Abstract
Generative AI (GenAI) is reshaping academic writing and research workflows, especially for scholars publishing in a language other than their own. This qualitative case study examines how Saudi non-native English-speaking (NNES) researchers use GenAI and how they evaluate its quality and integrity when writing for publication in English. Drawing on 34 semi-structured interviews across disciplines and career stages, analysed via reflexive thematic analysis, we identify five roles that structure GenAI use: literature navigator, writing enhancer, intellectual partner, research designer–analyst, and publishing mentor. The participants reported clear benefits: speed, reduced effort, improved translation and coherence, organized summaries and lower costs, especially in literature work and language polishing. They also stated systematic quality and integrity checks, citing risks of unreliable summaries, fabricated or misattributed references, privacy and copyright concerns and erosion of voice and creativity. Human agency and responsible use are emphasized across practices. Although anchored in the Saudi context, the findings are analytically transferable to systems in which English-language publication is compulsory outside English-dominant environments. The results inform policy and practice: international and local bodies can refine guidance and competency frameworks; institutions and NNES researchers can strategically leverage GenAI while safeguarding integrity and authorial voice; editors of ranked journals can clarify GenAI policies for fairness; and targeted training, especially for postgraduate researchers, can strengthen critical, analytical, ethical and legal GenAI competencies.
Keywords
Introduction
Research is crucial in the academic world. Whether pursuing a higher degree, advancing one’s career, securing promotions or obtaining funding, academic activities largely depend on research output (Chauhan & Currie, 2024). Academic writing is particularly complex and can be overwhelming, especially for early-career researchers at the postgraduate level or those seeking publication in high-ranked journals (Chauhan & Currie, 2024; Jin et al., 2025).
Academic researchers face growing pressure to publish in international high-ranked journals, most of which publish in English, the dominant global academic language. This raises a crucial question for scholars whose first language is not English—they must also develop sufficient proficiency in another language (Flowerdew, 2019). Aside from unfamiliar vocabulary and grammar, cultural differences are also embedded in academic discourse (Jin et al., 2025).
In many Saudi universities, postgraduate students and faculty are required to publish in English, often in high-ranked journals, as a condition for graduation or promotion (King Saud University, Scientific Council, 2025; Prince Sultan University, 2025). For non-native English-speaking (NNES) researchers, this policy creates a persistent linguistic and cognitive burden: drafting, translating, and polishing scholarship in a non-native language while following international style and integrity standards (Flowerdew, 2019).
In competitive environments that demand continued productivity, researchers seek support. Generative artificial intelligence (GenAI) has recently emerged as a promising assistant (Berg, 2023). GenAI consists of advanced algorithms capable of generating text, images and other content, and it has begun to reshape the research landscape by supporting literature work, writing quality, data analysis and overall academic tasks (Jin et al., 2025), which may relieve these pressures for NNES researchers. However, alongside these benefits are concerns about quality, reliability and ethical implications in scholarly work (Chauhan & Currie, 2024).
Previous studies on GenAI in academia have largely examined what GenAI tools can do, such as technical capacity, feature evaluations or what users think in the abstract; attitudes and intentions, often through cross-sectional surveys with students, researchers or journals’ peer reviewers (Al-Abdullatif & Alsubaie, 2024; Andersen et al., 2025; Hadan et al., 2024); editorials or scoping reviews that synthesise risks, such as plagiarism, detection and bias (Cohen & Moher, 2025; Jiang et al., 2025; Lim et al., 2023); and affordances and challenges (Andersen et al., 2025; Chauhan & Currie, 2024). Many of these studies have isolated single tools, such as ChatGPT and Elicit (Dwivedi et al., 2023; Whitfield & Hofmann, 2023); single stages, such as paraphrasing, idea generation and enhancing English abstracts (Chou & Chow, 2024; Yao et al., 2025); and specific levels of researchers, such as postgraduate students (Aladsani, 2025; Jin et al., 2025) or university faculties (Peres et al., 2023).
What remains missing is a close examination of how NNES researchers actually use GenAI end-to-end across the research–writing workflow for English-language publication and, more importantly, how they evaluate the quality and integrity of using it in situ.
Addressing these gaps, this qualitative case study offers context-specific evidence from Saudi researchers from various levels, from master’s degree to full professor, foregrounding NNES researchers’ lived practices and the competencies they mobilise to keep GenAI useful, trustworthy and under human control, with attention to the tension that arises, under compulsory English-publication policies, between utilising GenAI for linguistic support and maintaining authorial integrity.
The purpose of this study is to understand the role of GenAI in helping Saudi NNES researchers meet the demands of English-language publication and to identify how it safeguards scholarly quality and integrity in the process. The study is guided by the following question:
How do Saudi NNES researchers use GenAI and evaluate its quality and integrity when writing for publication in English?
Literature Review
NNES Researchers and English-Language Publications
Given the global dominance of English, most highly ranked journals publish in English, which places pressure on academics worldwide who seek to publish in these venues, whether due to institutional requirements, the pursuit of research grants or the desire to enhance academic reputation (Flowerdew, 2019).
Hyland (2016) argues that native and non-native English speakers face similar difficulties in academic writing and that language itself is not a barrier to publication, citing the many non-native authors who publish successfully and repeatedly in English. However, this view is contested. A substantial body of studies has reported that English constitutes a major obstacle for non-native authors. Flowerdew (2019) discusses this from two perspectives. First, NNES researchers carry an extra burden: learning and using an additional language at a level that meets the expectations of highly ranked English-language journals, which demands significant time and effort during writing. Second, NNES authors may be less familiar with disciplinary rhetorical norms and cultural–linguistic conventions, leading to frequent consultation of dictionaries or language editors to ensure fluency, accuracy and coherence. Flowerdew even notes that some researchers may resort to copying and pasting published papers, exposing themselves to plagiarism risks. Likewise, Soler (2021) asserts that NNES scholars face inequity in English-medium publishing because of the language itself. One manifestation is gatekeeping on linguistic quality: journals may desk-reject manuscripts that do not meet their English standards, and some may implicitly favour work by native English authors (Politzer-Ahles et al., 2016).
Building on this debate, publishing in English for NNES scholars adds extra layers to ordinary research work: Aside from reading complex texts, following English-language rhetorical norms and writing with discipline-specific style and precision (Flowerdew, 2019; Jin et al., 2025), they need to simplify complex findings while ensuring accuracy, logical flow and adequate evidence (Gupta et al., 2022). University policies that require publication in high-ranked English journals intensify these demands and can widen existing inequalities in global knowledge production, in which English already dominates what is considered high-ranked scholarship (Flowerdew, 2019; Gupta et al., 2022; Hadan et al., 2024).
GenAI in Scholarly Writing: Capabilities and Constraints
GenAI refers to a subset of AI systems designed to generate content based on the data on which they have been trained (Jin et al., 2025). Unlike traditional AI, which focuses on pattern recognition, data classification or decision-making within a predefined framework, GenAI creates new outputs that have not been explicitly programmed or encountered before involving text, images or other content (Dwivedi et al., 2023). These advancements enable it to produce advanced human-like content (Lim et al., 2023).
In academic work, GenAI tools can assist researchers at various points in the research–writing workflow, lowering routine burdens. Previous studies have consistently shown two core capabilities. First, GenAI can reduce the mental load of language work by proofreading, editing and paraphrasing text, and it can also improve local coherence and readability and flag awkward or inconsistent sentences (Aladsani, 2025; Jin et al., 2025; Katongtung et al., 2025; Liu et al., 2024). Second, AI-powered literature tools can reduce the time required to scan large amounts of literature by surfacing relevant papers, grouping related work and generating concise summaries that help researchers identify potential research gaps (Jain et al., 2024; Katongtung et al., 2025; Srivastava & Agarwal, 2024; Whitfield & Hofmann, 2023). These strengths are especially attractive to researchers writing in English as an additional language who confront time-intensive drafting and revision when targeting high-ranked journals (Chou & Chow, 2024; Jin et al., 2025).
Within this context, GenAI is beneficial because it helps with the heaviest language tasks. NNES researchers report using GenAI tools, such as ChatGPT and Grammarly, to strengthen grammar, improve coherence and clarify sentences, especially when aiming for journals with strict style expectations (Hadan et al., 2024). Previous studies have also noted that NNES researchers use GenAI more frequently for writing support than native speakers and that this assistance can reduce writing anxiety and increase self-efficacy (Baek et al., 2024).
At the same time, these benefits come with important constraints. Large language models (LLMs) can produce fluent but unreliable text; they can misread disciplinary nuance, invent details or fabricate and misattribute references (Aladsani, 2025; Nicholas et al., 2024). AI-powered literature tools can also imperfectly summarise, overlook key studies outside their indexed sources or present uneven coverage across fields (Jain et al., 2024; Whitfield & Hofmann, 2023). Translation support can help with clarity, but it may flatten field-specific terminology or rhetorical moves if unchecked (Da Silva et al., 2024).
Academic Integrity with GenAI
Academic research is built on integrity and ethics, which are fundamental to preserving scholarly trust and advancing knowledge. Despite its promising potential, GenAI presents several challenges that must be carefully considered.
The key limitations of GenAI include its tendency to produce inaccurate or non-factual outputs and insufficient or fabricated references (Aladsani, 2025). GenAI also lacks an understanding of the meanings behind the words it processes, often resulting in a lack of logical flow despite mimicking a confident human tone (Hadan et al., 2024; Nicholas et al., 2024). It tends to produce overly general and repetitive text and lacks the emotional depth that characterises human academic work (Jin et al., 2025; Liu et al., 2024).
These limitations pose significant threats to academic integrity and authenticity, potentially leading to a decline in research quality (Nicholas et al., 2024). The capability of GenAI to produce seemingly convincing academic content, laden with misinformation and misleading data, raises concerns about the proliferation of fabricated, low-quality research that journals may struggle to detect and manage effectively, further burdening already overextended editors and reviewers (Chauhan & Currie, 2024).
Despite their advancements, AI detection tools’ accuracy remains a concern, and these tools often misclassify human-authored work as AI-generated. NNES researchers are disproportionately affected, facing plagiarism allegations due to the potential similarity between their writing styles and GenAI (Baek et al., 2024; Chauhan & Currie, 2024).
Over-reliance on GenAI may also erode researchers’ creativity, reasoning, critical thinking and problem-solving skills (Al Lily et al., 2023; Jin et al., 2025; Liu et al., 2024), hindering the development of essential intellectual capabilities crucial for scholarly work (Hadan et al., 2024). Additional risks associated with GenAI include breaches of sensitive information and privacy concerns (Aladsani, 2025). Widespread GenAI use could deepen existing inequalities by reinforcing biases related to language efficiency, which negatively affect NNES researchers’ chances of publication (Jin et al., 2025), AI access, and the availability of adequate training to effectively utilise them (Nicholas et al., 2024). Researchers from lower socio-economic backgrounds may face disadvantages due to the cost of premium GenAI features, limiting their access to these resources (Al-Abdullatif & Alsubaie, 2024). The biases embedded within GenAI systems, stemming from training data, algorithmic design and societal contexts, may further perpetuate discrimination (Dwivedi et al., 2023).
As a result, some educational institutions have imposed bans on GenAI use, while certain scholars have expressed strong opposition to its integration into research (Liu et al., 2024). There is an urgent need for the establishment of ethical frameworks and policies that address the potential and risks of AI technologies (Baek et al., 2024; Dwivedi et al., 2023).
International and National GenAI Policies and Frameworks for Research
Internationally, United Nations Educational, Scientific and Cultural Organisation (UNESCO) has established coherent guidance on GenAI in education and research, articulating a human-centred approach to GenAI that sets the tone for research and education. UNESCO’s recommendations on the ethics of AI provide a normative anchor for addressing controversies around generative AI, maintaining that AI must serve the development of human capabilities and enable effective human–machine collaboration across life, learning and work. This approach is grounded in human rights and the protection of human dignity and cultural diversity within the global knowledge commons, ensuring that AI augments rather than displaces human decision-making (Holmes & Miao, 2023).
Complementing these policy tools, UNESCO’s AI Competency Framework for Students (Miao & Shiohira, 2024) offers a structured lens with clear implications for research training. The framework comprises 12 competencies across 4 dimensions: (1) a human-centred mindset (human agency, human accountability and citizenship in the era of AI), (2) ethics of AI (embodied ethics, safe and responsible use and ethics by design), (3) AI techniques and applications (AI foundations, application skills and creating AI tools) and (4) AI system design (problem scoping, architecture design, iteration and feedback loops). Although designed for learners, these dimensions map closely onto researchers’ responsibilities, selecting appropriate tools, exercising judgement, documenting methods with transparency and iterating responsibly with human oversight.
At the national level, Saudi Arabia’s Research, Development and Innovation Authority (RDIA) sets explicit expectations for scientific integrity and research ethics that directly address GenAI. More importantly, RDIA clarifies acceptable versus unacceptable uses of AI in research. AI use is permissible, when acknowledged, for language proofreading, style editing, summarising research ideas and translation. Conversely, claiming authorship of AI-generated content constitutes a violation of scientific integrity, whether in a publication or a proposal. Institutions are mandated to use up-to-date detection software to verify claims, and RDIA specifies that it is a clear breach if AI-generated content exceeds 20% of the total work. These provisions place disclosure, proportionality and human authorship at the centre of compliant practice (RDIA, 2025).
Taken together, UNESCO’s human-centred, rights-based architecture and Saudi RDIA’s AI guidance converge the pillars for research with GenAI. For NNES researchers, these frameworks legitimise supportive uses, such as language polishing and structured summarisation, while drawing clear lines around authorship, data protection and integrity, ensuring that GenAI enhances quality without eroding the human core of scholarly work.
Methodology
Study Rationale and Design
This study adopts a qualitative case study (Merriam & Tisdell, 2016; Yin, 2018) to examine how NNES researchers use and evaluate GenAI when writing for publication in English. The case focuses on Saudi NNES researchers who are required to write for English-language publication and who report using GenAI in that workflow.
The case study method is appropriate, as the research questions explore ‘how’ and ‘why’, the study examines a contemporary event, and control over behavioural variables is not required (Yin, 2018). This case study is analytically purposeful for two reasons. First, it constitutes a ‘critical’ case (Yin, 2018): In Saudi Arabia, many universities formally require English-language, high-ranked publications for graduation and promotion, creating strong, system-level pressures on NNES researchers (King Saud University, Scientific Council, 2025; Prince Sultan University, 2025). If there is any place where GenAI will be used intensively to meet such pressures, it is in Saudi Arabia. Second, it is a ‘typical’ case of higher education systems in which English-language publication is compulsory or strongly required for graduation, promotion or funding, even though English is not the national language. This requirement is common across many non-English-speaking countries, especially in the Arab and Gulf regions. Therefore, studying this case elucidates the general mechanisms of how researchers use GenAI and how they evaluate its quality and integrity, which are likely to be transferred to similar policy settings.
The case is time-bounded, that is, the use of GenAI during the current or most recent manuscript cycle, and participant-bounded, that is, researchers at postgraduate, early-career, junior and senior levels across various disciplines. The findings are not claimed as statistically generalisable but are offered as analytically transferable through a thick description of context, participants and processes.
Participants and Sample Selection
We employed purposive, maximum-variation sampling to capture the breadth across disciplines (science vs. social sciences and humanities), academic rank (postgraduate, early-career, junior and senior) and gender to enrich the data (Bryman, 2016). The inclusion criteria were Saudi nationality, NNES status, active engagement in writing for English-language publication within the past 12 months and (self-reported use of GenAI at any stage of the research–writing workflow, such as literature work, drafting, editing, analysis support or publishing tasks). The participants were recruited through face-to-face invitations at our university, professional networks, social media outreach and snowballing (Bryman, 2016). The 34 participants varied in gender, academic degree, discipline and geographic location of their universities (Tables 1 and 2).
Participant Information (n = 34).
Detailed Profile of Individual Participants (n = 34).
This study was approved by the Institutional Ethics Committee of King Faisal University (KFU-REC-2024-OCT-ETHICS2822) University for studies involving human participants. Participation was entirely voluntary, and informed consent was obtained before commencement. The participants were assured of their right to decline to answer any questions and withdraw from the study at any time. They were also informed that pseudonyms would be provided to them to analyse and report the data and that all data would be analysed by the researchers and securely disposed of following publication.
Data Collection Methods
We collected primary data through semi-structured interviews that lasted 40 to 90 min. Five interviews were conducted face-to-face, and the remainder were conducted online via Zoom. With permission, all interviews were audio-recorded, transcribed using Transkriptor software and accuracy-checked by the research team against the audio.
This study employed data source triangulation through maximum-variation sampling across disciplines and academic ranks (Lincoln & Guba, 1985). To collect the data, a semi-structured interview was chosen because it aligns well with the case study approach, providing in-depth insights into individuals’ experiences and opinions that might not be effectively captured through other methods (Merriam & Tisdell, 2016). The interview questions benefited from UNESCO’s AI competency framework for students, guidance for generative AI in education and research, and the literature on NNES publication pressures and academic integrity.
To maintain consistency, we used a standard interview guide comprising three core topics used with all the participants: (1) research context and English-publication requirements, (2) uses of GenAI across writing for publication in English and (3) evaluation of quality, integrity and ethics. Each topic included fixed core questions and optional probes for elaboration. Variations in interview length reflected the participants’ uses and data richness.
Some examples of the interview questions are as follows:
Which GenAI tools do you use for your research practices in publishing in high-ranked English journals? Why did you select these tools?
How do these tools help or hinder you in your writing for publication in English?
How do you verify GenAI outputs and decide what to accept or reject?
What challenges do you face when applying these tools in your writing for publication in English?
The interviews were conducted in the Arabic language because, as the participants’ mother tongue, it was more comfortable for them. The analysis was also carried out in Arabic to preserve cultural meaning (Bryman, 2016). Selected quotations for reporting were then translated into English and verified by a bilingual team member.
Data Analysis
As the research team, we comprised NNES academics with varied familiarity with GenAI and experience in publishing in English. We employed reflexive thematic analysis to analyse the data (Braun & Clarke, 2022). We followed six phases, starting first with repeated readings of interview transcripts, to gain a thorough understanding of the data.
In the second phase, each author generated initial codes to identify meaningful patterns across the data while focusing on the research questions (Table 3). The first author imported and analysed the data for accuracy using NVivo 14 software.
Examples of Generating Initial Codes from the Data.
In the third phase, these codes were organised into broader categories, forming initial themes through shared meanings and relationships (Table 4).
Examples of Generating Themes from Codes.
The themes were reviewed against the original dataset and refined for coherence and distinction. In the fifth phase, each theme was defined to reflect its essence. The three authors regularly met throughout the process to discuss and reach a consensus on the theme names. Finally, the analysis was written into a coherent narrative incorporating quotations for each theme.
Trustworthiness
Several strategies were employed to ensure the credibility of the findings (Lincoln & Guba, 1985), including triangulation. The three authors collaborated to collect and analyse the data, ensuring multiple perspectives and reducing bias. Detailed descriptions of the data collection and analysis methodologies demonstrated how the data ultimately informed the findings. Member checking was also used, in which the Word file of the initial findings was shared with the participants via WhatsApp to confirm the accuracy of the analysis of their interviews. In addition, an audit trail was maintained to enhance transparency, including a complete record of the procedures applied, such as raw data and detailed documentation of the data interpretation and analysis.
Findings
The data analysis identified five themes related to the research question: GenAI as a literature navigator, a writing enhancer, an intellectual partner, a method designer–analyser and a publication mentor (Figure 1). These themes presented how Saudi NNES researchers used GenAI and evaluated its quality and integrity when writing for publication in English.
Five themes explaining how Saudi NNES researchers use GenAI and evaluate its quality and integrity when writing for publication in English.
GenAI as a Literature Navigator
Most of the participants considered writing literature reviews the greatest challenge in their research, describing it as ‘boring and exhausting compared with analysing my own work and data’ (Dr Lama). Thus, many of them turned to GenAI services to ease this stage and reduce the heaviest load on English-language publishing.
Participants’ Use of GenAI as a Literature Navigator
The participants used GenAI to expedite and facilitate three tasks: finding appropriate references, reading and understanding them and summarising their content.
As English was not their first language, the participants first used GenAI to search for suitable references for their topics. Their tools varied from LLMs, such as ChatGPT, Gemini and Copilot, to AI-powered literature review tools, such as Elicit, SciSpace, Consensus, Paperpal and CoSchedule. Basmah, a PhD student, revealed, ‘I use Microsoft Copilot. I ask it to search for references on a specific topic, and it provides me with references and links. I then go in and verify them’. Prof. Sami said, ‘I use a tool called Consensus. I type the keywords for my topic, and it displays dissertations, theses and research papers on that topic’.
The participants emphasised speed and reduced effort compared with traditional methods, such as Google Scholar searches, scanning systematic reviews and journal alerts. Hasan, a PhD student, said, ‘I used to spend 2–3 weeks searching for, reading and understanding relevant references. Now, with GenAI, I can accomplish all of this in less than 1 week’.
After locating potential references, the participants used GenAI to simplify and clarify the papers. Many stated that reading English texts was difficult and time-consuming, so they would ask GenAI to explain and translate key sections into Arabic, their mother tongue. Some preferred AI-powered literature review tools with PDF chat, while others simply used LLMs. Dr Shahd explained, ‘I use SciSpace. I ask it to search for references relevant to my research. When I like a paper, I open the chat with the PDF, write my question in Arabic, and it replies in Arabic with highlighted English text’. By contrast, Safaa, a PhD student, said that she used ChatGPT to translate English articles into Arabic to make understanding easier.
The participants appreciated ChatGPT’s more accurate translations compared with older tools, particularly for specialised terminology. It outperformed Google Translate in rendering complex terms in fields such as neuropsychology and medicine and demonstrated accuracy by supporting translated concepts with evidence from Arabic references. This reflects the participants’ trust in its linguistic and contextual precision, as explained by Dr Maram.
The participants also relied on GenAI to produce brief, comparable summaries that helped them see patterns and gaps. This use was most common in scientific fields. According to Dr Fahad, ‘In medicine, we need to read a large number of papers in a short time. I use Elicit because it summarises studies in a table that shows the location, year and sample size on one page, making it easier to discover research gaps’. Researchers in the humanities and social sciences have also used this feature, although less frequently. Dr Bander said, “Some tools display reference summaries in a tree, arranged by date, language and country. They tell you, ‘This is 99% similar to your topic’ or ‘70% similar’.”
The participants further praised GenAI’s ability to present summaries in clear formats, such as tables, maps and trees, which helped them turn from reading many separate papers to having all key points gathered and organised in one place, ready for writing. As Dr Bander explained, ‘On Google Scholar, I need to open each link and read the abstract to judge relevance; GenAI tools like Elicit put multiple article summaries in one table on a single page, making it easier to compare studies and take notes’.
Participants’ Evaluation of GenAI’s Quality and Integrity
The participants focused on two issues in evaluating GenAI as a literature navigator: reliability and hallucination.
While the participants acknowledged great advantages, they repeatedly questioned the reliability of GenAI outputs. They noted that GenAI could produce human-like, persuasive text, but the information could be false, misleading or fabricated. Accordingly, they avoided accepting outputs without human scrutiny.
The participants also compared LLMs with AI-powered literature review tools and generally found dedicated platforms to be more accurate for suggested references, although not always reliable. As Basmah, a PhD student, revealed, ‘ChatGPT usually gives me fabricated or non-existent references’. In addition, Dr Mona expressed frustration that even specialised tools sometimes display references unrelated to her topic. Several participants observed that some literature tools draw on moderate or weak databases, which many journals may not accept. Dr Hajer explained, ‘I noticed that Elicit suggests references from Semantic Scholar, which includes research from weak journals. High-ranked journals refuse references from weak sources’. In response, the participants described the additional time needed to verify every reference suggested by GenAI, checking authenticity, journal quality and citation accuracy.
Moreover, the participants highlighted the risk of AI hallucinations in summaries and explanations. Some reported cases in which GenAI invented details that were not present in the original references but presented them confidently. Dr Omar said, ‘I use Gemini with extreme caution when explaining or summarising research because I have noticed that it sometimes hallucinates and gives me incorrect, incomplete or even redundant information. I must review it afterward’. Similarly, other participants noted that although GenAI saved significant time, they remained cautious about its outputs. They would quickly check each suggested study to verify journal quality and confirm that the extracted information genuinely existed, rather than being fabricated.
GenAI as a Writing Enhancer
Almost all participants used GenAI tools for writing enhancement tasks, such as proofreading, editing and paraphrasing, to meet English-language publication standards while protecting their meaning and authorial voice. Some relied on LLMs, such as ChatGPT and Gemini, while others preferred dedicated editing tools, such as Grammarly.
Participants’ Use of GenAI as a Writing Enhancer
The participants described four approaches. First, some participants drafted sections in their first language and then produced an English version either with traditional tools, such as Google Translate, followed by an LLM for refinement, or directly with an LLM for translation and polishing. They consistently felt that LLM translations sounded more natural than older tools. As Dr Njood said, ‘It is easy to write in the language you think in, so I write in Arabic, give ChatGPT the paragraph, and then ask it to translate it into English’.
Second, the other participants wrote the manuscript in English and used GenAI for grammar and style checks only. This was more common among researchers with greater English proficiency, such as those who studied abroad or obtained an English undergraduate degree. Prof. Ahmed said, ‘My subject of study is primarily Applied Linguistics, so I sometimes need ChatGPT for proofreading and correcting minor errors only’.
The third method is similar to the second. The participants were the authors of the text that they used in a previous study. The participants wanted to reuse it in a new study, but to avoid plagiarism, they needed to paraphrase it. Thus, they turned to GenAI for this purpose. Dr Lama stated, ‘To avoid self-plagiarism, I provide ChatGPT with specific paragraphs from the Structural Equation Modelling methodology, which I frequently use in my research, and ask it to slightly rephrase my writing for use in a new study’.
The fourth method, which is rare, involves asking GenAI tools to write the entire research paragraph. This emerged among a few postgraduate students. These participants took it as is, or with minor modifications, and included it in their research as their own writing. In this case, the participants did not ask the tool to write directly but rather provided it with the research content with proper references and only asked the tool to rephrase and link the paragraphs together. According to master’s student Reema, ‘When I write about literature reviews, I copy and paste paragraphs from different studies, write the citations under each section, and give them to GenAI and ask it to completely rephrase them’. While Reema provided GenAI with the references, PhD student Sara explained how she used this service in her doctoral thesis without providing any references: ‘I asked the tool for references on transitional services for people with disabilities. This tool wrote me a complete paragraph of approximately five lines, correctly documented from four references and also provided links to the references’. However, junior and senior researchers refused to use this method. Dr Bander pointed out that there are tools dedicated to literature reviews that offer paraphrasing and rewriting services upon request. Nevertheless, he expressed hesitation in using them for fear of plagiarism.
Participants’ Evaluation of GenAI’s Quality and Integrity
The participants discussed their evaluations of using GenAI for writing enhancement across five aspects: boundaries for acceptable copying, paraphrasing and plagiarism; disclosure when submitting to journals; limits to full-document proofreading; privacy and copyright concerns; and the erosion of creativity and expressive voice.
Some participants expressed confusion and questioned whether copying and pasting text generated by GenAI constituted plagiarism and whether journals could detect such use. They cited the vagueness of ethical guidelines and the lack of clarity regarding the use of GenAI in writing research. Dr Fouz said, ‘The problem lies in the ambiguity and lack of knowledge about what is and is not permitted in the use of GenAI tools in research. Because English is not our native language, we struggle to write texts correctly. Therefore, GenAI tools greatly assist us in paraphrasing and in the writing itself’.
Some believed that using GenAI to write or paraphrase theoretical sections, such as literature reviews, might be acceptable if researchers verified the references used. However, they considered using GenAI to write practical aspects, such as fieldwork procedures, findings and discussions, unacceptable, as these tasks should be performed by the researchers. Dr Omar said, ‘I believe that writing the discussion part is the responsibility of the researcher, as they are the one who analysed the data and wrote the findings. They must interpret them from their own perspective, not using GenAI’.
Most participants acknowledged that copying GenAI output and pasting it into their research is illegal and unethical. This perception is more widespread among junior and senior researchers than among postgraduate students. Prof. Amena likened this misuse of GenAI to the unethical practices of ‘illegal research offices that sell entire studies to researchers’. Dr Njood asserted, ‘Copying texts and attributing them to researchers is not permissible under Islamic law because it constitutes a lie and a false claim’.
Many participants expressed concern about the risk of high-ranked journals discovering their use of GenAI and rejecting their manuscripts for academic misconduct. Dr Adam revealed, ‘I received a desk rejection from a high-ranking journal because 48% of my study was identified as using GenAI, even though I only used ChatGPT to paraphrase some parts’. Consequently, many participants rewrote AI-generated texts in their own words to avoid detection. Dr Maram explained this process sarcastically, saying, ‘I use GenAI to write the literature review, then rewrite it myself. But because of my poor English, I send it back to GenAI for editing. Then, I get anxious and edit it a little myself to avoid the AI detection tools. It is a vicious loop’. Some participants also used AI detection tools to check similarity ratios, but they criticised their inaccuracy, as they sometimes flagged original work as AI-generated.
Some participants complained that GenAI tools refused to proofread an entire manuscript with one click. Thus, they had to break the manuscript into smaller parts and ask the tool to proofread each part. Alternatively, they sent the manuscript to a human proofreader, which was faster and easier. Some participants also noted that LLMs do not just check the language but also add sentences and words not written by the researcher. Prof. Salem said, “I have to review my writing after ChatGPT proofreads it because it adds extra words, even though I only wrote in the prompt, ‘proofread the language, do not change my words, and do not add any extra words’.”
Some participants also voiced their worry about using GenAI for proofreading and writing improvement tasks, fearing that their data or research content might be stolen or stored in its memory and then shared with others. Dr Noor said, ‘There is no security document between me and the GenAI tools that stipulates the protection of my data or that it would not be used for training. I do not give my work to ChatGPT for proofreading because I am afraid it might be stolen’.
To protect her work, master’s student Reema explained how she misleads GenAI tools so that they cannot memorise her entire manuscript: ‘I give ChatGPT unorganised parts of my research and ask it to proofread them. Then, I delete the chat, open another one, and give it the other parts. This way, it does not have my entire research’.
While the participants acknowledged that GenAI helps them improve their English writing, they expressed concern about their creativity and expressive abilities being undermined by their dependence on GenAI tools for writing improvement and paraphrasing tasks. In the same vein, some researchers likened the use of AI-powered literature review tools to students relying on summaries instead of textbooks, arguing that this practice impairs creativity, critical thinking and understanding of the research topic. They believed that GenAI should be limited to brainstorming and proofreading, while the responsibility for the literature review and the actual writing should remain with the researcher. They believed that this would enhance researchers’ skills and distinguish them from their peers.
GenAI as an Intellectual Partner
The participants described using GenAI primarily as an intellectual partner sought for advice and consultation on research ideas and questions.
Participants’ Use of GenAI as an Intellectual Partner
The participants discussed four tasks with which they utilised GenAI: brainstorming, organising outlines, interpreting and discussing the research findings and explaining specific research issues.
First, the participants used GenAI as an intellectual partner in brainstorming, especially when selecting and refining research ideas. They either requested ideas in their field or presented an unclear idea and asked GenAI to help develop it into clear and valuable research ideas. Dr Maram said, ‘In my last paper, I had an idea in mind, but it was not clear. So, I asked ChatGPT, and it answered. I kept asking until the research idea was organised. I benefited a lot from it’.
They also used GenAI to develop organised outlines for the manuscript, theoretical framework, literature review and even the methodology sections. As Prof. Sami explained, ‘Sometimes, I am confused about the most important aspects I should write about in a literature review, so I consult ChatGPT, give it my research objectives, and ask it to create an outline suggesting the most important points’.
Second, some participants struggled with individual brainstorming when interpreting and discussing the research findings, especially in English, and sought a partner to suggest interpretations. They found this support in GenAI. According to Dr Lama, ‘I benefited greatly from ChatGPT in interpreting the results. It is like a co-researcher: We sit and discuss – why did this result come out? What could be the reasons? What should we write to explain it?’ Some participants also asked GenAI to link their findings to prior studies or to interpret them in light of existing theories.
Third, the participants used GenAI as a private tutor for quick, targeted explanations about theories or statistical analyses, and they preferred it to YouTube videos or long courses. Dr Omar said, ‘I use Gemini as a private tutor to explain certain theories or statistical analyses that I could not understand. Honestly, it is faster and cheaper than taking a long course when I only need one topic!’
Fourth, some participants, especially postgraduate students, viewed GenAI as an expert supervisor guiding them from idea inception to publication. They praised GenAI’s availability and patience, answering anytime, anywhere, without boredom or frustration. Master’s student Nora said, ‘I consider ChatGPT a supervisor who answers me day and night without getting bored or complaining’.
Participants’ Evaluation of GenAI’s Quality and Integrity
Although the participants acknowledged the significant benefits of GenAI, they also expressed concerns about it. First, they were worried that heavy reliance on GenAI could weaken their critical and creative thinking, which required reflection and human-led brainstorming. Some feared that depending on GenAI for brainstorming would undermine their ability to think and even write independently in the future. In the same vein, Dr Yahya said, ‘A researcher who cannot paraphrase a text and relies solely on ChatGPT does not deserve to be in the research field’.
Second, the participants cautioned that adopting GenAI outputs directly could produce similar work and obscure the researcher’s unique voice, leading to a loss of originality and research identity. Master’s student Doa explained, ‘I do not like consulting ChatGPT too much in my research because it gives me duplicate and similar results to others. I like to think for myself and build a research personality that stands out from others’.
GenAI as a Research Designer and Analyst
The participants described GenAI as a practical aide for research design and analysis, such as drafting instruments and materials and running preliminary analysis.
Participants’ Use of GenAI as a Research Designer and Analyst
The participants used GenAI in tasks related to research procedures that can be summarised in three areas: producing specific research materials, designing data collection instruments and analysing collected data.
Several participants used GenAI’s services to produce research materials tailored to their objectives. These materials varied across audio, visual and textual formats. Master’s student Sama said, ‘Because I did not want my voice to be widely spread among students, I used a GenAI tool to convert text into high-quality, expressive audio, which I could share with students to practise expressive reading’. Some participants expressed appreciation for GenAI’s ability to generate high-quality images, save costs on hiring graphic designers, and assist researchers who lack the ability to create visual materials. Others noted that they used GenAI tools to create video clips for use in their research, and some even mentioned using GenAI to produce poetry for research purposes. Some researchers also used GenAI to generate graphs and tables from the data they provided, such as tables describing the research sample or graphs of selected findings.
Some participants indicated that they used GenAI tools to create and design data collection instruments, such as questionnaires or interviews. They initially asked GenAI to design the instruments based on the research objectives and the questions they provided. Others preferred to design their own tools and then submit them to GenAI for refinement and modification.
Regarding data analysis, the participants’ views varied. Most were sceptical of GenAI’s ability to perform analyses correctly and reliably and preferred traditional statistical software, such as SPSS and R, because they are more trustworthy. Conversely, some participants uploaded questionnaire data with their research questions to ChatGPT and asked it to determine the appropriate statistical methods and conduct the analysis. Others tested GenAI’s statistical capacity by first analysing the data using statistical software and then asking ChatGPT to analyse the same data using the same statistical method. As Dr Omar said, ‘I tested GenAI’s statistical analysis capabilities, and it performed remarkably well’.
For qualitative data analysis, most of the participants preferred their own human analysis, believing that no matter how intelligent GenAI was, it would not be able to generate appropriate codes or themes. PhD student Sara said, ‘I gave ChatGPT an interview excerpt and asked it to code using the thematic analysis method, guided by my research question. I was not satisfied with the analysis due to its superficiality and failure to capture implicit meanings between the lines’.
Participants’ Evaluation of GenAI’s Quality and Integrity
The most significant challenge in using GenAI as a research designer was the fear of intellectual property infringement. The participants were worried that GenAI might provide them with designs of images or videos that it would also provide to others, exposing researchers to potential plagiarism or theft. They also faced the challenge of whether these products could be considered their own intellectual property when they did not produce them themselves.
The greatest challenge in using GenAI as a data analyst was the potential for errors and inaccuracies, which could create problems with the journals they sought to publish. Dr Fahad said, ‘I noticed a problem with Gemini’s statistical analysis of my research data. It added fake data, which caused significant errors in the analysis’. Therefore, the participants who used GenAI for data analysis were careful to review the outputs closely and avoided relying on them entirely.
GenAI as a Publishing Mentor
As NNES researchers are required to publish in high-ranked journals, most of the participants emphasised their need for a mentor throughout the publishing process.
Participants’ Uses of GenAI as a Publishing Mentor
The participants described six stages in the publication process: selecting the appropriate journal, receiving guidance on how to write a manuscript suitable for submission, preparing the manuscript according to journal requirements, adjusting the documentation and citation style, understanding reviewers’ comments and modifying the manuscript accordingly and writing proposals for funding support or research ethics approval.
First, several participants reported using GenAI’s specialised publishing tools to help choose journals aligned with their topic and objectives. Dr Fahad explained, ‘I use tools like Trinka: I give it my research abstract, and it recommends suitable journals and suggests how to refine the paper for fit. This guidance helps me avoid desk rejection for being outside a journal’s scope’.
Second, some participants described how they benefited from GenAI when writing a manuscript ready for submission to highly ranked journals. Some consulted ChatGPT on how to write sections, such as methodology, findings and implications, in a professional manner suitable for ranked journals. Prof. Sami revealed, ‘Ranked journals pay great attention to the implications and how they can benefit the readership, so I consult ChatGPT to help me write my research implications in the best possible way’. PhD student Sara added, ‘I had to publish an English paper from my PhD thesis, but cutting it down was hard. I used ChatGPT to shorten and paraphrase sections while keeping the core ideas and meaning intact’.
Third, many participants praised GenAI services for preparing manuscripts according to journal requirements. Some used specialised GenAI tools to evaluate and review manuscripts against the target journal’s guidelines. Dr Ali said, ‘I subscribe to a GenAI tool. It reviews your manuscript and gives you implications for appropriate journals. It even reads the journal’s requirements, reviews your manuscript and provides you with recommendations or a report for modifications based on those requirements’. In the same context, many participants complained about spending time and effort formatting a manuscript for one journal only to face rejection and then having to reformat for another journal. They found relief and saved time and effort by using GenAI tools to advance the reformatting for the next submission.
Fourth, most of the participants confirmed using GenAI for documentation to avoid referencing errors. Some provided the references and asked GenAI to format them, both in-text citation and full bibliography, according to a specific style, such as APA or Harvard. Master’s student Hana said, ‘As a novice researcher, there are many things I do not know about documentation, such as conferences or book chapters. This requires precision. So, I have started using GenAI tools that automate and speed up documentation according to my requested citation format’. Others used GenAI to convert one documentation style to another based on journal requirements. Dr Shahd explained, ‘The most exhausting part is formatting references. I arrange everything in APA style. The paper is rejected, and the next journal wants IEEE. I used to pay a specialist, but now I convert styles on my own in seconds with ChatGPT’.
Fifth, several participants used GenAI to understand reviewers’ comments. They copied and pasted comments into ChatGPT or Gemini and asked for explanations, guidance on how to revise the manuscript accordingly and help to draft a response letter. Master’s student Eman explained, ‘I could not understand the reviewers’ comments due to my poor English, and there was no way to contact them. I typed comments into ChatGPT, asked it to explain them in Arabic, guide me on edits, and write responses in English’.
Sixth, the participants further benefited from GenAI when writing research proposals for funding support or ethics approval. They provided the research title, objectives and key information and then asked GenAI to draft the proposal. Dr Fahad said, ‘As a doctor, my research requires careful writing to ensure that ethics and patient rights are protected. Instead of drafting proposals myself, I write ideas on Gemini and ask it to produce a research proposal with my required conditions and aspects’.
Participants’ Evaluation of GenAI’s Quality and Integrity
Many participants mentioned inaccuracies when GenAI prepared manuscripts to meet journal requirements and formats. Some also reported mistakes in documentation and citations. Dr Saeed said, ‘I need to be careful with ChatGPT’s work in writing the bibliography of a manuscript I give it because it makes mistakes’. The other participants noted that, sometimes, ChatGPT tended to exaggerate the importance of the research and its implications in funding proposals, even adding content unrelated to the proposed study, as Dr Fahd mentioned.
Discussion
The findings show how Saudi NNES researchers used GenAI when writing for publication in English and how they evaluated the quality and integrity of their uses. The participants used GenAI as a writing enhancer, an intellectual partner, a method designer–analyser and a publication mentor. They evaluated each use from the quality and integrity perspectives.
For the participants, the most common uses of GenAI were as a literature navigator and as a writing enhancer. Writing the literature review is difficult and time consuming because of its complexity and multi-stage process: searching for appropriate references, reading and understanding them, summarising content, connecting ideas, linking sections drawn from different sources and writing academically. Therefore, the participants appreciated any support to assist them in this important process, including GenAI tools. Previous studies have proven the benefits of GenAI as a literature navigator, which can assist researchers in identifying a wide array of prior studies, summarising them (Srivastava & Agarwal, 2024) and finding research gaps (Liu et al., 2024).
We observed that researchers in scientific disciplines (e.g., medicine and engineering) were most likely to use GenAI for literature review tasks. A possible reason for this is their need to scan large volumes of papers, but their core writing work emphasises laboratory procedures rather than extended writing. By contrast, researchers in the social and humanities sciences were most likely to use GenAI to enhance their academic writing, as they require extensive writing because scholarship in these fields centres on critical argumentation, persuading readers and articulating the writer’s voice (Jin et al., 2025).
Although all disciplines require wide reading, scientific fields lean more towards rapid skimming and scanning, in which GenAI helps by providing organised summaries and comparison views, whereas social and humanities fields often require reading full papers to inform both the literature review and the discussion, which is less supported by GenAI.
Tenopir et al. (2019) found that 70.9% of their participants from the social sciences, and 76.0% of those from the humanities and fine arts, went beyond reading the title and abstract into the body of the article, showing great care to read and understand most or all of the articles they read.
Academic writing imposes an additional burden on NNES researchers seeking to publish in English. Writing itself is an exhausting process requiring sustained cognitive effort, complex multi-stage tasks and meticulous precision (Chauhan & Currie, 2024; Katongtung et al., 2025). GenAI can reduce the cognitive burden of writing, enabling researchers to concentrate on higher-order tasks (Baek et al., 2024; Jin et al., 2025), convey ideas more clearly and with fewer ambiguities and generate well-structured, readable text that enhances overall flow and clarity (Aladsani, 2025; Liu et al., 2024).
Although all the participants used GenAI as a writing enhancer, those with lower English proficiency were the most frequent users of substantive language improvement, whereas the participants with advanced English tended to apply GenAI for light proofreading rather than paraphrasing or heavy editing. Baek et al. (2024) found that in writing tasks, NNES users tended to use ChatGPT frequently in their writing compared with those who were native speakers or had high-level English proficiency.
Postgraduate researchers also relied on GenAI to improve their writing due to their poor English proficiency and developing research skills; a few even used it to draft complete passages, not just for proofreading or editing. By contrast, junior and senior researchers were more cautious about relying on GenAI for writing enhancement, likely due to their awareness of the risks of copying GenAI output directly into manuscripts and the potential to be flagged by plagiarism and AI-detection tools. Several studies have indicated that applying GenAI as an editor to improve academic writing is common among postgraduate students, particularly those whose English is not their first language. They used it for correcting grammatical errors, ensuring conciseness and achieving an academic tone (English et al., 2025; Yao et al., 2025). Ravi et al. (2025) examined why university faculty researchers expressed greater concern than students about GenAI-generated writing, content ownership and academic integrity. They argued that faculty tend to be more change averse, whereas students are often early adopters of new technologies. In addition, faculty members are typically more attuned to academic integrity standards and the consequences of plagiarism than students.
Our participants used GenAI to improve their academic writing, either to translate their Arabic version into an English academic version, polish the initial English version they produced or paraphrase their previous writing sections to be used in other manuscripts. These findings are consistent with those of Yao et al. (2025), who found that NNES researchers used GenAI primarily to correct grammatical errors and change their writing style to resemble academic writing.
The participants highlighted substantial benefits from using GenAI across research tasks, with speed and time saving being the most prominent. This was especially evident when GenAI was used as a literature navigator, and the participants contrasted it with the traditional methods of searching, reading, summarising, understanding and translating references. Instead of opening multiple papers and manually extracting key points, with GenAI tools, researchers can move on more quickly from searching to organising notes ready for writing (Srivastava & Agarwal, 2024).
The participants also emphasised translation accuracy, whether from English to Arabic or vice versa. They noted clearer renderings of specialist terms using GenAI tools rather than older tools, thus reducing misunderstandings and lowering the linguistic barrier to engaging with complex sources.
Besides the literature work, the participants reported benefits in brainstorming and idea generation. They used GenAI to discover and refine research gaps, plan the overall manuscript and outline sections before drafting. In this regard, Liu et al. (2024) and Nicholas et al. (2024) found that GenAI supported early-stage thinking by providing scaffolds that made research starting easier and progression more structured.
The participants further described GenAI as a unique intellectual partner, offering suggestions and alternative framings when discussing and interpreting findings while also serving as a personalised tutor and patient supervisor who guided learning and manuscript development from inception to publication. They also relied on GenAI as a publishing mentor, using it to identify suitable journals, interpret reviewer comments and draft response letters with the appropriate tone and structure. Similarly, Berg (2023) stated that GenAI can assist researchers to improve their manuscripts by playing the role of a critical reviewer, pinpointing weaknesses much like a peer reviewer or colleague and providing feedback that closely mirrored real peer-review reports. Berg (2023) also explained how GenAI can serve as a personalised training tool, delivering method-specific guidance tailored to the researcher’s needs, unlike traditional training programmes, which often cover more than is necessary.
In addition to saving time, GenAI reduced costs by providing free or low-cost services that had been previously paid. The participants most often mentioned proofreading and language polishing, artwork or figure design and reference formatting, all of which helped them reallocate limited resources to the substantive elements of their research. Alqahtani et al. (2023) emphasised that both postgraduate students and faculty suffer from less time or resources to dedicate to writing, as preparing a scientific article is time intensive and often requires expenses when authors hire professional editors or designers. By using GenAI to generate and refine text, researchers can reduce or avoid these outlays and replace paid services with low-cost, readily available tools.
When our participants evaluated their use of GenAI in writing for publication in English, they demonstrated strong awareness of embodied ethics, safe and responsible use and human agency, which are three key competencies in the UNESCO AI competency framework for students (Miao & Shiohira, 2024) and in the UNESCO guidance for generative AI in education and research (Holmes & Miao, 2023). This awareness was evident in their insistence on verifying GenAI outputs rather than accepting them as absolute. Their caution aligns with studies showing that GenAI can produce logically convincing text that appears valid to non-experts but contains hallucinations and false information supported by fabricated references (Aladsani, 2025; Chauhan & Currie, 2024; Jin et al., 2025).
Accordingly, most of the participants preferred AI-powered literature review tools over general LLMs because the former can highlight the location of information within the original source, improve traceability and reliability and address the risks of misleading information and fabricated references (Jain et al., 2024). Nevertheless, some participants noted that these tools can produce inaccurate references or flawed summaries; thus, they adopted cautious practices, such as uploading their own sources to be summarised rather than relying on suggested references. Whitfield and Hofmann (2023) similarly observed inaccuracies, including inappropriate reference suggestions and misleading summaries or comparisons. The participants also noted that some GenAI tools, such as Elicit, rely on open-access articles in the Semantic Scholar database, which could present research from weak or unranked journals.
As NNES researchers are required to publish in high-ranked English-language journals, they show heightened responsibility for selecting high-quality sources. This reflects embodied ethical competence, which involves choosing appropriate GenAI tools for specific purposes while taking full human responsibility for reviewing outputs, verifying accuracy and validity and bearing the consequences of any failure to do so (Holmes & Miao, 2023; Miao & Shiohira, 2024).
Despite their recognition of GenAI’s benefits, such as accuracy gains and savings in time and money, the participants raised concerns about data protection, fearing that uploaded text could be retained on external servers, leaked to other users or later flagged by detection tools as AI-generated rather than human-authored. This demonstrates their safe and responsible use competence, which means using GenAI in ways consistent with ethical principles and local regulations, recognising the risks of privacy breaches and taking care to protect personal and research data (Miao & Shiohira, 2024).
In addition to safe and responsible use, the participants demonstrated human agency competence, emphasising that AI must remain under human control (Holmes & Miao, 2023; Miao & Shiohira, 2024). They distinguished between tasks appropriately supported by GenAI and those requiring human expertise, such as data collection, writing up results and interpreting findings. Although some participants used GenAI to help develop research instruments, they collected data themselves, noting that AI-generated data would be unacceptable and would constitute scientific misconduct and fabrication (Chauhan & Currie, 2024; Katongtung et al., 2025; Nicholas et al., 2024). The participants generally limited GenAI’s role to writing to inspire, brainstorming and proofreading, not to generating entire sections, because passing off AI-produced text as one’s own was viewed as unethical and harmful to academic integrity (Aladsani, 2025; English et al., 2025; Hadan et al., 2024). This assessment aligns with human agency competence: GenAI must be guided by humans, remain under their control and never replace their judgement. The participants rejected being wholly controlled by, or entirely reliant on, GenAI for generating research texts (Holmes & Miao, 2023; Miao & Shiohira, 2024).
Moreover, the participants raised concerns that GenAI might erode creativity, writing ability and critical thinking. Many feared that habitual reliance on seemingly competent outputs could diminish distinctly human capacities. Previous studies have echoed these worries, suggesting that over-reliance on GenAI as an information source and writing partner may blur the distinction between human abilities and homogenise scholarly outputs (Al Lily et al., 2023), with potential downstream effects on future generations’ capacity to evaluate information critically and seek alternative sources, risking a form of machine dominance over human cognition and knowledge acquisition (Da Silva et al., 2024). These concerns reflect a mature sense of human agency: GenAI must remain subordinate to human intelligence, and uncritical dependence on its outputs carries serious consequences (Holmes & Miao, 2023; Miao & Shiohira, 2024). UNESCO similarly emphasises the need to enhance critical thinking and to evaluate GenAI outputs carefully, rather than accepting them as established facts (Holmes & Miao, 2023).
Despite the generally high level of ethical awareness, some participants, especially postgraduate researchers and novice users of GenAI, expressed uncertainty about what is and is not permissible when incorporating GenAI outputs into manuscripts and about the extent of their intellectual property rights over such outputs. This confusion may stem from limited experience with the technology and unfamiliarity with the evolving guidance from international and national bodies on ethical GenAI use in research and publication, such as UNESCO’s guidance and the Saudi RDIA (2025). This confirms the need to strengthen global and local efforts to build ethical awareness among researchers. Holmes and Miao (2023) note that the absence of national guidelines and policies exposes human values, ethical integrity and data privacy to risk. Therefore, UNESCO urges every educational institution to establish clear, sustainable policies that explain how to use GenAI safely and ethically in education and research, clarify what is permitted and what is prohibited and raise awareness of practices that protect and develop human capabilities rather than suppress them (Holmes & Miao, 2023).
Conclusion
This qualitative case study examined how Saudi NNES researchers use GenAI and evaluate its quality and integrity when writing for publication in English. Overall, they used GenAI as a literature navigator, a writing enhancer, an intellectual partner, a method designer–analyser and a publication mentor. Moreover, they evaluated these uses from a quality and integrity lens.
Implications
The qualitative findings are not statistically generalisable, but they are analytically transferable. While these findings are anchored in the Saudi context, they speak to wider systems in which publication in English is compulsory despite it not being the national language, a recurring reality in many academic communities (e.g. Gulf countries, Spain, Turkey, China and Japan).
International and local bodies, such as UNESCO and RDIA, may draw on these findings when refining guidance and competency frameworks that address AI use, language barriers and inequities in English-dominant publication systems.
Research institutions and NNES researchers can use these insights to use GenAI wisely and strategically, maximising its benefits for language support, design, idea generation and publication processes while safeguarding integrity, human agency and authorial voice.
Editors of ranked journals may use these findings to refine GenAI-related policies that protect fairness and equity for NNES authors, clarify acceptable uses and explain what is permitted and what requires disclosure, aligning with research integrity.
Targeted support must be provided through training programmes to postgraduate students who, in our study, seemed to rely more heavily on GenAI in writing for publication in English to improve their higher GenAI competencies, specifically critical and analytical thinking and ethical and legal competencies.
Limitations and Directions for Future Research
This study has several limitations that future research could address. First, it is a single-case study of Saudi NNES researchers who were required to publish in English in high-ranked journals. Therefore, we suggest cross-cultural or multi-country comparative studies in non-English-speaking contexts in which English publication is mandatory, focusing on disciplinary and career stage differences.
Second, as a qualitative case study and not a comparative design, it does not allow the systematic comparison of all findings by rank, discipline or gender. Future research could adopt quantitative or mixed-methods approaches to enable clearer and more precise comparisons.
Third, most of the participants were from the educational technology discipline, reflecting access and our purposive strategy to recruit information-rich cases that met the inclusion criteria. This limitation could be mitigated by sampling across additional disciplines. Future studies could focus on researchers in scientific fields or compare GenAI use in English-language publishing between the scientific science and the social sciences and humanities.
Footnotes
Ethical Considerations
The study was approved by the Research Ethics Committee (REC) of King Faisal University (KFU-REC-2024-OCT- ETHICS2822). Informed consents were obtained from all involved parties.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Deanship of Scientific Research, King Faisal University, under Grant (KFU253960).
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
The dataset generated and analysed in the current study is not publicly available due to the privacy policy. However, it can be made available upon reasonable request.