yildiz Durak. 2025. “Let’s Talk About EdTech! A Topic Modelling Analysis

of AI Tools and Pre-Service Teachers’ Perspectives.” European Journal of Education, 60(1), e12913.

https://doi.org/10.1111/ejed.12913

9. Fedoriv, Y., Pirozhenko, І., & Shuhai, A. (2024). Linguistic Analysis of Human-and AI-Created Content

in Academic Discourse. Journal of Vasyl Stefanyk Precarpathian National University. Philology.10, 47-

67. doi: 10.15330/jpnuphil.10.47-67

10. Frangieh, C., & Abdallah, M. (2024). A Linguistic Comparison between Human-and AI-Generated

Texts in Lebanon. CALR Linguistics Journal,15, 1-11. https://web.aou.edu.lb/research/online-

journals/PublishingImages/Pages/Calr-15/Article%2011.pdf

11. Ghiurău, D., & Popescu, D. E. (2024). Distinguishing reality from AI: approaches for detecting

synthetic content. Computers, 14(1), 1-33. https://doi.org/10.3390/computers14010001

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 534

www.rsisinternational.org

12. Godwin-Jones, R. (2025). Technology integration for less commonly taught languages: AI and

pedagogical translanguaging. Language Learning & Technology, 29(2), 11–34.

https://hdl.handle.net/10125/73609

13. Gupta, B. P. (2024). Can Artificial Intelligence Only be a Helper Writer for Science? Science Insights,

44(1), 1221-1227.

https://doi.org/10.15354/si.24.re872

14. Hakam, H.T., Prill, R., Korte, L., Lovreković, B., Ostojić, M., Ramadanov, N., & Muehlensiepen, F.

(2024). Human-written vs AI-generated texts in orthopedic academic literature: comparative qualitative

analysis. JMIR Formative Res. 2024;8: e52164. https://doi.org/10.2196/52164.

15. Hyland, K. (2000). Disciplinary discourses: Social interactions in academic writing. London, UK:

Longman.

16. Maurya, R. K., & Maurya, S. R. (2024) Content Reliability in the Age of AI: A Comparative Study of

Human vs. GPT-Generated Scholarly Articles. Library Progress International, 44(3), 1932-1943.

https://philarchive.org/archive/MAUCRI-3

17. Mazzi, F. (2024). Authorship in artificial intelligence‐generated works: Exploring originalit y in text

prompts and artificial intelligence outputs through philosophical foundations of copyright and collage

protection. The Journal of World Intellectual Property, 27(3), 410-427. DOI: 10.1111/jwip.12310

18. Melliti, M. (2016). Generic structure of research letters’ introductions: Create a research letter

introduction model (CARL introduction model). TAYR Quarterly, 3(1), 10-44.

https://doi.org/10.31561/2014tq

19. Mobilio, S. B. (2024). Utilizing Generative AI To Counter Deceptive Messaging. Calhoun: The NPS

Institutional Archive DSpace Repository.

https://hdl.handle.net/10945/73186

20. Muñoz-Ortiz A, Gómez-Rodríguez C, Vilares D. Contrasting linguistic patterns in human and LLM-

generated news text. Artif Intell Rev. 2024. https://doi.org/10.1007/s10462-024-10903-2.

21. Nanola, E. L., Arroyo, R. L., Hermosura, N. J. T., Ragil, M., Sabanal, J. N. U., & Mendoza, H. B.

(2025). Recognizing the artificial: A comparative voice analysis of AI-Generated and L2 undergraduate

student-authored academic essays. System, 130, 103611. https://doi.org/10.1016/j.system.2025.103611

22. Rakrak, M. (2025). The Genre Approach to Writing: A Socially Contextualized Pedagogy for Effective

Instruction. English Language Teaching Journal, 5(1), 1-9.

https://ejournal.alqolam.ac.id/index.php/eltj/article/view/1681

23. Ramazani, A., Bijani, H., & Oroji, M. R. (2025). Comparative analysis of AI vs. human feedback

effects on IELTS candidates' writing performance. Journal of Foreign Language Teaching and

Translation Studies, 10(1), 17–40. doi: 10.22034/efl.2025.493559.1334

24. Sardinha, T. B. (2024). AI-generated vs human-authored texts: A multidimensional comparison.

Applied Corpus Linguistics, 4(1), 100083. https://doi.org/10.1016/j.acorp.2023.100083

25. Swales, J. M. (1990). Genre analysis: English in academic and research settings. Cambridge University

Press.

26. Tan, X. (2024). How Chinese Doctors Do Things with Discursive Strategies in Palliative Care Family

Meetings: A Genre Theory Analysis. Health Communication, 1-21.

https://doi.org/10.1080/10410236.2024.2431179

27. Thane, A. A. (2024). A Critical Genre Analysis of Social Media News Reporting: The Case of

Instagram. [Doctoral dissertation, Open Access Te Herenga Waka-Victoria University of Wellington,

New Zealand].

28. Verhulsdonck, G., Weible, J., Stambler, D. M., Howard, T., & Tham, J. (2024). Incorporating human

judgment in AI-assisted content development: The HEAT heuristic. Technical Communication, 71(3),

60-72. doi.org/10.55177/tc286621

29. Zhao, C. (2025). Rethinking Authorship in the Age of AI: Reflections on the AI-Integrated

Writing Framework (AWAI). Journal of Educational Technology and Innovation, 7(2), 25–38.

https://doi.org/10.61414/h2a5bt21

30. Zhang, G. (2023). Authorial stance in citations: Variation by writer expertise and research article part-

genres. English for Specific Purposes, 70, 131-147. https://doi. org/10.1016/j.esp.2022.12.002

31. Zheng, W. (2024). AI vs. Human: A Comparative Study of Cohesion and Coherence in Academic Texts

between Human-Written and ChatGPT-Generated Texts [Master’s thesis, Universidad de Alicante,

Spain].

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue X October 2025

Page 535

www.rsisinternational.org

Appendix A: Analytical Codebook for Move Analysis

Example Entry:

 Move: Purpose of Study (PS)

 Definition: A sentence or clause that explicitly states the aim, objective, or research question of the

study.

 Linguistic Cues: "This study aims to...", "The purpose of this research is...", "We investigate..."

 Example (AI): "This study aims to investigate how gender representation is expressed in local novels."

 Example (Human): "This paper seeks to explore the interplay between post-colonial theory and

modern Malaysian poetry."

Appendix B: Anonymised Exemplars with Move Annotation

Exemplar 1: Human-Written Abstract (Anonymised)

Title: Navigating Identity: Code-Switching in Multilingual Professional Environments in Malaysia

[BI] The linguistic landscape of Malaysia is inherently multilingual, presenting a complex environment for

professional communication. [PR] Previous sociolinguistic research has established code-switching as a

common phenomenon, yet its strategic functions in specific professional domains remain

underexplored. [IG] There is a scarcity of research focusing on the intentionality and perceived efficacy of

code-switching among young professionals. [PS] This study aims to investigate the strategic functions and

perceived effectiveness of Malay-English code-switching in business meetings. [ME] Employing a qualitative

case study design, data were collected through semi-structured interviews with eight marketing executives and

through non-participant observation of five team meetings, which were subsequently analysed using thematic

analysis. [R] The findings reveal that participants strategically employ code-switching to build rapport, assert

authority, and clarify complex concepts. [R/C] This suggests that code-switching is not merely a linguistic

reflex but a nuanced communicative strategy that enhances professional interaction and operational

efficiency. [C] The study concludes that recognizing the strategic value of code-switching can inform more

effective, culturally-attuned communication training programmes in multinational corporations.

Exemplar 2: AI-Generated Abstract (Anonymised)

Title: An Analysis of Gender Representation in Contemporary Young Adult Fiction

[PS] This study aims to investigate the portrayal of gender roles and stereotypes in contemporary young adult

fiction. [IG] While gender representation has been a topic of scholarly interest, a focused analysis on recent

publications from the past five years is lacking. [BI] Young adult literature serves as a significant socializing

agent for adolescents, shaping their perceptions of societal norms. [ME] Utilizing a qualitative content analysis

approach, a purposive sample of ten bestselling young adult novels published between 2020 and 2024 was

systematically examined. [C] The analysis indicates that while there is a move towards more progressive

representations, traditional gender stereotypes persist in character archetypes and narrative outcomes. [C] This

study underscores the need for continued critical engagement with literary content to promote gender equity

among young readers.