INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
Lecturer's Perceptions and Strategies on ChatGPT Overreliance in  
ESL Academic Writing Among Undergraduates: A Case Study at a  
Malaysian Private University  
Fairuz Umira Binti Azmi1, Harwati Hashim2  
1Universiti Kebangsaan Malaysia (UKM)  
2Universiti Poly-Tech Malaysia (UPTM)  
DOI: https://doi.org/10.47772/IJRISS.2025.91200050  
Received: 10 December 2025; Accepted: 17 December 2025; Published: 31 December 2025  
ABSTRACT  
The emergence of artificial intelligence (AI) tools such as ChatGPT has transformed academic writing practices  
in English as a Second Language (ESL) contexts while simultaneously raising concerns about academic integrity  
and skill development. This study explored ESL lecturers’ perceptions of students’ overreliance on ChatGPT  
and the strategies adopted to manage this phenomenon at a Malaysian private university. Guided by the Theory  
of Planned Behaviour (TPB), a qualitative case study design was employed, and semi-structured interviews were  
conducted with three ESL lecturers teaching academic writing. Reflexive thematic analysis revealed that while  
ChatGPT offers linguistic scaffolding, lecturers perceived a decline in authentic writing processes, diminished  
metacognitive engagement, and increasing occurrences of AI-generated inaccuracies and fabricated references.  
Moreover, varied lecturer expectations and the lack of guidelines were found to encourage students’ dependence  
on AI applications. Consequently, lecturers introduced in-class writing tasks, structured assessments and oral  
defences to verify the authenticity of student submissions. These results are significant because they emphasise  
the institutional requirements for AI literacy education, unified governance and the restructuring of assessments  
to guarantee ethical and accountable AI application. As a result, this study contributes context-specific insights  
into sustainable AI integration aligned with SDG 4’s call for quality education in the digital era.  
Keywords: ESL academic writing, ChatGPT, AI overreliance, Theory of Planned Behaviour (TPB), higher  
education  
INTRODUCTION  
The integration of artificial intelligence (AI) tools like ChatGPT in English as a Second Language (ESL)  
education presents a dual-edged reality in achieving Sustainable Development Goal 4 (SDG 4) on quality  
education. AI writing assistants offer learners instant language support along with feedback, improving fairness  
and significantly boosting learning achievements for non-native English learners in tertiary education (Barrot,  
2023). Additionally, a recent student generative AI survey conducted by HEPI (2025) indicates that over two-  
thirds of undergraduates are now relying on generative AI weekly for academic tasks, which underscores  
growing pedagogical impact.  
Nonetheless, this educational instrument raises concerns regarding academic honesty. The reason is that learners  
may grow dependent on an AI-driven writing tool for cultivating essential critical writing abilities, which are  
regarded as fundamental to an academic perspective (Yan, 2023). In addition, overreliance on AI can undermine  
the development of authentic academic writing skills, as students increasingly submit AI-generated work that  
appears polished but lacks depth and originality (Cotton et al., 2023; Kasneci et al., 2023). This concept aligns  
with empirical studies showing that generative AI systems, like ChatGPT, Claude and Gemini, can generate text  
that is linguistically smooth yet superficial and inaccurate (Li et al., 2024; Farquhar et al., 2024).  
Additionally, unclear institutional guidelines and a lack of proficiency among teachers increasingly shift the  
responsibility of handling AI misuse onto educators (Setyaningsih et al., 2025). While numerous investigations  
Page 571  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
examine students’ overreliance on AI tools, only a limited number have focused on ESL instructors’ views and  
reactions to ChatGPT’s influence on writing. Since teachers are responsible for guiding students’ behaviour,  
shaping learning strategies, designing assessments, and providing ethical support, this research gap must be  
covered. Research on educator attitudes and readiness shows that teachers’ AI literacy and institutional guidance  
strongly influence whether AI is integrated responsibly or becomes a source of uncertainty (Setyaningsih et al.,  
2025; Ali et al., 2024).  
Despite growing AI adoption in Malaysian universities, many ESL programmes reported inconsistent classroom  
practices and a lack of institutional guidelines, which may indirectly normalise AI misuse. As a result, lecturers  
increasingly struggle to maintain academic integrity while supporting students’ learning needs. In this study,  
overreliance is defined as excessive reliance on the technological support of generative AI tools such as ChatGPT  
for completing independent thinking and cognitive processes involved in academic writing tasks, such as  
developing ideas, building arguments, and constructing unique compositions. Such reliance may appear through  
verbatim copying of AI-generated texts, failure to go through drafting processes, and incomplete reasoning  
processes or displays of minimal or negligible critical engagement, ultimately resulting in foundational erosion  
of writing and analysis skills (Teng, 2025; Kasneci et al., 2023).  
Meanwhile, academic writing is a set of core skills that include coherence, argumentation, source integration,  
linguistic appropriateness, and adherence to academic integrity requirements, which are necessary for students’  
roles as academic authors (Jiang & Hyland, 2025). Educators’ evaluations of students’ work, particularly in  
generative AI contexts, therefore serve as key indicators of these competencies’ development or decline. Existing  
articles highlight the pedagogical challenges educators face when reconciling AI’s technological affordances  
with the need to maintain authentic learning (, 2024; Bin-Nashwan et al., 2023). Nonetheless, empirical insights  
into Malaysian ESL lecturers’ perceptions remain limited.  
Addressing the mentioned gap is essential for developing informed policies, curricula, and professional  
development frameworks in ESL education. The objectives of this study are:  
1. To explore ESL lecturers’ perceptions towards overreliance of ChatGPT in ESL undergraduate students'  
academic writing.  
2. To explore the strategies for addressing the issues that arise from the overreliance of ChatGPT tools in ESL  
students' academic writing.  
Through these objectives, the research contributes vital insights into responsible AI integration in higher  
education, with particular relevance to Malaysia's evolving educational landscape.  
LITERATURE REVIEW  
Academic Writing and Artificial Intelligence in Education  
Writing academic discourse in ESL contexts is a multi-faceted higher-order cognitive process that involves  
students having to engage in two or more simultaneous prose processes. As outlined in Flower and Hayes (1981)  
framework, the writing process is seen as recursive, encompassing planning, translating and reviewing, all  
demanding significant cognitive effort (Barrot, 2024). With generative AI tools becoming widely available,  
scholars have begun examining how these recursive stages might be altered. Jiang & Hyland (2025) believe that  
overreliance on AI tools may hinder the development of academic writing skills. Wang & Fan (2025) supports  
the claim and also concludes that although ChatGPT reduces cognitive load for linguistic accuracy, it  
simultaneously decreases leaners’ engagement in idea generation and higher-order processing. This can weaken  
the cognitive complexity required in academic writing.  
The cognitive demands of writing are also evident in feedback and revision processes. According to Flower and  
Hayes (1981), metacognitive review is crucial for development because it encourages writers to recognize flaws,  
assess coherence and enhance clarity. Research highlights its shortcomings, although AI can assist in correcting  
surface-level mistakes (Barrot, 2023; Hwang et al., 2023). Steiss et al. (2024), conversely, caution that depending  
Page 572  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
heavily on automated feedback could interfere with the recursive stage of writing known as "reviewing." If AI  
takes control of the revision process, learners might fail to absorb techniques or identify their flaws, resulting in  
what recent cognitive studies refer to as "cognitive offloading," a form of passivity that hinders the growth of  
autonomous problem-solving skills (Gerlich, 2025).  
Baek et al., (2024) and Ali et al. (2024), also claim that AI tools can help novice writers by reducing cognitive  
load for grammar, structure and organisation, which enables them to focus on higher-level thinking drawn from  
long-term memory. However, this perceived benefit highlights a recurring tension. AI tools often bypass  
generative phases such as drafting, evaluating meaning and constructing arguments, which limit opportunities  
for authentic knowledge building. Subsequently, generative AI can encourage premature acceptance of machine  
outputs and reduce learners’ monitoring and evaluative behaviours (Espartinez, 2024). This research has  
observed a lack of drafting and revising among students, in addition to the reduction in self-regulated  
engagement with meaning-making.  
The “monitoring phase” component of Flower and Hayes’ (1981) model may also be compromised if AI is  
introduced early in the writing process. Yan (2023) suggests that students employing AI tools exhibit reduced  
awareness because the technology interferes with the organic writing flow by providing early answers. Research  
on LLM quality and hallucinations also adds complexity to depending on AI. Surveys and technical studies  
document hallucinations and factual errors in LLM outputs, resulting in fluently expressed paragraphs with  
inaccurate statements or misleading references (Farquhar et al., 2024; Hwang et al., 2023). The evidence suggests  
that although AI-generated text may appear rhetorically polished, it can be conceptually shallow and require  
careful pedagogical mediation.  
Artificial Intelligence in Education and the ESL Classroom  
Rapid advancement in AI has significantly reshaped English as a Second Language instruction, particularly in  
academic writing development. Modern adaptive learning tools can now provide personalized writing support  
that adapts to individual student needs with real-time feedback on grammar, vocabulary, and sentence structure  
(Barrot, 2024; Hwang et al., 2023; Espartinez, 2024). For many ESL learners, AI writing assistants serve as  
always-available language tutors, providing instant error corrections and suggestions which boost the learning  
process (Espartinez, 2024; Barrot, 2023).  
However, emerging research reveals concerning trends about overreliance on these technologies. AI tools can  
excel at improving surface-level writing features, but they may unintentionally discourage deeper intellectual  
engagement (Gerlich, 2025; Yan, 2023). Multiple studies document cases where students' dependence on AI-  
generated content resulted in writing that is grammatically correct but lacks original thought and critical analysis  
(Wang & Fan, 2025; Espartinez, 2024). The limitations become particularly evident in advanced academic  
writing, where AI often fails to replicate subject-area writing styles or generate nuanced arguments (Jiang &  
Hyland, 2025; Baek et al., 2024). Recent research also shows that overuse of tools such as ChatGPT may reduce  
metacognitive involvement and awareness on the part of L2 writers (Freeman, 2025; Espartinez, 2024).  
In response, educators now face the challenge of integrating these tools effectively while preserving essential  
writing skills. Current best practices emphasize using AI as an assisting tool rather than a replacement for  
traditional writing instruction (Espartinez, 2024; Cotton et al., 2023). Many experts support hybrid approaches  
that combine AI's efficiency with human-guided instruction in critical thinking and genre conventions (Wang &  
Fan, 2025; Jiang & Hyland, 2025). However, implementation remains uneven across institutional contexts.  
Many universities lack clear policy frameworks, leaving educators to adopt ad hoc approaches to AI regulation  
(Cotton et al., 2023). This challenge is particularly severe in Asian and Malaysian tertiary settings where human-  
centred pedagogies and concerns about maintaining students’ authentic voices shape more cautious adoption  
practices (Hu et al., 2025). Compounding these issues, teachers are increasingly expected to evaluate AI-  
mediated writing without the necessary AI literacy training (Espartinez, 2024). Additional evidence suggests  
that students using AI systems may show reduced self-editing tendencies compared to learners using automated  
writing evaluation systems (Steiss et al., 2024). These results highlight the necessity of pedagogical approaches  
that teach students when and how to use AI tools responsibly.  
Page 573  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
Although there has been cumulative scholarly recognition of both the strengths and limitations of AI writing,  
only a handful of studies advance practical, ethical or pedagogically grounded models for their use within the  
English language classroom (Ali et al., 2024; Kasneci et al., 2023). They also point out that there are wide-  
ranging professional development shortfalls and many lecturers feel unprepared to cope with AI-assisted writing,  
as they receive a low level of support from their institutions (Freeman, 2025; Yan, 2023). Successful integration  
of AI will involve designs that mediate what is relevant, which ultimately provide the answer to how the  
technology assists in achieving these aims and learners continue to engage in acts such as analysis and  
interpretation that cannot be automated (Farquhar et al., 2024). These studies collectively highlight the increasing  
need for context-specific investigations, particularly in Malaysian tertiary ESL settings, where institutional  
readiness and assessment practices vary widely (Hu et al., 2025; Espartinez, 2024).  
Previous Studies of Educators’ Perceptions in Using AI in ESL Classrooms  
Educators see both the advantages and the challenges of using AI in ESL classrooms. Recent research shows  
that not only teachers become in favour of AI as teaching assistance that can provide instant feedback on the  
students’ written texts and encourage different learning paths (Espartinez, 2024). Many educators mentioned  
using AI tools to handle grading responsibilities while ensuring uniform assessment criteria for large ESL classes  
(Freeman, 2025). However, these perceived benefits come with concerns about academic integrity, as evidenced  
by Malaysian university educators’ worries about ChatGPT-generated submissions in composition courses (Hu  
et al., 2025; Cotton et al., 2023).  
The adoption of AI technologies in ESL education environments encounters considerable institutional barriers.  
Many universities lack defined policy guidelines, resulting in educators creating improvised strategies for AI  
incorporation (Espartinez, 2024). According to Li et al., 2024), East Asian universities often emphasize human-  
centred pedagogy, which leads to more cautious AI integration compared to Western institutions. These cultural  
distinctions are especially noticeable in writing education, where preserving the genuineness of students’  
expression is an issue. Moreover, recent advancements indicate that teachers are progressively required to assess  
student writing influenced by AI without being provided professional education in AI literacy (Espartinez, 2024).  
Recent studies reveal significant gaps in comprehending AI’s impact on language instruction in higher education.  
Although prior studies report on both the affordances and limitations of AI writing technologies, few reports  
exist on how they could be used ethically within an academic ESL programme in a university setting  
(Espartinez, 2024; Cotton et al., 2023). Additionally, there remains insufficient investigation into effective  
professional development models that prepare writing instructors to leverage AI while maintaining academic  
standards (Yan, 2023). More recent research also reveals that teachers’ degree of assurance to rein in AI itself  
differs greatly by institutional direction (Freeman, 2025).  
There is growing recognition that successful AI integration in tertiary ESL education requires balanced, context-  
sensitive approaches. Rather than viewing AI as either a solution or a threat, developing pedagogical models that  
strategically incorporate these tools while preserving essential elements of writing instruction is crucial (Jiang  
& Hyland, 2025). This involves designing tasks that leverage AI’s capabilities in language support appropriately  
while necessitating cognitive skills and unique critical thinking, beyond automation (Graham & Milan, 2025;  
Barrot, 2024).  
The Effects of AI Overreliance on Writing Skills  
Recent empirical studies have shown that although these tools provide immediate gains in terms of language  
production, the uncritical use of these tools is likely to place at risk the development of long-term writing  
competencies in university settings. Research reveals that overreliance on AI tools can reduce the ability to think  
critically. Baek et al. (2024) found that undergraduate students who use AI writing assistance were shown to be  
much weaker at generating natural arguments and integrating source materials than those who made routine  
utilisation of nondigital means. This research is of special concern because at the tertiary-level writing  
assessments often have a critical analysis component to the prompt (Jiang & Hyland, 2025).  
Page 574  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
Graham & Milan (2025) found that students who used AI tools significantly engaged less in revising drafts and  
were less iterative writers. Textual analyses of AI-assisted submissions in the humanities and social sciences  
have established a pattern where these will include less field-specific information and demonstrate lower levels  
of personal engagement with source materials (Jiang & Hyland, 2025; Baek et al., 2024). Recent comparative  
work also demonstrates notable inconsistencies in LLM-generated academic content, including fabricated  
citations and factual errors in tools such as ChatGPT, Bard/Gemini and Claude (Farquhar et al., 2024; Huang et  
al., 2023).  
Additionally, preliminary findings indicate that students who have become over-reliant on AI tools exhibit less  
metacognitive awareness of their writing processes and are less able to self-identify areas where they might be  
weak (Freeman, 2025; Espartinez, 2024). This has potential long-lasting implications for writing skill  
development beyond just individual courses.  
Conceptual Framework  
This study employs the Theory of Planned Behavior (TPB) (Ajzen, 1991) to analyse lecturers’ strategies for  
addressing ChatGPT overreliance in ESL academic writing. The construct in the framework includes attitudes  
(perception of benefits and risks for writing development), subjective norms (institutional policies, peer  
expectations, and student demands) and perceived behavioural control (confidence controlling AI use,  
influenced by training resource availability). These influences in combination dictate teacher intention and  
thereby the resulting pedagogy associated with AI tools.  
The institutional context of Malaysian private universities moderates these relationships. Unclear AI policies  
could compound issues of control through perceived empowerment, whereas clear guidelines and faculty  
development can help educators to accommodate ChatGPT in a critical capacity (Ali et al., 2024; Cotton et al.,  
2023). The framework also reflects academic integrity theory to address the ethical questions that are necessary  
for any framework that purports to implement practice compatible with wider educational values and standards  
(Cotton et al., 2023; Bin-Nashwan et al., 2023)  
By integrating TPB with institutional and ethical concerns, this framework provides a comprehensive  
perspective on how lecturers can balance innovation with pedagogical integrity in ESL writing instruction. It  
illustrates how personal values, social pressures and situational constraints interact to influence responses to new  
AI interventions in higher education.  
Figure 1: Theory of Planned Behaviour (Adapted from Ajzen, 1991)  
Page 575  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
METHODOLOGY  
Research design  
This study employed a qualitative case study design to investigate ESL lecturers’ perceptions and strategies for  
addressing students’ overreliance on ChatGPT in academic writing. A qualitative approach was appropriate for  
capturing the depth, complexity, and contextual nuances of lecturers’ lived experiences, interpretations, and  
instructional practices within their institutional environment. As stated by Yin (2018), a case study design  
enables a detailed and contextualised examination of a bounded system or in this particular research, the practices  
and perspectives of ESL educators at Malaysian higher learning institutions.  
Guided by the Theory of Planned Behaviour (TPB) (Ajzen, 1991), the research design focused on examining  
lecturers’ attitudes, subjective norms, and perceived behavioural control in relation to AI-assisted writing. These  
constructs informed the development of interview questions and shaped the analytical lens applied during data  
interpretation.  
Semi-structured interviews served as the primary method of data collection. This format ensured flexibility while  
maintaining alignment with the TPB framework, allowing participants to elaborate on their teaching experiences,  
institutional challenges, and strategies for managing AI overreliance.  
Population, Participants and Sampling Techniques  
Purposive sampling was utilized to choose participants with knowledge regarding AI applications in academic  
writing classrooms. The sample included three ESL instructors teaching writing at a Malaysian private  
university. Eligible participants were required to (1) be actively involved in teaching English or academic writing  
at the tertiary level, and (2) have encountered or managed instances of student writing potentially influenced by  
AI-generated content. This sampling strategy ensured that participants were information-rich and able to provide  
meaningful reflections on AI-related writing behaviours.  
Data Collection Procedures  
Data were collected through individual, online semi-structured interviews conducted via Google Meet. Each  
interview lasted approximately 30–45 minutes and was audio-recorded with participants’ consent. The interview  
guide was organised around TPB constructs, covering attitudes toward ChatGPT, perceived social influences,  
and the ability to regulate AI use in writing tasks.  
Participants were provided with an information sheet outlining the study’s objectives, confidentiality measures,  
and rights to withdraw. Verbal and written consent were obtained prior to data collection. Interviews were  
transcribed verbatim shortly after completion.  
Data Analysis  
The interview data were analysed using reflexive thematic analysis based on Braun and Clarke’s (2006, 2021)  
six-phase framework. The researcher first familiarised herself with the transcripts through repeated readings  
before conducting manual line-by-line coding. A combined deductive–inductive approach was applied: the three  
constructs of the Theory of Planned Behaviour (attitudes, subjective norms, and perceived behavioural control)  
were used deductively as the thematic domains, while the specific codes under each domain were generated  
inductively from lecturers’ narratives. Thus, the TPB constructs formed the main themes, and the codes reflected  
the nuanced patterns that emerged from the data, which were later organised and presented in Table 1. Themes  
and codes were iteratively refined to ensure internal coherence and meaningful representation of participants’  
experiences. Manual coding was selected to maintain close engagement with the data and to support interpretive  
meaning-making appropriate for a small qualitative dataset. To enhance the trustworthiness of the instrument,  
the revised interview script was also validated by one interviewee, who confirmed that the questions were clear,  
relevant, and appropriate for the study context.  
Page 576  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
FINDINGS  
Lecturers’ Perceptions of Students’ Overreliance on ChatGPT  
Lecturers consistently expressed dual views regarding the growing dependence of students on ChatGPT in  
completing academic writing tasks. They acknowledged the AI tools as beneficial to assist weaker students,  
particularly in generating ideas or improving their grammar, but many students rely on them excessively by  
replacing genuine writing effort with AI-generated text.  
L1 explained that some students “paste the question straight into ChatGPT and submit whatever comes out,”  
which indicates a lack of engagement with essential writing processes such as planning and drafting. L2 also  
echoed similar concern where the gap between students’ in-class writing and take-home assignments is “too  
drastic to be genuine,”. This shows that AI tools heavily influence the final submissions, and many assignments  
appear overly polished. L2 also added that “the writing sounds very similar with generic points that students  
themselves cannot elaborate.”  
All lecturers observed a noticeable decline in students’ writing competence and critical thinking ability. L2  
reported that although assignments appeared well-polished, students’ in-class writing demonstrated weaker  
grammar and less developed ideas. Additionally, L1 stated that prolonged reliance on ChatGPT will result in a  
reduction in critical thinking and independent reasoning. L3 emphasised that students often could not justify the  
content produced in their assignments during consultations: “When I ask why they included a particular example,  
they cannot explain it at all.”  
Lecturers also raised concerns regarding the loss of students’ individual voices. Many assignments displayed a  
generic, uniform style characteristic of AI-generated text. L3 commented that “many assignments sound the  
same,” while L1 noted that sentence structures appeared “too perfect” and inconsistent with students’ typical  
writing patterns.  
Social influences and inconsistent expectations further contributed to students’ behaviour. According to L1,  
students frequently claimed that “everyone is using ChatGPT”, which reinforced the perception that dependency  
on the tool is acceptable. L2, on the other hand, also pointed out that students often compare lecturers’  
expectations across courses, noting that some lecturers allow extensive AI use while others discourage it. This  
inconsistency creates confusion regarding acceptable academic practice. L3 also stated that the absence of a clear  
institutional policy leaves both lecturers and students uncertain about what constitutes appropriate or excessive  
use of AI tools.  
Table 1 presents the study’s findings mapped onto the three TPB domains, which function as the primary  
analytical themes. Under each theme, inductively derived codes illustrate the specific forms of behaviour,  
perceptions, and challenges described by lecturers. These codes were developed from repeated patterns in the  
data and reflect both shared and individual experiences expressed by the participants. The supporting quotations  
provide authentic evidence for each code, demonstrating how lecturers interpret students’ writing practices and  
their reliance on ChatGPT.  
TPB Domain  
Codes  
Supporting Quotations  
Decline in authentic “Some students don’t even attempt to write a draft  
(1) Attitudes Toward the  
Behaviour  
writing  
anymore—they just paste the question into  
ChatGPT and submit whatever comes out.” (L1)  
“The jump in quality between their in-class writing  
and assignment writing is too drastic to be genuine.”  
(L2)  
Reduced  
thinking  
critical “When I ask them why they included a certain point,  
they cannot explain it at all.” (L3)  
Page 577  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
“Their essays look polished, but the reasoning is  
shallow. They can’t defend the ideas.” (L2)  
Loss of student voice “Many assignments sound the same—very generic  
and unnatural. You can immediately sense it’s from  
ChatGPT.” (L3)  
“You see perfect sentences that don’t match their  
usual style.” (L1)  
Presence  
of  
AI “Some examples are irrelevant or inaccurate;  
hallucinations  
sometimes the facts are just wrong.” (L2)  
“I’ve seen fabricated references—authors and titles  
that don’t exist.” (L3)  
Peer pressure  
“When we question them, they say ‘But everyone is  
using ChatGPT,’ so they think it’s acceptable.” (L1)  
(2) Subjective Norms  
“They genuinely believe it’s the new norm in  
university writing.” (L3)  
Lecturer  
expectations  
“Some lecturers allow unlimited AI use, so students  
think we’re too strict when we ask them to write on  
their own.” (L2)  
“Students are confused because every lecturer has  
different rules.” (L1)  
Lack of clear policy  
“There is no official guideline, so everyone  
interprets ChatGPT use differently.” (L3)  
“The lack of clear rules makes students feel  
anything goes.” (L2)  
Difficulty detecting “Essays look perfect, but the moment you ask them  
AI-generated writing to explain, they can’t say a single thing.” (L2)  
(3)Perceived  
Control  
Behavioural  
“It’s difficult to prove AI use because the detection  
tools are not reliable.” (L3)  
Limited ability to “We cannot control what they do at home. They can  
regulate out-of-class use ChatGPT anytime without us knowing.” (L1)  
writing  
“We can warn them, but we can’t monitor every  
step of their writing outside class.” (L2)  
Institutional  
expectations  
“We ourselves need more training. We’re still  
learning how to deal with AI in the classroom.” (L3)  
“I wish we had proper workshops or at least  
standard guidelines to guide us.” (L1)  
Table 1: Themes (TPB Domains), Codes, and Supporting Quotations  
Page 578  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
Strategies Employed to Address ChatGPT Overreliance  
To counter the challenges posed by students’ overreliance on ChatGPT, the lecturers implemented several  
pedagogical and assessment-based strategies.  
A key approach involved increasing in-class writing activities. L1 stated that weekly in-class tasks “help show  
their real ability,” as students are unable to depend on AI tools during class. L2 used short-term writing activities  
to observe students’ spontaneous writing performance. Additionally, L3 also explained that the direct  
comparison between in-class drafts and take-home assignments helps identify inconsistencies.  
Lecturers also emphasised scaffold assignments to ensure engagement with the writing processes. L2 required  
the students to submit brainstorming notes, an outline and multiple drafts, which explain “forces them to show  
their thinking process instead of depending entirely on ChatGPT”. L1 and L3 similarly used staged drafting to  
monitor progress and detect sudden changes in style or quality.  
Oral checks such as mini presentations, oral defences and brief interviews were another effective strategy. L3  
frequently used oral defences to verify whether students understood the content of their assignments, noting that  
“students who rely heavily on ChatGPT struggle the moment you ask them to justify what they wrote”.  
According to L2, incorporating short interviews during drafting is meant to ensure that students can explain the  
key arguments.  
In addition, lecturers redesigned assessments to discourage AI misuse. L1 adopted more personalised writing  
prompts, which require students to connect topics to their own authentic experiences, making AI-generated  
responses less suitable. L2 incorporated interview-based or observation-based tasks that required original data.  
L3 embedded reflective components where students described their decision-making throughout the writing  
process.  
Overall, lecturers actively employ a combination of instructional and assessment strategies to promote authentic  
writing and reduce dependence on ChatGPT.  
DISCUSSION  
Lecturers’ Perceptions of Students’ Overreliance on ChatGPT  
The findings revealed that lecturers hold negative and cautious perceptions toward students’ increasing reliance  
on ChatGPT for academic writing. These perceptions are related directly to the Attitude, Subjective Norms, and  
Perceived Behavioural Control components of the Theory of Planned Behaviour (TPB).  
Lecturers noticed that many students skip the cognitive stages of writing, like planning and drafting, and instead  
depend heavily on ChatGPT. This corresponds with the Attitudes Toward the Behaviour aspect of TPB, which  
showed that this behaviour (overdependence) leads to outcomes. These worries closely mirror studies suggesting  
generative AI tools can disrupt genuine writing involvement and diminish vital cognitive tasks such as idea  
generation, organisation and iterative revision (Graham & Milan, 2025; Steiss et al., 2024). Similarly, the  
concerns raised by lecturers, such as over-polished assignments and inconsistency of students’ in-class writing,  
mirror the findings that ChatGPT can mask fundamental proficiency and undermine writing development.  
A critical issue highlighted by the lecturers was the decline in students’ critical thinking and content  
understanding. Students were unable to justify arguments or explain their own work. This reflects earlier findings  
by Teng (2025) and Espartinez (2024), who found that dependence on AI diminishes students’ metacognitive  
involvement and restricts their capacity to assess the quality of ideas. Moreover, the occurrence of AI  
hallucinations, such as examples and made-up citations, supports the findings of Farquhar et al. (2024) and Li et  
al. (2024), who demonstrated how AI systems produce believable yet incorrect data. These errors increased  
instructors’ scepticism towards AI-supported tasks.  
Subjective norms also played a significant role in shaping students’ reliance on ChatGPT. Lecturers noted that  
students often justified their behaviour by asserting that everyone is doing it, which echoed TPB’s view that  
Page 579  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
perceived social expectations influence behavioural choices. Similar patterns were also reported by Fajt &  
Schiller, (2025) and Setyaningsih et al., (2025) who found that students tend to normalise questionable practices  
when they believe peers are engaging in the same behaviour. In this study, inconsistency in the lecturer’s policies,  
where some permit AI use and others forbid it, confused and reinforced students’ perception that reliance was  
acceptable.  
The findings also show that lecturers experience reduced perceived behavioural control in detecting AI-  
generated writing. Many expressed uncertainties about distinguishing genuine student writing from AI-assisted  
text. Recent research similarly reports that educators struggle to verify AI-generated submissions due to  
unreliable detection tools (Cotton, 2024; Bin-Nashwan et al., 2023). Additionally, the absence of clear  
institutional guidelines mirrors concerns that higher education institutions are not yet adequately equipped to  
address AI-integrity issues (Freeman, 2025; Espartinez, 2024). As TPB suggests, when individuals perceive low  
control over a behaviour, their ability to regulate or influence it becomes limited. This was reflected in lecturers’  
uncertainty and inconsistent approaches to managing AI use. Overall, lecturers’ perceptions in this study align  
closely with the literature, highlighting the disruptive impact of AI overreliance on cognitive, ethical, and  
pedagogical dimensions of writing.  
Strategies Employed to Address ChatGPT Overreliance  
Lecturers responded to students’ overreliance by implementing various instructional and assessment practices  
aimed at promoting authentic writing. These strategies relate to behavioural responses and intentions which are  
influenced by their beliefs, the expectations of others, and their sense of control, in line with TPB.  
The use of in-class writing was one of the most prominent strategies. Assigning students to write paragraphs,  
drafts or reflections in real-time enables the lecturers to observe the students’ actual writing abilities and identify  
inconsistencies with take-home submissions. This aligns with Shin et al (2024), who emphasize the importance  
of process-based work to reduce AI misuse. Such strategies reinforce authentic learning experiences and are  
widely encouraged as adequate safeguards against AI-generated academic dishonesty.  
Lecturers also implemented scaffolded writing processes, which require brainstorming notes, outlines, and  
multiple drafts. This approach encourages students to engage meaningfully with each stage of the writing process  
and reduces the likelihood of relying on AI-generated content. Such practices are consistent with the work of  
Teng (2025) and Steiss et al. (2024), who highlight that scaffolding improves reasoning, coherence and  
metacognitive awareness, which usually deteriorate when students overly rely on AI tools. Research further  
backs the focus on scaffolding, indicating that systematic drafting cycles assist students in assimilating  
techniques and enhance their ability to oversee their thoughts (Teng, 2025; Steiss et al., 2024); Graham, 2022).  
Moreover, scaffolded assignments support the iterative processes described in known writing frameworks, such  
as the phases recognized by Flower and Hayes (1981) and subsequently elaborated by Hayes (2012).  
Lecturers also highlighted evaluations like consultations, brief presentations and oral defences. These function  
as verification methods since students who depend greatly on ChatGPT frequently struggle to defend, justify, or  
clarify their thoughts. Studies indicate that oral assessments rank among the dependable techniques for  
identifying AI-produced content and confirming authorship responsibility (Freeman, 2025; Cotton, 2024).  
Another strategy was assessment redesign, where lecturers introduced personalised, reflective or experience-  
based tasks. These tasks require students to draw from their own prior knowledge, observations and making AI-  
generated responses less relevant. Studies emphasise that authentic, contextualised assessments significantly  
reduce reliance on generative AI (Shin et al., 2024; Kasneci et al., 2023). Findings from this study demonstrate  
that such tasks led to more original submissions and deeper engagement with course content.  
Across all strategies, lecturers highlighted the need for institutional support, including clearer guidelines, AI  
literacy training, and programme-level consistency. Although individual strategies were effective in the  
classroom, they also highlighted the need for clearer guidelines, AI literacy training and assessment policies to  
ensure consistent practice across the programme. Even though individual strategies were effective, lecturers  
stressed that sustainable change requires institutional alignment, which is consistent with TPB’s emphasis on  
Page 580  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
enhancing perceived behavioural control. Recent research echoes this need, showing that educator confidence  
and regulatory consistency improve when institutions provide clear governance (Espartinez, 2024; Cotton et al.,  
2023).  
Overall, these strategies align closely with global educational recommendations and reflect adaptive, context-  
sensitive responses to the challenges of AI-assisted writing. They demonstrate lecturers’ commitment to  
maintaining academic integrity, fostering authentic writing, and safeguarding learning outcomes despite  
institutional uncertainty.  
CONCLUSION  
This study explored ESL lecturers’ perceptions of students’ overreliance on ChatGPT and strategies employed  
to address this issue in academic writing classrooms at a Malaysian private university. Overreliance on ChatGPT  
was identified as detrimental to students’ writing skills, critical thinking and sense of responsibility for their  
assignments. Educators noted growing differences between AI-generated writing, shaped by peer influences and  
diverse course demands. The lack of institutional policies and ineffective detection tools also diminished their  
assurance in handling AI-related concerns. In response, lecturers implemented classroom-based and assessment-  
oriented strategies to encourage genuine engagement with the writing process. Although this study was limited  
to a single institutional context, the issues identified may mirror emerging trends in other Malaysian higher  
education settings experiencing similar challenges with generative AI integration. These findings highlight the  
need for clearer policies, stronger AI literacy, and more cohesive institutional support to ensure responsible and  
ethical use of generative AI in academic writing.  
REFERENCES  
1. Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision  
Processes, 50(2), 179–211. https://doi.org/10.1016/0749-5978(91)90020-T  
2. Ali, D., Fatemi, Y., Boskabadi, E., Nikfar, M., Ugwuoke, J., & Ali, H. (2024). ChatGPT in teaching and  
learning: A systematic review. Education Sciences, 14(6), 643. https://doi.org/10.3390/educsci14060643  
3. Baek, C., Tate, T., & Warschauer, M. (2024). “ChatGPT seems too good to be true”: College students’  
use and perceptions of generative AI. Computers and Education: Artificial Intelligence, 7, 100294.  
https://doi.org/10.1016/j.caeai.2024.100294  
4. Balázs Fajt, & Schiller, E. (2025). ChatGPT in academia: University students’ attitudes towards the use  
of ChatGPT and plagiarism. Journal of Academic Ethics. https://doi.org/10.1007/s10805-025-09603-5  
5. Barrot, J. S. (2023). ChatGPT as a language learning tool: An emerging technology report. Technology,  
Knowledge and Learning, 29. https://doi.org/10.1007/s10758-023-09711-4  
6. Barrot, J. S. (2024). Leveraging ChatGPT in the writing classrooms: Theoretical and practical insights.  
Language Teaching Research Quarterly, 43, 43. https://doi.org/10.32038/ltrq.2024.43.03  
7. Bin-Nashwan, S. A., Sadallah, M., & Bouteraa, M. (2023). Use of ChatGPT in academia: Academic  
integrity  
https://doi.org/10.1016/j.techsoc.2023.102370  
8. Braun, V., & Clarke, V. (2021). One size fits all? What counts as quality practice in (reflexive) thematic  
analysis? Qualitative Research in Psychology, 18(3), 328–352.  
https://doi.org/10.1080/14780887.2020.1769238  
hangs  
in  
the  
balance.  
Technology  
in  
Society,  
75,  
102370.  
9. Cotton, D. R. E., Cotton, P. A., & Shipway, J. R. (2023). Chatting and cheating: Ensuring academic  
integrity in the era of ChatGPT. Innovations in Education and Teaching International, 61(2), 228–239.  
https://doi.org/10.1080/14703297.2023.2190148  
10. Espartinez, A. S. (2024). Exploring student and teacher perceptions of ChatGPT use in higher education:  
A Q-methodology study.  
Computers and  
Education: Artificial Intelligence, 7, 100264.  
https://doi.org/10.1016/j.caeai.2024.100264  
11. Farquhar, S., Kossen, J., Kuhn, L., & Gal, Y. (2024). Detecting hallucinations in large language models  
using semantic entropy. Nature, 630(8017), 625–630. https://doi.org/10.1038/s41586-024-07421-0  
12. Flower, L., & Hayes, J. R. (1981). A cognitive process theory of writing. College Composition and  
Communication, 32(4), 365–387. https://doi.org/10.58680/ccc198115885  
Page 581  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)  
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XII December 2025  
13. Freeman,  
J.  
(2025).  
Student  
generative  
AI  
survey  
2025.  
https://www.hepi.ac.uk/wp-  
content/uploads/2025/02/HEPI-Kortext-Student-Generative-AI-Survey-2025.pdf  
14. Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking.  
Societies, 15(1), 6. https://doi.org/10.3390/soc15010006  
15. Graham, O., & Milan, Y. (2025). The impact of ChatGPT reliance on the development of student critical  
thinking skills. Preprints. https://doi.org/10.20944/preprints202506.0861.v1  
16. Hu, H., Du, K., & Hashim, H. (2025). ChatGPT in English writing assessment: Can AI accurately  
measure complexity, accuracy, and fluency indices? 2025 International Conference on Distance  
Education and Learning (ICDEL), 189–195. https://doi.org/10.1109/icdel65868.2025.11193569  
17. Huang, L., Yang, Y., Ma, W., Zhong, W., Feng, Z., Wang, H., Chen, Q., Peng, W., Feng, X., Qin, B., &  
Liu, T. (2023). A survey on hallucination in large language models: Principles, taxonomy, challenges,  
and open questions. ArXiv. https://doi.org/10.48550/arxiv.2311.05232  
18. Hwang, G.-J., & Chen, N.-S. (2023). Editorial position paper: Exploring the potential of generative  
artificial intelligence in education: Applications, challenges, and future research directions. Educational  
Technology & Society, 26(2). https://doi.org/10.30191/ETS.202304_26(2).0014  
19. Jiang, F. K., & Hyland, K. (2025). Does ChatGPT write like a student? Engagement markers in  
argumentative essays. Written Communication, 42(3). https://doi.org/10.1177/07410883251328311  
20. Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G.,  
Günnemann, S., Hüllermeier, E., Krusche, S., Kutyniok, G., Michaeli, T., Nerdel, C., Pfeffer, J., Poquet,  
O., Sailer, M., Schmidt, A., Seidel, T., & Stadler, M. (2023). ChatGPT for good? On opportunities and  
challenges of large language models for education. Learning and Individual Differences, 103, 102274.  
https://doi.org/10.1016/j.lindif.2023.102274  
21. Li, J., Chen, J., Ren, R., Cheng, X., Zhao, X., Nie, J.-Y., & Wen, J.-R. (2024). The dawn after the dark:  
An empirical study on factuality hallucination in large language models. Proceedings of the 62nd Annual  
Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 10879–10899.  
https://doi.org/10.18653/v1/2024.acl-long.586  
22. Pritpal Singh Bhullar, Joshi, M., & Chugh, R. (2024). ChatGPT in higher education: A synthesis of the  
literature and a future research agenda.  
https://doi.org/10.1007/s10639-024-12723-x  
Education and Information Technologies, 29.  
23. Setyaningsih, E., Zainnuri, H., Wahyuni, D. S., & Hariyanti, Y. (2025). EFL students’ use, perceptions,  
and reliance on Chat-GPT for editing and proofreading: A technology acceptance model perspective.  
Journal of Languages and Language Teaching, 13(3), 1367. https://doi.org/10.33394/jollt.v13i3.13484  
24. Steiss, J., Tate, T., Graham, S., Cruz, J., Hebert, M., Wang, J., Moon, Y., Tseng, W., Warschauer, M., &  
Olson, C. B. (2024). Comparing the quality of human and ChatGPT feedback of students’ writing.  
Learning and Instruction, 91, 101894. https://doi.org/10.1016/j.learninstruc.2024.101894  
25. Teng, M. F. (2025). Examining longitudinal development of writing motivation in the GenAI context: A  
self-determination  
theory  
perspective.  
Learning  
and  
Motivation,  
91,  
102157.  
https://doi.org/10.1016/j.lmot.2025.102157  
26. Wang, J., & Fan, W. (2025). The effect of ChatGPT on students’ learning performance, learning  
perception, and higher-order thinking: Insights from a meta-analysis. Humanities and Social Sciences  
Communications, 12, 621. https://doi.org/10.1057/s41599-025-04787-y  
27. Yan, D. (2023). Impact of ChatGPT on learners in a L2 writing practicum: An exploratory investigation.  
Education and Information Technologies, 28, 13943–13967. https://doi.org/10.1007/s10639-023-11742-  
4
Page 582  
www.rsisinternational.org