Generative Artificial Intelligence (GenAI) for Academic Writing in Higher Education: A Scoping Review of Applications, Challenges, and Implications

Generative Artificial Intelligence (GenAI) for Academic Writing in Higher Education: A Scoping Review of Applications, Challenges, and Implications | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Systematic Review Generative Artificial Intelligence (GenAI) for Academic Writing in Higher Education: A Scoping Review of Applications, Challenges, and Implications Renz Alvin E. Gabay, Aaron A. Funa, Jhonner D. Ricafort This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7440784/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Generative artificial intelligence (GenAI) is reshaping academic writing in higher education faster than institutions can develop evidence-informed guidance, leaving practice ahead of proof. To clarify what is happening and where benefits and risks cluster, the researchers conducted a scoping review structured by a Population–Concept–Context (PCC) frame and aligned with PRISMA-ScR procedures. Peer-reviewed, English-language empirical studies published from 2024 through Q2 2025 in higher-education settings were included, and findings were synthesized via convergent integration that juxtaposed quantitative distributions with qualitative themes. A total of 25 studies met criteria. Across populations and contexts, GenAI was most often positioned as assistive scaffolding across the planning-to-revision span of writing; reported benefits concentrated on organization, fluency, efficiency, and language support (notably for multilingual writers). Recurrent risks included hallucinations and unreliable or fabricated citations, inconsistent disclosure or attribution, and overreliance when use was unscaffolded; the limited reliability of AI-detection tools complicated integrity judgments. Context shaped practice: clearer policies and better access supported more constructive use, while the evidence base skews toward English-medium, well-resourced institutions and relies heavily on short-term or proxy outcomes. By integrating counts and themes within a PCC frame, this review offers an up-to-date evidence map that distinguishes where benefits reliably cluster (process-level supports) and where risks persist (source work and attribution), while surfacing salient gaps (faculty/postgraduate cohorts and Global South contexts). Overall, the pattern supports an assistive, not substitutive stance in which GenAI complements—rather than replaces—human judgment in argument construction, source interrogation, and synthesis. Artificial Intelligence and Machine Learning Academic writing ChatGPT GAI higher education large language models scoping review Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. Introduction Generative artificial intelligence (GenAI; referred to as “GAI” in some studies) is transforming how people approach writing, communication, and knowledge creation. In the post-pandemic period, as remote learning and digital platforms became integral to education, tools such as ChatGPT, Gemini, and other large language models (LLMs) entered mainstream academic spaces and are now used for brainstorming, paraphrasing, and even drafting research papers (Emran et al., 2024 ; Meyer et al., 2023 ). Scholars, educators, and students increasingly recognize that these technologies can enhance productivity and creativity in academic writing (Funa & Gabay, 2025a ; Khalifa & Albadawy, 2024 ). At the same time, their implications for academic integrity, authorship, and writing practice remain contested and not yet fully understood (Acut et al., 2024 ; Funa & Gabay, 2025b ). This review examines the emerging literature on GenAI’s role in academic writing, mapping current evidence, identifying gaps, and clarifying ongoing debates through a comprehensive scoping review. Understanding GenAI’s influence is crucial for shaping future academic policies and pedagogies. GenAI refers to AI systems designed to produce new content—text, images, or code—rather than merely analyze existing data. Its conceptual foundation can be traced to research on generative models, particularly the introduction of generative adversarial networks (GANs) by Goodfellow et al. ( 2014 ), which showed that AI could learn patterns in data and create novel outputs. Contemporary tools such as ChatGPT and Gemini are built on what the Stanford Center for Research on Foundation Models (CRFM) terms foundation models: large-scale, self-supervised deep learning systems trained on vast datasets and adaptable to many tasks. These models provide the underlying architectures that power modern GenAI applications and enable language, image, and code generation at scale (Bommasani et al., 2021 ). In academic settings, GenAI tools are widely used for idea generation, paraphrasing, literature summarization, and drafting, reshaping how students and scholars approach writing (Acut et al., 2024 ; Emran et al., 2024 ; Funa & Gabay, 2025a ; Kasneci et al., 2023 ; Meyer et al., 2023 ). A recent survey of medical students in the United States found that 48.9% had used ChatGPT in their studies; among users, 43.7% reported weekly to daily use, most commonly for writing, revising, editing, and summarizing. Notably, 37.5% and 41.3% reported using ChatGPT for these tasks for more than 25% of their working time (Zhang et al., 2024 ). Funa and Gabay ( 2025a ) similarly observed that faculty across generations in higher education use GenAI primarily for ideation and rapid feedback, with younger participants tending to trust outputs more readily. Kasneci et al. ( 2023 ) highlight potential benefits for students and educators—such as quiz generation, simplification of complex content, and adaptive feedback—while cautioning about bias, overreliance, and ethical concerns. Consistent with these cautions, Funa and Gabay ( 2025b ) and Meyer et al. ( 2023 ) emphasize that although GenAI can improve clarity, grammar, and readability, including for non-native English users, scholars should remain vigilant about factual inaccuracies, ethical issues, and model biases. To interpret these emerging practices, this review draws on complementary theoretical frameworks. It is anchored in socio-constructivist perspectives on writing and learning, which view writing as a socially mediated process shaped by tools, contexts, and interaction. From this perspective, learners develop writing proficiency through dialogue with peers, engagement with cultural tools, and iterative practice supported by scaffolds (Flower & Hayes, 1981 ; Vygotsky, 1978 ). GenAI tools such as ChatGPT and Gemini can therefore be conceptualized as cognitive and metacognitive scaffolds that provide immediate feedback, offer alternative phrasings, and suggest structural improvements to support planning, revision, and refinement. These tools function as mediators in the social process of writing and may extend a writer’s zone of proximal development by offering access to language models and ideas that might otherwise be unavailable. In parallel, the paradigm of foundation models (Bommasani et al., 2021 ) situates GenAI within a broader shift toward flexible, generalizable AI systems that mediate knowledge creation across disciplines. Integrating these perspectives provides a robust basis for examining how GenAI reshapes writing practices and for considering the pedagogical, ethical, and institutional implications that follow. Against this backdrop, current scholarship on GenAI and academic writing continues to expand but often addresses isolated tools, single-discipline applications, or specific aspects of writing support. Comprehensive syntheses that map patterns of use across educational contexts remain limited (Kasneci et al., 2023 ). Early advances in LLMs began influencing educational technologies around 2020 and laid the groundwork for new AI-mediated learning tools. The public release of ChatGPT in November 2022 marked a turning point, prompting rapid growth in research and adoption, particularly as digital and AI-assisted platforms became integral to academic workflows in the post-pandemic period (Bisi et al., 2023 ; Bommasani et al., 2021 ; Huh, 2023 ; OpenAI, 2022 ). This review adopts a scoping review approach, which is well suited to this topic because the field of GenAI is evolving rapidly, the available evidence is heterogeneous, and the objective is to map the breadth of existing literature rather than evaluate intervention effectiveness or test a narrowly defined hypothesis (Arksey & O’Malley, 2005 ; Tricco et al., 2018 ). It builds on a systematic review by Chanpradit ( 2025 ), who synthesized 30 empirical studies from 2023 to 2024 and reported gains in cohesion, clarity, creativity, fluency, and proficiency, alongside risks such as plagiarism, overreliance, hallucinations, bias, and unequal access. The review also recommended institutional guidelines, transparent data practices, human oversight, and structured training. Extending this work, the present scoping review covers studies published from 2024 through the second quarter of 2025 to capture both ongoing adoption and emerging developments in higher education. By synthesizing peer-reviewed studies from this period, the review maps current applications, challenges, and research gaps in the use of GenAI for academic writing and identifies priorities for further investigation. Guided by the Population–Concept–Context (PCC) framework from the Joanna Briggs Institute (Arksey & O’Malley, 2005 ; Levac et al., 2010 ; Peters et al., 2020 ), the researchers define the population as individuals engaged in academic writing in higher education (students, faculty, and researchers), the concept as the use of GenAI in academic writing, and the context as higher-education settings. Accordingly, the researchers address: (RQ1) How is GenAI being used in academic writing? (RQ2) What benefits and opportunities are reported? (RQ3) What challenges, risks, or ethical issues are identified? (RQ4) What gaps and future research directions are highlighted by existing studies? 2. Methodology 2.1. Research Design This scoping review followed the methodological framework of Arksey and O’Malley ( 2005 ), refined by Levac et al. ( 2010 ) and Peters et al. ( 2020 ), guided by procedures outlined in Funa et al. ( 2024 ), and reported in accordance with the PRISMA Extension for Scoping Reviews (PRISMA-ScR; Tricco et al., 2018 ). The approach was selected to map the breadth of evidence on the use of GenAI in academic writing and to identify research gaps in a rapidly evolving field. Eligibility and synthesis were structured using the PCC mnemonic: Population (students, faculty, and researchers in higher education), Concept (use of GenAI in academic writing), and Context (higher-education scholarly settings). 2.2. Search Strategy and Study Selection An initial search was performed using Harzing’s Publish or Perish (Harzing, 2007 ) to explore and retrieve relevant literature from multiple databases, including Google Scholar, Scopus, PubMed, Semantic Scholar, and Web of Science. A comprehensive list of keywords and descriptors was developed, refined, and iteratively tested across these databases. Boolean operators (AND, OR) were applied, and search terms were systematically combined and interchanged to ensure coverage across the three focal constructs: academic writing, GenAI, and higher education. For example, the primary term “academic writing” or “scientific writing” was combined with secondary terms such as “generative artificial intelligence,” “generative AI,” or “GenAI,” together with “higher education” or “university students,” using AND/OR. 2.3. Inclusion and Exclusion Criteria Following the recommendations of Levac et al. ( 2010 ), the inclusion and exclusion criteria were designed to align directly with the research questions and were refined through iterative team discussions during the initial screening phase. The study selection process is shown in Fig. 1 . A study was included if it met all of the following conditions: (a) it was a peer-reviewed journal article published between 2024 and the second quarter of 2025, reflecting the rapid advancements and evolving applications of GenAI during this period; (b) it made an explicit reference to the use, application, or impact of GenAI—such as ChatGPT, Gemini, or other LLMs—within the context of academic writing/scientific writing; (c) it was written in English to ensure accurate interpretation; (d) it reported original empirical findings, ensuring the synthesis was based on primary data or firsthand analyses rather than secondary syntheses; and (e) it focused on higher education institutions. Conference papers and other non-journal sources were excluded to maintain the rigor and comparability of the included evidence. Restricting the review to empirical studies in higher education ensured that the findings provided robust, evidence-based insights into the phenomenon under investigation. The criteria were piloted and refined before the full screening process to enhance clarity and ensure consistent application. Figure 1 presents the PRISMA-ScR flow for study identification, screening, eligibility, and inclusion for this scoping review (2024–Q2 2025). Across databases, 1,599 records were identified; 1,334 were removed prior to screening, leaving 265 titles/abstracts screened. Of these, 180 were excluded at screening and 85 full texts were assessed for eligibility. 60 full-text articles were excluded (primary reasons summarized in Fig. 1 ; the most common was no clear link between GenAI use and academic writing in higher education), yielding 25 studies included in the final synthesis. 2.4. Coding Procedures Data from the included studies were extracted and organized using a standardized data charting form developed by the review team. In line with Levac et al. ( 2010 ), the coding process was iterative, allowing the team to refine categories as familiarity with the literature increased. Each article was coded for key variables such as publication details (author & year), country/region, population (students, faculty, researchers), study context, study design/methodology, type of GenAI tool/s used, specific applications in academic writing (e.g., brainstorming, drafting, paraphrasing, summarizing), reported benefits and opportunities, identified challenges and ethical issues, and stated research gaps or recommendations for future studies. Two researchers independently coded each included study to enhance reliability. Discrepancies were discussed and resolved through consensus, with a third researcher consulted when necessary. Excel was used to facilitate organization and thematic grouping of data. Codes were grouped into higher-order categories aligned with the research questions, and emerging themes were refined through ongoing team discussion. Consistent with scoping review guidance, the objective was to map the evidence; therefore, no formal critical appraisal of study quality was undertaken. Findings were synthesized using a descriptive numerical summary of study characteristics and a qualitative thematic synthesis mapped to the PCC framework and the review questions. 2.5. Characteristics of the Included Studies This scoping review synthesized peer-reviewed empirical studies published from 2024 to Q2 2025 (n = 25) that investigated GenAI in academic writing within higher education contexts (see Appendix A). To aid interpretation, settings are grouped by continent (see Fig. 2 ): Europe (n = 11: United Kingdom [n = 7], Denmark, Norway, Greece, Switzerland), Asia (n = 7: China [n = 3], Hong Kong [n = 2], United Arab Emirates, Türkiye), Americas (n = 4: Ecuador [n = 2], Chile, United States), and Africa (n = 1: South Africa), with two cross-regional studies (one bi-national Finland–New Zealand; one global/multi-country survey). Across 25 studies, participants were predominantly student-only samples: 19 of 25 (76%); faculty-only: 3 of 25 (12%); and mixed student-faculty: 3 of 25 (12%). Sample sizes ranged from small, course-embedded cohorts (e.g., 6 students in Denmark [Jensen & Jensen, 2025 ]; 20 students at a Sino-British EMI university [Kim et al., 2025 ]) to large cross-sectional surveys (e.g., 2,555 students at a UK university [Johnston et al., 2024 ]; a global mixed sample of n = 1,217 students and lecturers [Yusuf et al., 2024 ]). Staff-only samples also appeared (e.g., n = 284 UK academics [Watermeyer et al., 2024 ]; n = 184 university teachers in Ecuador [Cordero et al., 2025 ]). Study designs were mainly descriptive and exploratory, comprising quantitative surveys (Johnston et al., 2024 ), qualitative interviews or focus groups (Hysaj et al., 2025 ), mixed-methods designs (Han, 2025 ), task-based observational studies (Johnston et al., 2025 ), and a small number of course-embedded interventions (Wang & Ren, 2024 ). The most commonly examined tools were ChatGPT (n = 23), followed by Grammarly (n = 4), Microsoft Copilot (n = 3), Perplexity (n = 2), QuillBot (n = 1), and local or institution-provided LLMs (n = 2). Reported writing applications centered on brainstorming/idea generation and planning/outlining (Nguyen et al., 2024 ; Johnston et al., 2025 ), as well as summarizing literature and drafting/refinement within course tasks (Wang & Ren, 2024 ). Across contexts, authors noted benefits such as efficiency/time savings and greater confidence/self-efficacy (Campbell & Cox, 2024 ; Hysaj et al., 2025 ), along with language support for L2 writers and organizational help (planning/structuring, grammar support) (Johnston et al., 2024 ). Recurrent challenges included risks of plagiarism/contract cheating and policy gaps/uneven AI literacy (Campbell & Cox, 2024 ; Hysaj et al., 2025 ), hallucinations and unverifiable sources (Johnston et al., 2025 ), overreliance that may erode writing skills and limitations of detection tools (Jensen & Jensen, 2025 ), and broader academic-integrity concerns (Nelson et al., 2025 ). 2.6. Data Analysis Procedures Data analysis followed established scoping-review methodology that prioritizes mapping the breadth of evidence and key concepts over effect estimation (Arksey & O’Malley, 2005 ; Levac et al., 2010 ; Peters et al., 2020 ; Tricco et al., 2018 ). Guided by the PCC framework, the researchers conducted two complementary analytic strands and then integrated findings to address the review questions. Using Microsoft Excel, the researchers obtained quantitative data by coding publication year, country and continent, participant group, study design, and GenAI tools. Multi-country items were coded as cross-regional. Tool tallies reflect studies that examined or instructed the use of a tool at least once; because some studies included multiple tools, tool counts can exceed n = 25. Cross-tabulations (for example, tool and participant group; design and continent) were generated to surface distributional patterns. Outputs included summary tables and figures (for example, PRISMA flow; continent map) consistent with PRISMA-ScR reporting (Tricco et al., 2018 ). Qualitative data were extracted during charting text fields (applications in academic writing, benefits and opportunities, challenges and ethical issues, and gaps or recommendations) were analyzed using a deductive–inductive thematic analysis (Braun & Clarke, 2006 ; 2021 ). The researchers began with a deductive coding frame aligned to PCC and the review questions, conducted line-by-line open coding to allow additional categories to emerge, iteratively refined a shared codebook, and developed higher-order themes through constant comparison and memoing. Quantitative and qualitative results were combined using a convergent integrated approach in the Discussion section: numerical patterns (for example, continent or tool frequencies) were juxtaposed with qualitative themes in side-by-side matrices and narrative weaving to produce an evidence map that directly answers each review question (Fetters et al., 2013 ; Peters et al., 2020 ). 3. Results Findings are organized around the four research questions and integrate a descriptive numerical summary with a qualitative thematic synthesis. Drawing on the 25 included studies, the researchers first map how GenAI is being used in academic writing (RQ1), then synthesize reported benefits and opportunities (RQ2), followed by challenges, risks, and ethical issues (RQ3), and finally identify gaps and future research directions (RQ4). This structure aligns with the review’s PCC framing and the stated research questions. 3.1. Ways GenAI is being used in academic writing (RQ1) Across the 25 studies, GenAI is used across the full writing cycle, from pre-writing through finalization (see Table 1 ). Students most commonly employ chatbots to clarify concepts, request definitions and examples, and generate outlines or essay plans, indicating early-stage support for planning and structuring (Johnston et al., 2025 ). GenAI is then used to draft and reorganize text, polish language, and refine coherence, with classroom data showing perceived utility for idea generation, vocabulary support, grammar correction, and argument organization (Wang & Ren, 2024 ). In multilingual contexts, learners report using GenAI to paraphrase, simplify readings, and address language mechanics, describing these uses as practical supports for completing written assessments (Hysaj et al., 2025 ). Source-work and referencing support also appear frequently, although students’ practices are uneven; for example, students plan and search with GenAI but often omit acknowledging the tool itself in references, underscoring an ongoing need for information-literacy guidance (Johnston et al., 2025 ; Jensen & Jensen, 2025 ). ChatGPT is the most frequently used tool, with additional use of Grammarly, Copilot, Perplexity, QuillBot, and institution-provided or local LLMs in several studies (Johnston et al., 2025 ). Table 1 Ways GenAI is used across stages of academic writing (n = 25). Writing stage Typical GenAI uses Representative tools Example sources Pre-writing Clarify concepts, generate ideas/topics, produce examples ChatGPT; local LLMs Johnston et al. ( 2025 ); Johnston et al. ( 2024 ) Planning & outlining Create outlines, reorganize structure, plan sections ChatGPT; Copilot; Perplexity Johnston et al. ( 2025 ); Nguyen et al. ( 2024 ) Drafting Produce first drafts, expand points, suggest wording ChatGPT Wang & Ren ( 2024 ) Revising & refining Rewrite passages, improve coherence/flow, style and tone ChatGPT Wang & Ren ( 2024 ); Hysaj et al. ( 2025 ) Language support (L2) Paraphrase/translate, vocabulary support, grammar/mechanics Grammarly; QuillBot; ChatGPT Hysaj et al. ( 2025 ); Johnston et al. ( 2024 ) Working with sources Generate search terms, summarize articles, format references Perplexity; ChatGPT; Copilot Johnston et al. ( 2025 ); Jensen & Jensen ( 2025 ) Multimodal support Create/plan visuals to accompany text; slide notes ChatGPT (image features) Wang & Ren ( 2024 ) As shown in Table 1 , the studies depict whole-process integration rather than single-point use. Planning and conceptual scaffolding are especially prominent, followed by text production and language polishing. Source-work support is common, but attribution and referencing practices lag; students frequently use GenAI when planning or searching yet omit citing the tool, indicating a gap for AI-informed information-literacy instruction (Johnston et al., 2025 ; Jensen & Jensen, 2025 ). While ChatGPT dominates use, several studies encourage broadening tool awareness to alternatives such as Copilot and Perplexity, and to institution-provided LLMs where available (Johnston et al., 2025 ). 3.2. Benefits and opportunities reported regarding GenAI-assisted writing (RQ2) As summarized in Fig. 3 , the researchers identified six, recurrent benefit clusters that map onto the writing process (planning, drafting, revising, and finalizing). First, students consistently reported efficiency and time savings, with GenAI handling lower-level mechanics (grammar, phrasing, formatting) and thereby freeing attention for higher-order concerns such as argumentation and evidence use (Campbell & Cox, 2024 ; Han, 2025 ; Wang & Ren, 2024 ). Second, GenAI offered substantial language support, particularly in EAP/L2 contexts, where learners used it to paraphrase, translate, expand vocabulary, and improve clarity and fluency; these functions were frequently linked to increased confidence and self-efficacy in completing written assessments (Hysaj et al., 2025 ; Moorhouse et al., 2025 ; Nelson et al., 2025 ). Third, tools were widely used for planning and organization—brainstorming, outlining, and structuring paragraphs—often improving task interpretation and the perceived coherence of drafts (Johnston et al., 2025 ; Mo & Crosthwaite, 2025 ; Wang & Ren, 2024 ). Fourth, several studies highlighted formative feedback and scaffolding: rapid explanations, exemplars, and revision suggestions supported iterative improvement; in some cases, source-display features helped students plan searches and check claims (Jensen & Jensen, 2025 ; Johnston et al., 2025 ). Fifth, the literature points to access and inclusion opportunities. Students with disabilities described GenAI as helpful for planning, drafting, and multimodal expression (e.g., generating alternatives or simplifying language), indicating potential to reduce participation barriers when used with appropriate guidance (Zhao et al., 2025 ). Finally, at the pedagogical and institutional levels, studies framed GenAI as an opportunity space for course/assessment redesign and AI-literacy development—for example, integrating transparent AI use into authentic assessment, and embedding guidance on prompting, verification, and attribution (Cordero et al., 2025 ; Johnston et al., 2024 ; Kofinas et al., 2025 ). Taken together, the pattern across contexts suggests GenAI’s strongest contributions cluster in the planning-to-polishing span of the writing cycle. Benefits are maximized when use is transparent, scaffolded, and paired with verification practices, positioning GenAI as an assistive resource rather than a substitute for disciplinary thinking and academic integrity (Jensen & Jensen, 2025 ; Johnston et al., 2025 ). 3.3. Challenges, risks, and ethical issues identified in the literature (RQ3) Synthesizing the 25 studies, the researchers identified five recurrent risk domains that cut across the writing process (Fig. 4 ). First, source reliability and epistemic risk remain prominent. Studies document hallucinations, unverifiable claims, and fabricated citations that can be difficult for novice writers to detect, underscoring the need for systematic verification and triangulation when GenAI is used for content generation or source work (Johnston et al., 2025 ; Jensen & Jensen, 2025 ; Zizka, 2025 ). Second, issues of authorship, attribution, and academic integrity are widely reported. Surveys and classroom investigations note inconsistent acknowledgment of tool use, uncertainty about the boundary between acceptable support and misconduct, and risks of plagiarism or contract cheating when outputs are submitted with minimal transformation (Johnston et al., 2024 ; Kofinas et al., 2025 ; Nelson et al., 2025 ; Johnston et al., 2025 ). Third, the literature points to overreliance and skill development concerns. Without scaffolding, students may defer critical reading, argumentation, and revision to GenAI, exhibiting automation bias and reduced practice with higher-order writing skills; several papers caution that “hands-off” use can erode competence over time (Jensen & Jensen, 2025 ; Han, 2025 ; Watermeyer et al., 2024 ). Fourth, assessment alignment and detection limits pose persistent challenges. Tasks that can be solved by generic prompts invite superficial engagement, while AI-detection tools are unreliable, susceptible to manipulation, and not suitable as sole evidence in integrity processes. The literature instead recommends process-focused assessment, multi-source evidence in investigations, and explicit expectations about permissible GenAI support (Kofinas et al., 2025 ; van Niekerk et al., 2025 ; Jensen & Jensen, 2025 ). Fifth, studies surface governance, equity, and privacy issues. Uneven AI literacy and unclear policies produce inconsistent practice across courses; students also express concerns about sharing assignments with third-party systems, potential data exposure, and unequal access to paid tools and connectivity that can amplify existing inequities (Campbell & Cox, 2024 ; Johnston et al., 2024 ; Rodafinos, 2025 ; Zhao et al., 2025 ; Stanford, 2025 ). The synthesized studies suggest that risk is highest where GenAI intersects with source credibility, authorship norms, and assessment design. Consistent recommendations across studies include transparent policies with required disclosure of tool use, explicit training in verification and source work, redesign of assessments to emphasize process and originality (e.g., staged drafts and checkpointing), avoidance of detectors as the sole basis for misconduct decisions, and privacy-safe, equitable access practices (Kofinas et al., 2025 ; Johnston et al., 2024 ; van Niekerk et al., 2025 ). 3.4. Gaps and future research directions highlighted by synthesized studies (RQ4) The researchers distilled seven cross-cutting gaps and aligned each with suggested methods and target outcomes (see Table 2 ). Collectively, the evidence base remains largely descriptive; stronger causal designs are needed to test whether GenAI improves writing quality, learning, or transfer across time and genres. Priority studies include longitudinal cohorts and classroom experiments with transparent comparison conditions and validated rubrics (Kofinas et al., 2025 ; Wang & Ren, 2024 ; Kim et al., 2025 ; Moorhouse et al., 2025 ; Zizka, 2025 ). Table 2 Summary of Research Gaps, Suggested Methods, and Target Outcomes in GenAI-Assisted Academic Writing (n = 25). Research gaps (what needs evidence) Suggested methods (how to study it) Target outcomes to report (what to measure) Representative studies 1. Causal effects on writing quality, learning, and transfer Longitudinal cohorts; classroom experiments or quasi-experiments with transparent comparison conditions; multi-site replications Writing quality using validated rubrics; learning gains and transfer to new genres; persistence over time; time-on-task and revision productivity Kofinas et al., 2025 ; Wang & Ren, 2024 ; Kim et al., 2025 ; Moorhouse et al., 2025 ; Zizka, 2025 2. Reporting and attribution practices Field experiments embedding disclosure requirements; audit studies of assignments; mixed-methods studies of student/marker perceptions Disclosure rates; accuracy/completeness of AI-use statements; impacts on grades and feedback; perceived fairness/integrity Johnston et al., 2025 ; Jensen & Jensen, 2025 ; van Niekerk et al., 2025 ; Johnston et al., 2024 3. AI-literacy and pedagogy interventions Design-based research (DBR) on curricula; randomized or quasi-experimental evaluation of modules on prompting, verification, and source work AI-literacy competency gains; verification accuracy; quality of source work; metacognitive strategy use; student confidence/self-efficacy Campbell & Cox, 2024 ; Mo & Crosthwaite, 2025 ; Hysaj et al., 2025 ; Nguyen et al., 2024 4. Assessment and policy design Comparative studies of assessment formats (staged drafts, in-class writing, viva/checkpoints); policy implementation evaluations; process tracing of drafting workflows Misconduct allegations and outcomes; detector false-positive/negative rates; policy compliance/fidelity; marker workload; student satisfaction and perceived fairness Kofinas et al., 2025 ; van Niekerk et al., 2025 ; Johnston et al., 2024 5. Equity, accessibility, and privacy Studies with disability subgroups; usability testing; surveys/interviews on access; privacy impact assessments Accessibility gains; accommodation effectiveness; access gaps (device, bandwidth, paid tools); privacy incidents and data sharing; differential outcomes by subgroup Zhao et al., 2025 ; Watermeyer et al., 2024 ; Rodafinos, 2025 ; Stanford, 2025 6. Multilingual and cross-cultural contexts Cross-site studies in non-Anglophone HEIs; L1/L2 comparisons; corpus-informed analyses of genre and register Language quality (clarity, cohesion, accuracy) in L2 writing; translation/paraphrase fidelity; genre conformity; cultural/disciplinary fit Nelson et al., 2025 ; Adalı & Bilgili, 2025; Hysaj et al., 2025 7. Measurement standards and replication Consensus methods (e.g., Delphi) to define taxonomies; preregistered protocols; shared prompts and model versions; open materials for replication Reporting-checklist compliance; reproducibility of results; sensitivity to model/version/prompt; open datasets and code availability Mo & Crosthwaite, 2025 ; Han, 2025 A second cluster concerns reporting and attribution. Multiple papers document inconsistent disclosure of GenAI use and uncertainty about acknowledgment norms, especially when tools assist with planning or paraphrase. Table 3 therefore recommends field experiments that embed disclosure requirements and mixed-methods audits of student and marker perceptions, with outcomes such as disclosure rates, accuracy of AI-use statements, and effects on grading and feedback (Johnston et al., 2025 ; Jensen & Jensen, 2025 ; van Niekerk et al., 2025 ; Johnston et al., 2024 ). Third, the literature calls for systematic evaluation of AI-literacy and pedagogy. While many studies advocate instruction in prompting, verification, and source work, few evaluate structured curricula at scale. Design-based research and quasi-experimental modules should report competency gains, verification accuracy, quality of source work, and metacognitive strategy use (Campbell & Cox, 2024 ; Mo & Crosthwaite, 2025 ; Hysaj et al., 2025 ; Nguyen et al., 2024 ). Fourth, assessment and policy design require rigorous testing. Recommended directions include comparative evaluations of process-oriented formats (e.g., staged drafts, in-class writing, oral checkpoints), policy implementation studies, and process tracing of drafting workflows; outcomes should include integrity incidents, false-positive rates from detectors, policy fidelity, marker workload, and perceived fairness (Kofinas et al., 2025 ; van Niekerk et al., 2025 ; Johnston et al., 2024 ). Fifth, equity, accessibility, and privacy remain under-researched. Early findings suggest potential to reduce participation barriers for students with disabilities but also raise concerns about unequal access to paid tools and data-sharing risks. Future work should incorporate subgroup analyses, usability testing, and privacy-impact assessments with clear reporting of accessibility gains and differential outcomes (Zhao et al., 2025 ; Watermeyer et al., 2024 ; Rodafinos, 2025 ; Stanford, 2025 ). Sixth, the field needs broader coverage of multilingual and cross-cultural contexts. Studies should move beyond Anglophone settings to examine L1/L2 differences, translation/paraphrase fidelity, and genre conformity in non-English HEIs (Nelson et al., 2025 ; Karahan Adalı & Bilgili, 2025 ; Hysaj et al., 2025 ). Finally, measurement standards and replication would improve comparability. The researchers recommend consensus on use-case taxonomies (planning, drafting, revising), preregistered protocols, and sharing of prompts, model versions, datasets, and code; core outcomes should include rubric-based writing quality, learning/transfer, integrity outcomes, and sensitivity to model/version changes (Mo & Crosthwaite, 2025 ; Han, 2025 ). 4. Discussion The researchers interpreted the map of evidence through the PCC lens using convergent integration, aligning numerical distributions with qualitative themes to explain what the patterns mean for academic writing in higher education rather than restating procedures or counts (Fetters et al., 2013 ; Tricco et al., 2018 ). As depicted in Fig. 5 , a PCC-aligned convergent model guides this interpretation: Population, Concept, and Context flow into a convergent-integration step (shown as a dashed band) that positions GenAI as assistive scaffolding while core scholarly practices remain grounded in human judgment. This process emphasis aligns with evidence that explicit argument scaffolds measurably improve conceptual understanding across levels and delivery modes, indicating that GenAI should be paired with argument-based routines rather than replace them (Ramallosa et al., 2022 ). With respect to RQ1, the researchers found convergence between tool prevalence and narratives of practice: widespread reliance on general-purpose chatbots alongside grammar and paraphrase tools corresponds to reported gains in clarity, organization, and confidence, especially among multilingual writers. Yet, qualitative accounts also underline that idea development, disciplinary reasoning, and source-critical reading are not reliably automated, which situates GenAI as a facilitator of process rather than a substitute for scholarly authorship (Jensen & Jensen, 2025 ; Johnston et al., 2025 ). This emphasis on process support is represented on the left of Fig. 5 as “Assistive scaffolding (Planning to Revision).” For RQ2 and RQ3, the researchers observed a stable pattern of benefits—efficiency, fluency, and reduced language barriers—counterbalanced by recurrent risks, notably hallucinations, fragile or fabricated citations, uneven disclosure, and signs of overreliance where scaffolding is absent. Divergence between stakeholder perspectives is salient: students often report improved fluency and task confidence, whereas instructors more frequently raise concerns about originality, source quality, and process visibility. The synthesis therefore supports the view that GenAI should augment, not replace, disciplinary thinking and source work, and that verification practices must be made explicit within assignments and feedback cycles (Campbell & Cox, 2024 ; Johnston et al., 2024 ; Zizka, 2025 ). The central “Human judgment” node in Fig. 5 visually anchors this requirement for verification and attribution/disclosure. Positioning GenAI as a tool that supports claim–evidence–reasoning and structured rebuttal is consistent with meta-analytic gains from argument-based learning (Ramallosa et al., 2022 ), while keeping human verification central. Interpreting the evidence through PCC also highlights where convergence and divergence matter. On population, the student-heavy corpus limits insight into supervisory practices, disclosure norms, and assessment decisions among faculty and postgraduate researchers. The researchers therefore note the need for more staff-focused and mixed-cohort studies to illuminate how expectations are translated into grading and feedback in authentic settings (Watermeyer et al., 2024 ; Cordero et al., 2025 ). On concept, whole-process assistance rather than single-step substitution explains the robust improvements in organization and language contrasted with mixed results for higher-order reasoning (Jensen & Jensen, 2025 ). On context, institutions with educative, transparent policies report more constructive uses—such as declared assistance with verification—while resource-constrained settings surface equity, access, and privacy concerns more sharply, suggesting that policy effectiveness is contingent on local conditions and support (Moorhouse et al., 2025 ; Zhao et al., 2025 ). Findings from secondary science during distance education show that learner-centered, action-oriented, and transformative practices emerged despite constraints, but inadequate equipment and poor connectivity were persistent barriers (Funa et al., 2023 ); these realities should shape GenAI policy to avoid deepening access gaps. These PCC elements are shown at the top of Fig. 5 , with directional arrows into the convergent-integration band. These patterns help resolve the assistive–substitution debate. Where courses embed GenAI as taught scaffolding within a staged writing process, the researchers observed improvements without systematic loss of authorial voice. Where use is broad and unscaffolded, automation bias, shallow source engagement, and dependence are more likely. The literature converges on three guardrails: verifiability of content and references, transparent attribution or disclosure, and process-oriented assessment that makes thinking visible; by contrast, sole reliance on AI-detection tools is widely considered insufficient evidence for adjudicating integrity (Han, 2025 ; Kofinas et al., 2025 ; van Niekerk et al., 2025 ). Converging with these findings, a recent meta-analysis of inquiry-based learning likewise reports substantial improvements in students’ conceptual understanding, particularly under open-inquiry conditions (Mediana Jr. et al., 2025 ); by analogy, GenAI should be embedded as a scaffold that sustains inquiry and verification rather than as an autonomous text generator. The bidirectional arrow between the two teal panels in Fig. 5 signals interaction—GenAI can support process without supplanting core scholarly practices. The researchers’ implications also resonate with the I-STEM-PBL-ESD instructional framework, which integrates problem-based learning with education for sustainable development; GenAI tasks can be embedded as supports for problem framing, scenario exploration, and reflective synthesis while safeguarding attribution and process visibility (Funa et al., 2024 ). Implications follow directly for curriculum and policy. The researchers interpret the evidence as supporting AI-literacy that teaches prompting for thinking, verification strategies, and explicit attribution norms; assessment designs that document process via staged drafts, in-class checkpoints, and brief oral or written justifications; and equity-minded implementation that addresses differential access, accessibility needs, and data-privacy risks. These directions align with emerging institutional scholarship that emphasizes responsible, transparent adoption calibrated to local contexts (Funa & Gabay, 2025a , 2025b ; Cordero et al., 2025 ). Positioning this review within prior syntheses, the researchers corroborate early findings of clarity and fluency gains alongside integrity and equity concerns, while extending the field by focusing on 2024 to Q2 2025 studies and by integrating numerical distributions with qualitative themes. The added value lies in specifying where benefits reliably cluster (planning-to-polishing stages), which risks remain unresolved (verification and attribution), and which methodological moves are now necessary to advance knowledge—namely causal and longitudinal designs testing learning and transfer, systematic studies of disclosure and authorship practices, cross-cultural comparisons beyond Anglophone/EMI contexts, and clearer reporting standards for AI involvement (Chanpradit, 2025 ; Mo & Crosthwaite, 2025 ; Tricco et al., 2018 ). Consistent with Fig. 5 , the implications panel synthesizes these priorities: educative policy; verification and attribution/disclosure; process-oriented assessment; equity and privacy; and a forward research agenda. Overall, the researchers interpret GenAI as most effective when treated as an assistive, transparent scaffold embedded in educative policy and process-forward pedagogy. Under these conditions, gains in organization, clarity, and productivity are achievable without compromising authorial ownership; under substitution framings, the risks noted above intensify. The field therefore benefits from designs and policies that preserve human judgment while leveraging GenAI’s strengths in planning, feedback, and language support (Jensen & Jensen, 2025 ; Johnston et al., 2025 ; Zhao et al., 2025 ). 5. Limitations of the Study The researchers conducted a scoping review to map GenAI use in higher-education writing, not to judge study quality or estimate effects. Inclusion filters (peer-reviewed, English, higher education, 2024–Q2 2025) may introduce language/publication bias and exclude preprints and conference work. Rapidly evolving terminology and tool stacks mean some relevant studies may have been missed, and conclusions rest on primary studies that often use small, single-site samples and proxy outcomes. The evidence base is skewed toward better-resourced, English-medium institutions with limited faculty/postgraduate and Global South representation. Finally, results may shift as GenAI capabilities and institutional policies change, and theme coding involved researcher judgment. 6. Conclusions and Recommendations The researchers conclude that GenAI currently functions most productively as assistive scaffolding across the planning-to-revision span of academic writing in higher education. When paired with explicit expectations for verification, attribution or disclosure, and visible processes, GenAI can improve organization, clarity, and productivity without displacing core scholarly practices such as argument construction, source interrogation, and synthesis. Divergences in outcomes appear when use is unscaffolded, when originality is treated as a product metric rather than a process, and when institutions rely on detection tools as sole evidence of integrity. The present map also shows important evidence gaps, including limited representation of faculty and postgraduate cohorts, under-representation of the Global South, and an over-reliance on short-term proxy measures of learning. Taken together, these patterns support an assistive-not-substitutive stance that centers human judgment while leveraging GenAI for formative support. The researchers recommend an institution-wide approach that treats GenAI as assistive scaffolding within clearly articulated policies and taught practices. Specifically, universities should implement AI-literacy programs that teach prompting for thinking, verification strategies, and explicit attribution or disclosure; course policies should define permitted uses, privacy expectations, and data handling; and faculty development should emphasize task designs that keep human reasoning visible and feedback that targets argument quality and source use. Assessment should be process-centered—requiring staged drafts, in-class checkpoints, and brief oral or written justifications—to make authorship and decision pathways transparent; institutions should avoid sole reliance on AI-detection tools and instead use multi-source evidence that includes process artifacts and instructor judgment. To promote equity and access, the researchers advise providing institutionally vetted tools, accessibility features, and low-bandwidth or offline pathways so that GenAI support does not widen existing gaps. In course design, GenAI should scaffold question generation, planning, formative feedback, and reflective synthesis while reserving claim-evidence-reasoning, source evaluation, and final interpretations for learners. Finally, the research agenda should prioritize causal and longitudinal studies of learning and transfer, broaden representation to include faculty and postgraduate cohorts and non-Anglophone or resource-constrained contexts, and improve reporting standards by documenting model versions, prompts, guardrails, and disclosure practices while employing outcomes that capture reasoning, ethics, and sustained learning rather than only surface features. References Acut, D. P., Malabago, N. K., Malicoban, E. V., Galamiton, N. S., & Garcia, M. B. (2024). “ChatGPT 4.0 Ghosted Us While Conducting Literature Search:” Modeling the Chatbot’s Generated Non-Existent References Using Regression Analysis. Internet Reference Services Quarterly , 29 (1), 27–54. https://doi.org/10.1080/10875301.2024.2426793 Arksey, H., & O’Malley, L. (2005). Scoping studies: Towards a methodological framework. International Journal of Social Research Methodology, 8 (1), 19–32. https://doi.org/10.1080/1364557032000119616 Bisi, T., Risser, A., Clavert, P., Migaud, H., & Dartus, J. (2023). What is the rate of text generated by artificial intelligence over a year of publication in Orthopedics & Traumatology: Surgery & Research? Analysis of 425 articles before versus after the launch of ChatGPT in November 2022. Orthopaedics & Traumatology: Surgery & Research , 109 (8), 103694. Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., Brynjolfsson, E., Buch, V., Card, D., Castellon, D., Chen, A., Creel, K. A., Davis, J. Q., Demszky, D., Donahue, C., Doumbouya, M., … Liang, P. (2021). On the opportunities and risks of foundation models. arXiv. https://doi.org/10.48550/arXiv.2108.07258 Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3 (2), 77–101. https://doi.org/10.1191/1478088706qp063oa Braun, V., & Clarke, V. (2021). Thematic analysis: A practical guide . SAGE. Campbell, L. O., & Cox, T. D. (2024). Utilizing generative AI in higher education teaching and learning. Journal of the Scholarship of Teaching and Learning, 24 (4), 162–173. https://doi.org/10.14434/josotl.v24i4.36575 Chanpradit, T. (2025). Generative artificial intelligence in academic writing in higher education: A systematic review. Edelweiss Applied Science and Technology, 9 (4), 889–906. https://doi.org/10.55214/25768484.v9i4.6128 Cordero, J., Torres-Zambrano, J., & Cordero-Castillo, A. (2025). Integration of generative artificial intelligence in higher education: Best practices. Education Sciences, 15 (1), 32. https://doi.org/10.3390/educsci15010032 Emran, A. Q., Mohammed, M. N., Saeed, H., Keir, M. Y. A., Alani, Z. N., & Ibrahim, F. M. (2024, January). Paraphrasing ChatGPT answers as a tool to enhance university students’ academic writing skills. In 2024 ASU International Conference in Emerging Technologies for Sustainability and Intelligent Systems (ICETSIS) (pp. 501-505). IEEE. Fetters, M. D., Curry, L. A., & Creswell, J. W. (2013). Achieving integration in mixed methods designs: Principles and practices. Health Services Research, 48 (6 Pt 2), 2134–2156. https://doi.org/10.1111/1475-6773.12117 Flower, L., & Hayes, J. R. (1981). A cognitive process theory of writing. College Composition and Communication, 32 (4), 365–387. https://doi.org/10.2307/356600 Funa, A. A., & Gabay, R. A. E. (2025a). Exploring perspectives toward artificial intelligence integration in science education: A cross-generational study [Preprint]. SSRN. https://doi.org/10.2139/ssrn.5240513 Funa, A. A., & Gabay, R. A. E. (2025b). Policy guidelines and recommendations on AI use in teaching and learning: A meta-synthesis study. Social Sciences & Humanities Open , 11 , 101221. Funa, A. A., Gabay, R. A. E., Esdicul, K. L., & Prudente, M. S. (2023). Secondary teachers’ and students’ perceptions of distance education in science: Focus on learner-centered, action-oriented, and transformative learning. Dalat University Journal of Science, 13 (3), 156–181. https://doi.org/10.37569/DalatUniversity.13.3.1108(2023) Funa, A. A., Ricafort, J. D., Jetomo, F. G. J., & Lasala Jr, N. L. (2024). Effectiveness of brain-based learning toward improving students’ conceptual understanding: a meta-analysis. International Journal of Instruction , 17 (1), 361-380. Funa, A. A., Roleda, L. S., & Prudente, M. S. (2024). Integrated Science, Technology, Engineering, and Mathematics—Problem-Based Learning—Education for Sustainable Development (I-STEM-PBL-ESD) framework. In D. Luberto, G. Sharma, N. Capraro, & R. Yadav (Eds.), A Diversity of Pathways Through Science Education . Springer. https://doi.org/10.1007/978-981-97-2607-3_9 Goodfellow, I. J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., ... & Bengio, Y. (2014). Generative adversarial nets. Advances in neural information processing systems , 27 . Han, Y. (2025). Beyond the algorithm: Reconciling generative AI and human agency in academic writing education. International Journal of Learning and Teaching, 11 (1), 39–42. https://doi.org/10.18178/ijlt.11.1.39-42 Harzing, A.W. (2007) Publish or Perish, available from https://harzing.com/resources/publish-or-perish Huh, S. (2023). Emergence of the metaverse and ChatGPT in journal publishing after the COVID-19 pandemic. Science Editing , 10 (1), 1-4. Hysaj, A., Dean, B. A., & Freeman, M. (2025). Exploring the purposes and uses of generative artificial intelligence tools in academic writing for multicultural students. Higher Education Research & Development . Advance online publication. https://doi.org/10.1080/07294360.2025.2488862 Jensen, T. W., & Jensen, S. W. (2025). AI literacy in the context of working with sources: Pitfalls and possibilities of generative AI models in academic writing. Journal of Academic Writing, 15 (S2), 1–12. https://doi.org/10.18552/joaw.v15iS2.1224 Johnston, H., Eaton, M., Henry, I., Deeley, E.-M., & Parsons, B. N. (2025). Discovering how students use generative artificial intelligence tools for academic writing purposes. Journal of Learning Development in Higher Education , (34). Johnston, H., Wells, R. F., Shanks, E. M., Boey, T., & Parsons, B. N. (2024). Student perspectives on the use of generative artificial intelligence technologies in higher education. International Journal for Educational Integrity, 20 , Article 2. https://doi.org/10.1007/s40979-024-00149-4 Karahan Adalı, G., & Bilgili, A. (2025). Generative AI in higher education: Students’ perspectives on adoption, ethical concerns, and academic impact. Acta Infologica, 9 (1), 147–166. https://doi.org/10.26650/acin.1670197 Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., Peters, J., & Kasneci, G. (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 Khalifa, M., & Albadawy, M. (2024). Using artificial intelligence in academic writing and research: An essential productivity tool. Computer Methods and Programs in Biomedicine Update , 5 , 100145. Kim, J., Lee, S.-S., Detrick, R., Wang, J., & Li, N. (2025). Students–generative AI interaction patterns and its impact on academic writing. Journal of Computing in Higher Education . Advance online publication. https://doi.org/10.1007/s12528-025-09444-6 Kim, J., Yu, S., Detrick, R., & Li, N. (2025). Exploring students’ perspectives on generative AI-assisted academic writing. Education and Information Technologies, 30 , 1265–1300. https://doi.org/10.1007/s10639-024-12878-7 Kofinas, A. K., Tsay, C.-H., & Pike, D. (2025). The impact of generative AI on academic integrity of authentic assessments within a higher education context. British Journal of Educational Technology, 00 , 1–28. https://doi.org/10.1111/bjet.13585 Levac, D., Colquhoun, H., & O’Brien, K. K. (2010). Scoping studies: Advancing the methodology. Implementation Science, 5 (69). https://doi.org/10.1186/1748-5908-5-69 Maphoto, K. B., Sevnarayan, K., Mohale, N. E., Suliman, Z., Ntsopi, T. J., & Mokoena, D. (2024). Advancing students’ academic excellence in distance education: Exploring the potential of generative AI integration to improve academic writing skills. Open Praxis, 16 (2), 142–159. https://doi.org/10.55982/openpraxis.16.2.649 Mediana Jr., N. L., Funa, A. A., & Dio, R. V. (2025). Effectiveness of inquiry-based learning (IbL) on improving students’ conceptual understanding in science and mathematics: A meta-analysis. International Journal of Education in Mathematics, Science and Technology, 13 (2), 532–552. https://doi.org/10.46328/ijemst.4769 Meyer, J. G., Urbanowicz, R. J., Martin, P. C., O’Connor, K., Li, R., Peng, P. C., ... & Moore, J. H. (2023). ChatGPT and large language models in academia: opportunities and challenges. BioData mining , 16 (1), 20. Mo, Z., & Crosthwaite, P. (2025). Exploring the affordances of generative AI large language models for stance and engagement in academic writing. Journal of English for Academic Purposes, 75 , Article 101499. https://doi.org/10.1016/j.jeap.2025.101499 Moorhouse, B. L., Wan, Y., Wu, C., Wu, M., & Ho, T. Y. (2025). Generative AI tools and empowerment in L2 academic writing. System, 133 , Article 103779. https://doi.org/10.1016/j.system.2025.103779 Nelson, A. S., Santamaría, P. V., Javens, J. S., & Ricaurte, M. (2025). Students’ perceptions of generative artificial intelligence (GenAI) use in academic writing in English as a foreign language. Education Sciences, 15 (5), 611. https://doi.org/10.3390/educsci15050611 Nguyen, A., Hong, Y., Dang, B., & Huang, X. (2024). Human–AI collaboration patterns in AI-assisted academic writing. Studies in Higher Education, 49 (5), 847–864. https://doi.org/10.1080/03075079.2024.2323593 OpenAI. (2022, November 30). Introducing ChatGPT. https://openai.com/blog/chatgpt Peters, M. D. J., Marnie, C., Tricco, A. C., Pollock, D., Munn, Z., Alexander, L., McInerney, P., Godfrey, C. M., & Khalil, H. (2020). Updated methodological guidance for the conduct of scoping reviews. JBI Evidence Synthesis, 18 (10), 2119–2126. https://doi.org/10.11124/JBIES-20-00167 Ramallosa, M., Funa, A. A., Geron, C. A., Ibardaloza, R. P., & Prudente, M. S. (2022). Meta-analysis on the effectiveness of argument-based learning on students’ conceptual understanding. In Proceedings of the 2022 13th International Conference on E-Education, E-Business, E-Management, and E-Learning (IC4E ’22) (pp. 128–133). ACM. https://doi.org/10.1145/3502485.3502495 Rodafinos, A. (2025). The integration of generative AI tools in academic writing: Implications for student research. Social Education Research, 6 (2). https://doi.org/10.37256/ser.6220256517 Stanford, J. (2025, March 17). How disruptive is AI in students’ academic writing in higher education? Studies in Technology Enhanced Learning, 4 (1). https://doi.org/10.21428/8c225f6e.0390b853 Tricco, A. C., Lillie, E., Zarin, W., O’Brien, K. K., Colquhoun, H., Levac, D., Moher, D., Peters, M. D. J., Horsley, T., Weeks, L., Hempel, S., Akl, E. A., Chang, C., McGowan, J., Stewart, L., Hartling, L., Aldcroft, A., Wilson, M. G., Garritty, C., … Straus, S. E. (2018). PRISMA extension for scoping reviews (PRISMA‑ScR): Checklist and explanation. Annals of Internal Medicine, 169 (7), 467–473. https://doi.org/10.7326/M18-0850 van Niekerk, J., Delport, P. M. J., & Sutherland, I. (2025). Addressing the use of generative AI in academic writing. Computers and Education: Artificial Intelligence, 8 , Article 100342. https://doi.org/10.1016/j.caeai.2024.100342 Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes. Harvard University Press. Wang, L., & Ren, B. (2024). Enhancing academic writing in a linguistics course with generative AI: An empirical study in a higher education institution in Hong Kong. Education Sciences, 14 , 1329. https://doi.org/10.3390/educsci14121329 Watermeyer, R., Phipps, L., Lanclos, D., & Knight, C. (2024). Generative AI and the automating of academia. Postdigital Science and Education, 6 , 446–466. https://doi.org/10.1007/s42438-023-00440-6 Yusuf, A., Pervin, N., & Román-González, M. (2024). Generative AI and the future of higher education: A threat to academic integrity or reformation? Evidence from multicultural perspectives. International Journal of Educational Technology in Higher Education, 21 , Article 21. https://doi.org/10.1186/s41239-024-00453-6 Zamorano, C. (2025). Enhancing academic writing in English language education through generative AI integration. Research Studies in English Language Teaching and Learning, 3 (3), 424–447. https://doi.org/10.62583/rseltl.v3i3.87 Zhang, J. S., Yoon, C., Williams, D. K. A., & Pinkas, A. (2024). Exploring the usage of ChatGPT among medical students in the United States. Journal of Medical Education and Curricular Development, 11, 1–7. https://doi.org/10.1177/23821205241264695 Zhao, X., Cox, A., & Chen, X. (2025). The use of generative AI by students with disabilities in higher education. The Internet and Higher Education, 66 , Article 101014. https://doi.org/10.1016/j.iheduc.2025.101014 Zizka, L. (2025). “It looks good enough”: Recognizing the quality of generative AI output in academic writing tasks in higher education. Journal of Hospitality & Tourism Education . Advance online publication. https://doi.org/10.1080/10963758.2025.2496663 Additional Declarations The authors declare no competing interests. Supplementary Files AppendixA.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7440784","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Systematic Review","associatedPublications":[],"authors":[{"id":504552432,"identity":"bb5d835f-3fb3-46bf-8943-20e12502482a","order_by":0,"name":"Renz Alvin E. Gabay","email":"","orcid":"https://orcid.org/0000-0002-8108-4589","institution":"Sorsogon State University","correspondingAuthor":false,"prefix":"","firstName":"Renz","middleName":"Alvin E.","lastName":"Gabay","suffix":""},{"id":504552433,"identity":"652b00ef-caff-4721-8267-36176a694758","order_by":1,"name":"Aaron A. Funa","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8UlEQVRIiWNgGAWjYBAC9gY4k/kAjGUAxAcwlMIAD0KKLYFkLTwGRGqRyD34gaHGLp9fuufzh4976hIb2Ju3STDuuINHS16yBMOxZMuZc85uk5zx7HBiA8+xMgnGM89warGXyDGQYGBjNjC4kbuNmefAgcQGiRwzCca2w3hsyTH+wfCvHqgl5/FnngNAh8m/IagFrACkhUGa5wAz0BYeAlp43phZJPYdN5CckWYmOePAYeM2nrRii8QzeLSw5xjf+PCt2oBfIvnxhw8H6mT72Q9vvPFxB24tYJCAzGEDEYkN+HVgAYykaxkFo2AUjILhCwD2CFMNoBK4lwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0002-6648-8825","institution":"Sorsogon State University","correspondingAuthor":true,"prefix":"","firstName":"Aaron","middleName":"A.","lastName":"Funa","suffix":""},{"id":504552434,"identity":"4a70d7ce-ce2b-46bc-90d7-775cf96ee213","order_by":2,"name":"Jhonner D. Ricafort","email":"","orcid":"https://orcid.org/0000-0001-8980-6681","institution":"Sorsogon State University","correspondingAuthor":false,"prefix":"","firstName":"Jhonner","middleName":"D.","lastName":"Ricafort","suffix":""}],"badges":[],"createdAt":"2025-08-23 10:58:43","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":true,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":true},"doi":"10.21203/rs.3.rs-7440784/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7440784/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":89880547,"identity":"3be70008-cbb1-4c11-ba81-a81c0f9a61e7","added_by":"auto","created_at":"2025-08-26 05:46:50","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":159064,"visible":true,"origin":"","legend":"\u003cp\u003ePRISMA study selection flow diagram for GenAI in academic writing.\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-7440784/v1/d24041e65157c1ee4c41bf0d.png"},{"id":89881349,"identity":"789a62a8-1e0f-4eba-9541-77f56edcbcae","added_by":"auto","created_at":"2025-08-26 05:54:50","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":115320,"visible":true,"origin":"","legend":"\u003cp\u003eGenAI and academic writing studies by continent (2024–Q2 2025, n = 25). \u003cem\u003eNote\u003c/em\u003e: Türkiye counted under Asia.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7440784/v1/b5c627e1f2dd7b4494b1050c.png"},{"id":89883092,"identity":"98d1ac3b-66de-484e-bfb4-fb23e2898f22","added_by":"auto","created_at":"2025-08-26 06:02:50","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":114548,"visible":true,"origin":"","legend":"\u003cp\u003eBenefits and opportunities of GenAI-assisted academic writing (2024–Q2 2025; n = 25). \u003cem\u003eNote\u003c/em\u003e. Themes synthesized from the included studies and mapped to writing stages.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-7440784/v1/3bb03ed5f83da939397f4429.png"},{"id":89880552,"identity":"10da689e-5b22-4bd5-a7ba-fac53181c194","added_by":"auto","created_at":"2025-08-26 05:46:50","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":207070,"visible":true,"origin":"","legend":"\u003cp\u003eBow-tie map of challenges, risks, and ethical issues in GenAI-assisted academic writing (2024–Q2 2025; n = 25). \u003cem\u003eNote.\u003c/em\u003e Items synthesize findings across the included studies; arrows indicate progression from enabling conditions to risk domains and consequences. Controls/mitigations summarize commonly recommended responses.\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-7440784/v1/3888f36abe9768e644895cd0.png"},{"id":89880548,"identity":"fca42736-ef04-4bca-9013-0a85f716fdc0","added_by":"auto","created_at":"2025-08-26 05:46:50","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":127852,"visible":true,"origin":"","legend":"\u003cp\u003ePCC-aligned convergent integration model for GenAI in academic writing.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-7440784/v1/11621323abb9408fcceb2387.png"},{"id":89884225,"identity":"5656ba6d-93ff-4dc6-ae86-d406904d96cd","added_by":"auto","created_at":"2025-08-26 06:10:51","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1827713,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7440784/v1/5ff13ec8-30a2-46af-a330-61ae7d233f92.pdf"},{"id":89880546,"identity":"9175810e-d46d-4946-a136-3613eb91ce27","added_by":"auto","created_at":"2025-08-26 05:46:50","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":36522,"visible":true,"origin":"","legend":"","description":"","filename":"AppendixA.docx","url":"https://assets-eu.researchsquare.com/files/rs-7440784/v1/91f5a675b8e85548db28978b.docx"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eGenerative Artificial Intelligence (GenAI) for Academic Writing in Higher Education: A Scoping Review of Applications, Challenges, and Implications\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eGenerative artificial intelligence (GenAI; referred to as \u0026ldquo;GAI\u0026rdquo; in some studies) is transforming how people approach writing, communication, and knowledge creation. In the post-pandemic period, as remote learning and digital platforms became integral to education, tools such as ChatGPT, Gemini, and other large language models (LLMs) entered mainstream academic spaces and are now used for brainstorming, paraphrasing, and even drafting research papers (Emran et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Meyer et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Scholars, educators, and students increasingly recognize that these technologies can enhance productivity and creativity in academic writing (Funa \u0026amp; Gabay, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e; Khalifa \u0026amp; Albadawy, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). At the same time, their implications for academic integrity, authorship, and writing practice remain contested and not yet fully understood (Acut et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Funa \u0026amp; Gabay, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2025b\u003c/span\u003e). This review examines the emerging literature on GenAI\u0026rsquo;s role in academic writing, mapping current evidence, identifying gaps, and clarifying ongoing debates through a comprehensive scoping review. Understanding GenAI\u0026rsquo;s influence is crucial for shaping future academic policies and pedagogies.\u003c/p\u003e\u003cp\u003eGenAI refers to AI systems designed to produce new content\u0026mdash;text, images, or code\u0026mdash;rather than merely analyze existing data. Its conceptual foundation can be traced to research on generative models, particularly the introduction of generative adversarial networks (GANs) by Goodfellow et al. (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), which showed that AI could learn patterns in data and create novel outputs. Contemporary tools such as ChatGPT and Gemini are built on what the Stanford Center for Research on Foundation Models (CRFM) terms foundation models: large-scale, self-supervised deep learning systems trained on vast datasets and adaptable to many tasks. These models provide the underlying architectures that power modern GenAI applications and enable language, image, and code generation at scale (Bommasani et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eIn academic settings, GenAI tools are widely used for idea generation, paraphrasing, literature summarization, and drafting, reshaping how students and scholars approach writing (Acut et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Emran et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Funa \u0026amp; Gabay, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e; Kasneci et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Meyer et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). A recent survey of medical students in the United States found that 48.9% had used ChatGPT in their studies; among users, 43.7% reported weekly to daily use, most commonly for writing, revising, editing, and summarizing. Notably, 37.5% and 41.3% reported using ChatGPT for these tasks for more than 25% of their working time (Zhang et al., \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Funa and Gabay (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e) similarly observed that faculty across generations in higher education use GenAI primarily for ideation and rapid feedback, with younger participants tending to trust outputs more readily. Kasneci et al. (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) highlight potential benefits for students and educators\u0026mdash;such as quiz generation, simplification of complex content, and adaptive feedback\u0026mdash;while cautioning about bias, overreliance, and ethical concerns. Consistent with these cautions, Funa and Gabay (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2025b\u003c/span\u003e) and Meyer et al. (\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) emphasize that although GenAI can improve clarity, grammar, and readability, including for non-native English users, scholars should remain vigilant about factual inaccuracies, ethical issues, and model biases.\u003c/p\u003e\u003cp\u003eTo interpret these emerging practices, this review draws on complementary theoretical frameworks. It is anchored in socio-constructivist perspectives on writing and learning, which view writing as a socially mediated process shaped by tools, contexts, and interaction. From this perspective, learners develop writing proficiency through dialogue with peers, engagement with cultural tools, and iterative practice supported by scaffolds (Flower \u0026amp; Hayes, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e1981\u003c/span\u003e; Vygotsky, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e1978\u003c/span\u003e). GenAI tools such as ChatGPT and Gemini can therefore be conceptualized as cognitive and metacognitive scaffolds that provide immediate feedback, offer alternative phrasings, and suggest structural improvements to support planning, revision, and refinement. These tools function as mediators in the social process of writing and may extend a writer\u0026rsquo;s zone of proximal development by offering access to language models and ideas that might otherwise be unavailable. In parallel, the paradigm of foundation models (Bommasani et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) situates GenAI within a broader shift toward flexible, generalizable AI systems that mediate knowledge creation across disciplines. Integrating these perspectives provides a robust basis for examining how GenAI reshapes writing practices and for considering the pedagogical, ethical, and institutional implications that follow.\u003c/p\u003e\u003cp\u003eAgainst this backdrop, current scholarship on GenAI and academic writing continues to expand but often addresses isolated tools, single-discipline applications, or specific aspects of writing support. Comprehensive syntheses that map patterns of use across educational contexts remain limited (Kasneci et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Early advances in LLMs began influencing educational technologies around 2020 and laid the groundwork for new AI-mediated learning tools. The public release of ChatGPT in November 2022 marked a turning point, prompting rapid growth in research and adoption, particularly as digital and AI-assisted platforms became integral to academic workflows in the post-pandemic period (Bisi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Bommasani et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Huh, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; OpenAI, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThis review adopts a scoping review approach, which is well suited to this topic because the field of GenAI is evolving rapidly, the available evidence is heterogeneous, and the objective is to map the breadth of existing literature rather than evaluate intervention effectiveness or test a narrowly defined hypothesis (Arksey \u0026amp; O\u0026rsquo;Malley, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Tricco et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). It builds on a systematic review by Chanpradit (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), who synthesized 30 empirical studies from 2023 to 2024 and reported gains in cohesion, clarity, creativity, fluency, and proficiency, alongside risks such as plagiarism, overreliance, hallucinations, bias, and unequal access. The review also recommended institutional guidelines, transparent data practices, human oversight, and structured training. Extending this work, the present scoping review covers studies published from 2024 through the second quarter of 2025 to capture both ongoing adoption and emerging developments in higher education. By synthesizing peer-reviewed studies from this period, the review maps current applications, challenges, and research gaps in the use of GenAI for academic writing and identifies priorities for further investigation.\u003c/p\u003e\u003cp\u003eGuided by the Population\u0026ndash;Concept\u0026ndash;Context (PCC) framework from the Joanna Briggs Institute (Arksey \u0026amp; O\u0026rsquo;Malley, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Levac et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Peters et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), the researchers define the population as individuals engaged in academic writing in higher education (students, faculty, and researchers), the concept as the use of GenAI in academic writing, and the context as higher-education settings. Accordingly, the researchers address: (RQ1) How is GenAI being used in academic writing? (RQ2) What benefits and opportunities are reported? (RQ3) What challenges, risks, or ethical issues are identified? (RQ4) What gaps and future research directions are highlighted by existing studies?\u003c/p\u003e"},{"header":"2. Methodology","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1. Research Design\u003c/h2\u003e\u003cp\u003eThis scoping review followed the methodological framework of Arksey and O\u0026rsquo;Malley (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2005\u003c/span\u003e), refined by Levac et al. (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) and Peters et al. (\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), guided by procedures outlined in Funa et al. (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), and reported in accordance with the PRISMA Extension for Scoping Reviews (PRISMA-ScR; Tricco et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The approach was selected to map the breadth of evidence on the use of GenAI in academic writing and to identify research gaps in a rapidly evolving field. Eligibility and synthesis were structured using the PCC mnemonic: Population (students, faculty, and researchers in higher education), Concept (use of GenAI in academic writing), and Context (higher-education scholarly settings).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2. Search Strategy and Study Selection\u003c/h2\u003e\u003cp\u003eAn initial search was performed using Harzing\u0026rsquo;s Publish or Perish (Harzing, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2007\u003c/span\u003e) to explore and retrieve relevant literature from multiple databases, including Google Scholar, Scopus, PubMed, Semantic Scholar, and Web of Science. A comprehensive list of keywords and descriptors was developed, refined, and iteratively tested across these databases. Boolean operators (AND, OR) were applied, and search terms were systematically combined and interchanged to ensure coverage across the three focal constructs: academic writing, GenAI, and higher education. For example, the primary term \u0026ldquo;academic writing\u0026rdquo; or \u0026ldquo;scientific writing\u0026rdquo; was combined with secondary terms such as \u0026ldquo;generative artificial intelligence,\u0026rdquo; \u0026ldquo;generative AI,\u0026rdquo; or \u0026ldquo;GenAI,\u0026rdquo; together with \u0026ldquo;higher education\u0026rdquo; or \u0026ldquo;university students,\u0026rdquo; using AND/OR.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3. Inclusion and Exclusion Criteria\u003c/h2\u003e\u003cp\u003eFollowing the recommendations of Levac et al. (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2010\u003c/span\u003e), the inclusion and exclusion criteria were designed to align directly with the research questions and were refined through iterative team discussions during the initial screening phase. The study selection process is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. A study was included if it met all of the following conditions: (a) it was a peer-reviewed journal article published between 2024 and the second quarter of 2025, reflecting the rapid advancements and evolving applications of GenAI during this period; (b) it made an explicit reference to the use, application, or impact of GenAI\u0026mdash;such as ChatGPT, Gemini, or other LLMs\u0026mdash;within the context of academic writing/scientific writing; (c) it was written in English to ensure accurate interpretation; (d) it reported original empirical findings, ensuring the synthesis was based on primary data or firsthand analyses rather than secondary syntheses; and (e) it focused on higher education institutions. Conference papers and other non-journal sources were excluded to maintain the rigor and comparability of the included evidence. Restricting the review to empirical studies in higher education ensured that the findings provided robust, evidence-based insights into the phenomenon under investigation. The criteria were piloted and refined before the full screening process to enhance clarity and ensure consistent application.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents the PRISMA-ScR flow for study identification, screening, eligibility, and inclusion for this scoping review (2024\u0026ndash;Q2 2025). Across databases, 1,599 records were identified; 1,334 were removed prior to screening, leaving 265 titles/abstracts screened. Of these, 180 were excluded at screening and 85 full texts were assessed for eligibility. 60 full-text articles were excluded (primary reasons summarized in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e; the most common was no clear link between GenAI use and academic writing in higher education), yielding 25 studies included in the final synthesis.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4. Coding Procedures\u003c/h2\u003e\u003cp\u003e Data from the included studies were extracted and organized using a standardized data charting form developed by the review team. In line with Levac et al. (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2010\u003c/span\u003e), the coding process was iterative, allowing the team to refine categories as familiarity with the literature increased. Each article was coded for key variables such as publication details (author \u0026amp; year), country/region, population (students, faculty, researchers), study context, study design/methodology, type of GenAI tool/s used, specific applications in academic writing (e.g., brainstorming, drafting, paraphrasing, summarizing), reported benefits and opportunities, identified challenges and ethical issues, and stated research gaps or recommendations for future studies.\u003c/p\u003e\u003cp\u003eTwo researchers independently coded each included study to enhance reliability. Discrepancies were discussed and resolved through consensus, with a third researcher consulted when necessary. Excel was used to facilitate organization and thematic grouping of data. Codes were grouped into higher-order categories aligned with the research questions, and emerging themes were refined through ongoing team discussion. Consistent with scoping review guidance, the objective was to map the evidence; therefore, no formal critical appraisal of study quality was undertaken. Findings were synthesized using a descriptive numerical summary of study characteristics and a qualitative thematic synthesis mapped to the PCC framework and the review questions.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5. Characteristics of the Included Studies\u003c/h2\u003e\u003cp\u003eThis scoping review synthesized peer-reviewed empirical studies published from 2024 to Q2 2025 (n\u0026thinsp;=\u0026thinsp;25) that investigated GenAI in academic writing within higher education contexts (see Appendix A). To aid interpretation, settings are grouped by continent (see Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e): Europe (n\u0026thinsp;=\u0026thinsp;11: United Kingdom [n\u0026thinsp;=\u0026thinsp;7], Denmark, Norway, Greece, Switzerland), Asia (n\u0026thinsp;=\u0026thinsp;7: China [n\u0026thinsp;=\u0026thinsp;3], Hong Kong [n\u0026thinsp;=\u0026thinsp;2], United Arab Emirates, T\u0026uuml;rkiye), Americas (n\u0026thinsp;=\u0026thinsp;4: Ecuador [n\u0026thinsp;=\u0026thinsp;2], Chile, United States), and Africa (n\u0026thinsp;=\u0026thinsp;1: South Africa), with two cross-regional studies (one bi-national Finland\u0026ndash;New Zealand; one global/multi-country survey).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eAcross 25 studies, participants were predominantly student-only samples: 19 of 25 (76%); faculty-only: 3 of 25 (12%); and mixed student-faculty: 3 of 25 (12%). Sample sizes ranged from small, course-embedded cohorts (e.g., 6 students in Denmark [Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e]; 20 students at a Sino-British EMI university [Kim et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e]) to large cross-sectional surveys (e.g., 2,555 students at a UK university [Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e]; a global mixed sample of n\u0026thinsp;=\u0026thinsp;1,217 students and lecturers [Yusuf et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2024\u003c/span\u003e]). Staff-only samples also appeared (e.g., n\u0026thinsp;=\u0026thinsp;284 UK academics [Watermeyer et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2024\u003c/span\u003e]; n\u0026thinsp;=\u0026thinsp;184 university teachers in Ecuador [Cordero et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e]).\u003c/p\u003e\u003cp\u003eStudy designs were mainly descriptive and exploratory, comprising quantitative surveys (Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), qualitative interviews or focus groups (Hysaj et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), mixed-methods designs (Han, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), task-based observational studies (Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), and a small number of course-embedded interventions (Wang \u0026amp; Ren, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The most commonly examined tools were ChatGPT (n\u0026thinsp;=\u0026thinsp;23), followed by Grammarly (n\u0026thinsp;=\u0026thinsp;4), Microsoft Copilot (n\u0026thinsp;=\u0026thinsp;3), Perplexity (n\u0026thinsp;=\u0026thinsp;2), QuillBot (n\u0026thinsp;=\u0026thinsp;1), and local or institution-provided LLMs (n\u0026thinsp;=\u0026thinsp;2). Reported writing applications centered on brainstorming/idea generation and planning/outlining (Nguyen et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), as well as summarizing literature and drafting/refinement within course tasks (Wang \u0026amp; Ren, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Across contexts, authors noted benefits such as efficiency/time savings and greater confidence/self-efficacy (Campbell \u0026amp; Cox, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Hysaj et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), along with language support for L2 writers and organizational help (planning/structuring, grammar support) (Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Recurrent challenges included risks of plagiarism/contract cheating and policy gaps/uneven AI literacy (Campbell \u0026amp; Cox, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Hysaj et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), hallucinations and unverifiable sources (Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), overreliance that may erode writing skills and limitations of detection tools (Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), and broader academic-integrity concerns (Nelson et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e2.6. Data Analysis Procedures\u003c/h2\u003e\u003cp\u003eData analysis followed established scoping-review methodology that prioritizes mapping the breadth of evidence and key concepts over effect estimation (Arksey \u0026amp; O\u0026rsquo;Malley, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Levac et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Peters et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Tricco et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Guided by the PCC framework, the researchers conducted two complementary analytic strands and then integrated findings to address the review questions.\u003c/p\u003e\u003cp\u003eUsing Microsoft Excel, the researchers obtained quantitative data by coding publication year, country and continent, participant group, study design, and GenAI tools. Multi-country items were coded as cross-regional. Tool tallies reflect studies that examined or instructed the use of a tool at least once; because some studies included multiple tools, tool counts can exceed n\u0026thinsp;=\u0026thinsp;25. Cross-tabulations (for example, tool and participant group; design and continent) were generated to surface distributional patterns. Outputs included summary tables and figures (for example, PRISMA flow; continent map) consistent with PRISMA-ScR reporting (Tricco et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Qualitative data were extracted during charting text fields (applications in academic writing, benefits and opportunities, challenges and ethical issues, and gaps or recommendations) were analyzed using a deductive\u0026ndash;inductive thematic analysis (Braun \u0026amp; Clarke, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The researchers began with a deductive coding frame aligned to PCC and the review questions, conducted line-by-line open coding to allow additional categories to emerge, iteratively refined a shared codebook, and developed higher-order themes through constant comparison and memoing. Quantitative and qualitative results were combined using a convergent integrated approach in the Discussion section: numerical patterns (for example, continent or tool frequencies) were juxtaposed with qualitative themes in side-by-side matrices and narrative weaving to produce an evidence map that directly answers each review question (Fetters et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Peters et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003eFindings are organized around the four research questions and integrate a descriptive numerical summary with a qualitative thematic synthesis. Drawing on the 25 included studies, the researchers first map how GenAI is being used in academic writing (RQ1), then synthesize reported benefits and opportunities (RQ2), followed by challenges, risks, and ethical issues (RQ3), and finally identify gaps and future research directions (RQ4). This structure aligns with the review\u0026rsquo;s PCC framing and the stated research questions.\u003c/p\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e3.1. Ways GenAI is being used in academic writing (RQ1)\u003c/h2\u003e\u003cp\u003eAcross the 25 studies, GenAI is used across the full writing cycle, from pre-writing through finalization (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Students most commonly employ chatbots to clarify concepts, request definitions and examples, and generate outlines or essay plans, indicating early-stage support for planning and structuring (Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). GenAI is then used to draft and reorganize text, polish language, and refine coherence, with classroom data showing perceived utility for idea generation, vocabulary support, grammar correction, and argument organization (Wang \u0026amp; Ren, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In multilingual contexts, learners report using GenAI to paraphrase, simplify readings, and address language mechanics, describing these uses as practical supports for completing written assessments (Hysaj et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Source-work and referencing support also appear frequently, although students\u0026rsquo; practices are uneven; for example, students plan and search with GenAI but often omit acknowledging the tool itself in references, underscoring an ongoing need for information-literacy guidance (Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). ChatGPT is the most frequently used tool, with additional use of Grammarly, Copilot, Perplexity, QuillBot, and institution-provided or local LLMs in several studies (Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eWays GenAI is used across stages of academic writing (n\u0026thinsp;=\u0026thinsp;25).\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWriting stage\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTypical GenAI uses\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eRepresentative tools\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eExample sources\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003ePre-writing\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eClarify concepts, generate ideas/topics, produce examples\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eChatGPT; local LLMs\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eJohnston et al. (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e); Johnston et al. (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003ePlanning \u0026amp; outlining\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCreate outlines, reorganize structure, plan sections\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eChatGPT; Copilot; Perplexity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eJohnston et al. (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e); Nguyen et al. (\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eDrafting\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eProduce first drafts, expand points, suggest wording\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eChatGPT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eWang \u0026amp; Ren (\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eRevising \u0026amp; refining\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRewrite passages, improve coherence/flow, style and tone\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eChatGPT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eWang \u0026amp; Ren (\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e); Hysaj et al. (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eLanguage support (L2)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eParaphrase/translate, vocabulary support, grammar/mechanics\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGrammarly; QuillBot; ChatGPT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eHysaj et al. (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e); Johnston et al. (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eWorking with sources\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGenerate search terms, summarize articles, format references\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePerplexity; ChatGPT; Copilot\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eJohnston et al. (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e); Jensen \u0026amp; Jensen (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eMultimodal support\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCreate/plan visuals to accompany text; slide notes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eChatGPT (image features)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eWang \u0026amp; Ren (\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the studies depict whole-process integration rather than single-point use. Planning and conceptual scaffolding are especially prominent, followed by text production and language polishing. Source-work support is common, but attribution and referencing practices lag; students frequently use GenAI when planning or searching yet omit citing the tool, indicating a gap for AI-informed information-literacy instruction (Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). While ChatGPT dominates use, several studies encourage broadening tool awareness to alternatives such as Copilot and Perplexity, and to institution-provided LLMs where available (Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e3.2. Benefits and opportunities reported regarding GenAI-assisted writing (RQ2)\u003c/h2\u003e\u003cp\u003eAs summarized in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, the researchers identified six, recurrent benefit clusters that map onto the writing process (planning, drafting, revising, and finalizing). First, students consistently reported efficiency and time savings, with GenAI handling lower-level mechanics (grammar, phrasing, formatting) and thereby freeing attention for higher-order concerns such as argumentation and evidence use (Campbell \u0026amp; Cox, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Han, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Wang \u0026amp; Ren, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Second, GenAI offered substantial language support, particularly in EAP/L2 contexts, where learners used it to paraphrase, translate, expand vocabulary, and improve clarity and fluency; these functions were frequently linked to increased confidence and self-efficacy in completing written assessments (Hysaj et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Moorhouse et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Nelson et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThird, tools were widely used for planning and organization\u0026mdash;brainstorming, outlining, and structuring paragraphs\u0026mdash;often improving task interpretation and the perceived coherence of drafts (Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Mo \u0026amp; Crosthwaite, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Wang \u0026amp; Ren, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Fourth, several studies highlighted formative feedback and scaffolding: rapid explanations, exemplars, and revision suggestions supported iterative improvement; in some cases, source-display features helped students plan searches and check claims (Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eFifth, the literature points to access and inclusion opportunities. Students with disabilities described GenAI as helpful for planning, drafting, and multimodal expression (e.g., generating alternatives or simplifying language), indicating potential to reduce participation barriers when used with appropriate guidance (Zhao et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Finally, at the pedagogical and institutional levels, studies framed GenAI as an opportunity space for course/assessment redesign and AI-literacy development\u0026mdash;for example, integrating transparent AI use into authentic assessment, and embedding guidance on prompting, verification, and attribution (Cordero et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Kofinas et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eTaken together, the pattern across contexts suggests GenAI\u0026rsquo;s strongest contributions cluster in the planning-to-polishing span of the writing cycle. Benefits are maximized when use is transparent, scaffolded, and paired with verification practices, positioning GenAI as an assistive resource rather than a substitute for disciplinary thinking and academic integrity (Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e3.3. Challenges, risks, and ethical issues identified in the literature (RQ3)\u003c/h2\u003e\u003cp\u003eSynthesizing the 25 studies, the researchers identified five recurrent risk domains that cut across the writing process (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). First, source reliability and epistemic risk remain prominent. Studies document hallucinations, unverifiable claims, and fabricated citations that can be difficult for novice writers to detect, underscoring the need for systematic verification and triangulation when GenAI is used for content generation or source work (Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zizka, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eSecond, issues of authorship, attribution, and academic integrity are widely reported. Surveys and classroom investigations note inconsistent acknowledgment of tool use, uncertainty about the boundary between acceptable support and misconduct, and risks of plagiarism or contract cheating when outputs are submitted with minimal transformation (Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Kofinas et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Nelson et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Third, the literature points to overreliance and skill development concerns. Without scaffolding, students may defer critical reading, argumentation, and revision to GenAI, exhibiting automation bias and reduced practice with higher-order writing skills; several papers caution that \u0026ldquo;hands-off\u0026rdquo; use can erode competence over time (Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Han, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Watermeyer et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eFourth, assessment alignment and detection limits pose persistent challenges. Tasks that can be solved by generic prompts invite superficial engagement, while AI-detection tools are unreliable, susceptible to manipulation, and not suitable as sole evidence in integrity processes. The literature instead recommends process-focused assessment, multi-source evidence in investigations, and explicit expectations about permissible GenAI support (Kofinas et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; van Niekerk et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Fifth, studies surface governance, equity, and privacy issues. Uneven AI literacy and unclear policies produce inconsistent practice across courses; students also express concerns about sharing assignments with third-party systems, potential data exposure, and unequal access to paid tools and connectivity that can amplify existing inequities (Campbell \u0026amp; Cox, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Rodafinos, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhao et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Stanford, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe synthesized studies suggest that risk is highest where GenAI intersects with source credibility, authorship norms, and assessment design. Consistent recommendations across studies include transparent policies with required disclosure of tool use, explicit training in verification and source work, redesign of assessments to emphasize process and originality (e.g., staged drafts and checkpointing), avoidance of detectors as the sole basis for misconduct decisions, and privacy-safe, equitable access practices (Kofinas et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; van Niekerk et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e3.4. Gaps and future research directions highlighted by synthesized studies (RQ4)\u003c/h2\u003e\u003cp\u003eThe researchers distilled seven cross-cutting gaps and aligned each with suggested methods and target outcomes (see Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Collectively, the evidence base remains largely descriptive; stronger causal designs are needed to test whether GenAI improves writing quality, learning, or transfer across time and genres. Priority studies include longitudinal cohorts and classroom experiments with transparent comparison conditions and validated rubrics (Kofinas et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Wang \u0026amp; Ren, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Kim et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Moorhouse et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zizka, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eSummary of Research Gaps, Suggested Methods, and Target Outcomes in GenAI-Assisted Academic Writing (n\u0026thinsp;=\u0026thinsp;25).\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eResearch gaps\u003c/p\u003e\u003cp\u003e(what needs evidence)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSuggested methods\u003c/p\u003e\u003cp\u003e(how to study it)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTarget outcomes to report\u003c/p\u003e\u003cp\u003e(what to measure)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRepresentative studies\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e1. Causal effects on writing quality, learning, and transfer\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLongitudinal cohorts; classroom experiments or quasi-experiments with transparent comparison conditions; multi-site replications\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eWriting quality using validated rubrics; learning gains and transfer to new genres; persistence over time; time-on-task and revision productivity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eKofinas et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Wang \u0026amp; Ren, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Kim et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Moorhouse et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zizka, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2025\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e2. Reporting and attribution practices\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eField experiments embedding disclosure requirements; audit studies of assignments; mixed-methods studies of student/marker perceptions\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eDisclosure rates; accuracy/completeness of AI-use statements; impacts on grades and feedback; perceived fairness/integrity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eJohnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; van Niekerk et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e3. AI-literacy and pedagogy interventions\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDesign-based research (DBR) on curricula; randomized or quasi-experimental evaluation of modules on prompting, verification, and source work\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAI-literacy competency gains; verification accuracy; quality of source work; metacognitive strategy use; student confidence/self-efficacy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eCampbell \u0026amp; Cox, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Mo \u0026amp; Crosthwaite, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Hysaj et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Nguyen et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e4. Assessment and policy design\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eComparative studies of assessment formats (staged drafts, in-class writing, viva/checkpoints); policy implementation evaluations; process tracing of drafting workflows\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMisconduct allegations and outcomes; detector false-positive/negative rates; policy compliance/fidelity; marker workload; student satisfaction and perceived fairness\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eKofinas et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; van Niekerk et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e5. Equity, accessibility, and privacy\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eStudies with disability subgroups; usability testing; surveys/interviews on access; privacy impact assessments\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAccessibility gains; accommodation effectiveness; access gaps (device, bandwidth, paid tools); privacy incidents and data sharing; differential outcomes by subgroup\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eZhao et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Watermeyer et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Rodafinos, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Stanford, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2025\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e6. Multilingual and cross-cultural contexts\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCross-site studies in non-Anglophone HEIs; L1/L2 comparisons; corpus-informed analyses of genre and register\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLanguage quality (clarity, cohesion, accuracy) in L2 writing; translation/paraphrase fidelity; genre conformity; cultural/disciplinary fit\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNelson et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Adalı \u0026amp; Bilgili, 2025; Hysaj et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e7. Measurement standards and replication\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eConsensus methods (e.g., Delphi) to define taxonomies; preregistered protocols; shared prompts and model versions; open materials for replication\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eReporting-checklist compliance; reproducibility of results; sensitivity to model/version/prompt; open datasets and code availability\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eMo \u0026amp; Crosthwaite, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Han, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eA second cluster concerns reporting and attribution. Multiple papers document inconsistent disclosure of GenAI use and uncertainty about acknowledgment norms, especially when tools assist with planning or paraphrase. Table\u0026nbsp;3 therefore recommends field experiments that embed disclosure requirements and mixed-methods audits of student and marker perceptions, with outcomes such as disclosure rates, accuracy of AI-use statements, and effects on grading and feedback (Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; van Niekerk et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Third, the literature calls for systematic evaluation of AI-literacy and pedagogy. While many studies advocate instruction in prompting, verification, and source work, few evaluate structured curricula at scale. Design-based research and quasi-experimental modules should report competency gains, verification accuracy, quality of source work, and metacognitive strategy use (Campbell \u0026amp; Cox, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Mo \u0026amp; Crosthwaite, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Hysaj et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Nguyen et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eFourth, assessment and policy design require rigorous testing. Recommended directions include comparative evaluations of process-oriented formats (e.g., staged drafts, in-class writing, oral checkpoints), policy implementation studies, and process tracing of drafting workflows; outcomes should include integrity incidents, false-positive rates from detectors, policy fidelity, marker workload, and perceived fairness (Kofinas et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; van Niekerk et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Fifth, equity, accessibility, and privacy remain under-researched. Early findings suggest potential to reduce participation barriers for students with disabilities but also raise concerns about unequal access to paid tools and data-sharing risks. Future work should incorporate subgroup analyses, usability testing, and privacy-impact assessments with clear reporting of accessibility gains and differential outcomes (Zhao et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Watermeyer et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Rodafinos, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Stanford, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eSixth, the field needs broader coverage of multilingual and cross-cultural contexts. Studies should move beyond Anglophone settings to examine L1/L2 differences, translation/paraphrase fidelity, and genre conformity in non-English HEIs (Nelson et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Karahan Adalı \u0026amp; Bilgili, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Hysaj et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Finally, measurement standards and replication would improve comparability. The researchers recommend consensus on use-case taxonomies (planning, drafting, revising), preregistered protocols, and sharing of prompts, model versions, datasets, and code; core outcomes should include rubric-based writing quality, learning/transfer, integrity outcomes, and sensitivity to model/version changes (Mo \u0026amp; Crosthwaite, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Han, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThe researchers interpreted the map of evidence through the PCC lens using convergent integration, aligning numerical distributions with qualitative themes to explain what the patterns mean for academic writing in higher education rather than restating procedures or counts (Fetters et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Tricco et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). As depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, a PCC-aligned convergent model guides this interpretation: Population, Concept, and Context flow into a convergent-integration step (shown as a dashed band) that positions GenAI as assistive scaffolding while core scholarly practices remain grounded in human judgment. This process emphasis aligns with evidence that explicit argument scaffolds measurably improve conceptual understanding across levels and delivery modes, indicating that GenAI should be paired with argument-based routines rather than replace them (Ramallosa et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eWith respect to RQ1, the researchers found convergence between tool prevalence and narratives of practice: widespread reliance on general-purpose chatbots alongside grammar and paraphrase tools corresponds to reported gains in clarity, organization, and confidence, especially among multilingual writers. Yet, qualitative accounts also underline that idea development, disciplinary reasoning, and source-critical reading are not reliably automated, which situates GenAI as a facilitator of process rather than a substitute for scholarly authorship (Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). This emphasis on process support is represented on the left of Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e as \u0026ldquo;Assistive scaffolding (Planning to Revision).\u0026rdquo;\u003c/p\u003e\u003cp\u003eFor RQ2 and RQ3, the researchers observed a stable pattern of benefits\u0026mdash;efficiency, fluency, and reduced language barriers\u0026mdash;counterbalanced by recurrent risks, notably hallucinations, fragile or fabricated citations, uneven disclosure, and signs of overreliance where scaffolding is absent. Divergence between stakeholder perspectives is salient: students often report improved fluency and task confidence, whereas instructors more frequently raise concerns about originality, source quality, and process visibility. The synthesis therefore supports the view that GenAI should augment, not replace, disciplinary thinking and source work, and that verification practices must be made explicit within assignments and feedback cycles (Campbell \u0026amp; Cox, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zizka, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). The central \u0026ldquo;Human judgment\u0026rdquo; node in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e visually anchors this requirement for verification and attribution/disclosure. Positioning GenAI as a tool that supports claim\u0026ndash;evidence\u0026ndash;reasoning and structured rebuttal is consistent with meta-analytic gains from argument-based learning (Ramallosa et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), while keeping human verification central.\u003c/p\u003e\u003cp\u003eInterpreting the evidence through PCC also highlights where convergence and divergence matter. On population, the student-heavy corpus limits insight into supervisory practices, disclosure norms, and assessment decisions among faculty and postgraduate researchers. The researchers therefore note the need for more staff-focused and mixed-cohort studies to illuminate how expectations are translated into grading and feedback in authentic settings (Watermeyer et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Cordero et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). On concept, whole-process assistance rather than single-step substitution explains the robust improvements in organization and language contrasted with mixed results for higher-order reasoning (Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). On context, institutions with educative, transparent policies report more constructive uses\u0026mdash;such as declared assistance with verification\u0026mdash;while resource-constrained settings surface equity, access, and privacy concerns more sharply, suggesting that policy effectiveness is contingent on local conditions and support (Moorhouse et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhao et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Findings from secondary science during distance education show that learner-centered, action-oriented, and transformative practices emerged despite constraints, but inadequate equipment and poor connectivity were persistent barriers (Funa et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2023\u003c/span\u003e); these realities should shape GenAI policy to avoid deepening access gaps. These PCC elements are shown at the top of Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, with directional arrows into the convergent-integration band.\u003c/p\u003e\u003cp\u003eThese patterns help resolve the assistive\u0026ndash;substitution debate. Where courses embed GenAI as taught scaffolding within a staged writing process, the researchers observed improvements without systematic loss of authorial voice. Where use is broad and unscaffolded, automation bias, shallow source engagement, and dependence are more likely. The literature converges on three guardrails: verifiability of content and references, transparent attribution or disclosure, and process-oriented assessment that makes thinking visible; by contrast, sole reliance on AI-detection tools is widely considered insufficient evidence for adjudicating integrity (Han, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Kofinas et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; van Niekerk et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Converging with these findings, a recent meta-analysis of inquiry-based learning likewise reports substantial improvements in students\u0026rsquo; conceptual understanding, particularly under open-inquiry conditions (Mediana Jr. et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e); by analogy, GenAI should be embedded as a scaffold that sustains inquiry and verification rather than as an autonomous text generator. The bidirectional arrow between the two teal panels in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e signals interaction\u0026mdash;GenAI can support process without supplanting core scholarly practices.\u003c/p\u003e\u003cp\u003eThe researchers\u0026rsquo; implications also resonate with the I-STEM-PBL-ESD instructional framework, which integrates problem-based learning with education for sustainable development; GenAI tasks can be embedded as supports for problem framing, scenario exploration, and reflective synthesis while safeguarding attribution and process visibility (Funa et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Implications follow directly for curriculum and policy. The researchers interpret the evidence as supporting AI-literacy that teaches prompting for thinking, verification strategies, and explicit attribution norms; assessment designs that document process via staged drafts, in-class checkpoints, and brief oral or written justifications; and equity-minded implementation that addresses differential access, accessibility needs, and data-privacy risks. These directions align with emerging institutional scholarship that emphasizes responsible, transparent adoption calibrated to local contexts (Funa \u0026amp; Gabay, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025a\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2025b\u003c/span\u003e; Cordero et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003ePositioning this review within prior syntheses, the researchers corroborate early findings of clarity and fluency gains alongside integrity and equity concerns, while extending the field by focusing on 2024 to Q2 2025 studies and by integrating numerical distributions with qualitative themes. The added value lies in specifying where benefits reliably cluster (planning-to-polishing stages), which risks remain unresolved (verification and attribution), and which methodological moves are now necessary to advance knowledge\u0026mdash;namely causal and longitudinal designs testing learning and transfer, systematic studies of disclosure and authorship practices, cross-cultural comparisons beyond Anglophone/EMI contexts, and clearer reporting standards for AI involvement (Chanpradit, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Mo \u0026amp; Crosthwaite, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Tricco et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Consistent with Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, the implications panel synthesizes these priorities: educative policy; verification and attribution/disclosure; process-oriented assessment; equity and privacy; and a forward research agenda.\u003c/p\u003e\u003cp\u003eOverall, the researchers interpret GenAI as most effective when treated as an assistive, transparent scaffold embedded in educative policy and process-forward pedagogy. Under these conditions, gains in organization, clarity, and productivity are achievable without compromising authorial ownership; under substitution framings, the risks noted above intensify. The field therefore benefits from designs and policies that preserve human judgment while leveraging GenAI\u0026rsquo;s strengths in planning, feedback, and language support (Jensen \u0026amp; Jensen, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Johnston et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhao et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e"},{"header":"5. Limitations of the Study","content":"\u003cp\u003eThe researchers conducted a scoping review to map GenAI use in higher-education writing, not to judge study quality or estimate effects. Inclusion filters (peer-reviewed, English, higher education, 2024\u0026ndash;Q2 2025) may introduce language/publication bias and exclude preprints and conference work. Rapidly evolving terminology and tool stacks mean some relevant studies may have been missed, and conclusions rest on primary studies that often use small, single-site samples and proxy outcomes. The evidence base is skewed toward better-resourced, English-medium institutions with limited faculty/postgraduate and Global South representation. Finally, results may shift as GenAI capabilities and institutional policies change, and theme coding involved researcher judgment.\u003c/p\u003e"},{"header":"6. Conclusions and Recommendations","content":"\u003cp\u003eThe researchers conclude that GenAI currently functions most productively as assistive scaffolding across the planning-to-revision span of academic writing in higher education. When paired with explicit expectations for verification, attribution or disclosure, and visible processes, GenAI can improve organization, clarity, and productivity without displacing core scholarly practices such as argument construction, source interrogation, and synthesis. Divergences in outcomes appear when use is unscaffolded, when originality is treated as a product metric rather than a process, and when institutions rely on detection tools as sole evidence of integrity. The present map also shows important evidence gaps, including limited representation of faculty and postgraduate cohorts, under-representation of the Global South, and an over-reliance on short-term proxy measures of learning. Taken together, these patterns support an assistive-not-substitutive stance that centers human judgment while leveraging GenAI for formative support.\u003c/p\u003e\u003cp\u003eThe researchers recommend an institution-wide approach that treats GenAI as assistive scaffolding within clearly articulated policies and taught practices. Specifically, universities should implement AI-literacy programs that teach prompting for thinking, verification strategies, and explicit attribution or disclosure; course policies should define permitted uses, privacy expectations, and data handling; and faculty development should emphasize task designs that keep human reasoning visible and feedback that targets argument quality and source use. Assessment should be process-centered\u0026mdash;requiring staged drafts, in-class checkpoints, and brief oral or written justifications\u0026mdash;to make authorship and decision pathways transparent; institutions should avoid sole reliance on AI-detection tools and instead use multi-source evidence that includes process artifacts and instructor judgment. To promote equity and access, the researchers advise providing institutionally vetted tools, accessibility features, and low-bandwidth or offline pathways so that GenAI support does not widen existing gaps. In course design, GenAI should scaffold question generation, planning, formative feedback, and reflective synthesis while reserving claim-evidence-reasoning, source evaluation, and final interpretations for learners. Finally, the research agenda should prioritize causal and longitudinal studies of learning and transfer, broaden representation to include faculty and postgraduate cohorts and non-Anglophone or resource-constrained contexts, and improve reporting standards by documenting model versions, prompts, guardrails, and disclosure practices while employing outcomes that capture reasoning, ethics, and sustained learning rather than only surface features.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAcut, D. P., Malabago, N. K., Malicoban, E. V., Galamiton, N. S., \u0026amp; Garcia, M. B. (2024). \u0026ldquo;ChatGPT 4.0 Ghosted Us While Conducting Literature Search:\u0026rdquo; Modeling the Chatbot\u0026rsquo;s Generated Non-Existent References Using Regression Analysis. \u003cem\u003eInternet Reference Services Quarterly\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(1), 27\u0026ndash;54. https://doi.org/10.1080/10875301.2024.2426793\u003c/li\u003e\n\u003cli\u003eArksey, H., \u0026amp; O\u0026rsquo;Malley, L. (2005). Scoping studies: Towards a methodological framework. \u003cem\u003eInternational Journal of Social Research Methodology, 8\u003c/em\u003e(1), 19\u0026ndash;32. https://doi.org/10.1080/1364557032000119616\u003c/li\u003e\n\u003cli\u003eBisi, T., Risser, A., Clavert, P., Migaud, H., \u0026amp; Dartus, J. (2023). What is the rate of text generated by artificial intelligence over a year of publication in Orthopedics \u0026amp; Traumatology: Surgery \u0026amp; Research? Analysis of 425 articles before versus after the launch of ChatGPT in November 2022. \u003cem\u003eOrthopaedics \u0026amp; Traumatology: Surgery \u0026amp; Research\u003c/em\u003e, \u003cem\u003e109\u003c/em\u003e(8), 103694.\u003c/li\u003e\n\u003cli\u003eBommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., Brynjolfsson, E., Buch, V., Card, D., Castellon, D., Chen, A., Creel, K. A., Davis, J. Q., Demszky, D., Donahue, C., Doumbouya, M., \u0026hellip; Liang, P. (2021). On the opportunities and risks of foundation models. \u003cem\u003earXiv.\u003c/em\u003e https://doi.org/10.48550/arXiv.2108.07258\u003c/li\u003e\n\u003cli\u003eBraun, V., \u0026amp; Clarke, V. (2006). Using thematic analysis in psychology. \u003cem\u003eQualitative Research in Psychology, 3\u003c/em\u003e(2), 77\u0026ndash;101. https://doi.org/10.1191/1478088706qp063oa \u003c/li\u003e\n\u003cli\u003eBraun, V., \u0026amp; Clarke, V. (2021). \u003cem\u003eThematic analysis: A practical guide\u003c/em\u003e. SAGE.\u003c/li\u003e\n\u003cli\u003eCampbell, L. O., \u0026amp; Cox, T. D. (2024). Utilizing generative AI in higher education teaching and learning. \u003cem\u003eJournal of the Scholarship of Teaching and Learning, 24\u003c/em\u003e(4), 162\u0026ndash;173. https://doi.org/10.14434/josotl.v24i4.36575 \u003c/li\u003e\n\u003cli\u003eChanpradit, T. (2025). Generative artificial intelligence in academic writing in higher education: A systematic review. \u003cem\u003eEdelweiss Applied Science and Technology, 9\u003c/em\u003e(4), 889\u0026ndash;906. https://doi.org/10.55214/25768484.v9i4.6128 \u003c/li\u003e\n\u003cli\u003eCordero, J., Torres-Zambrano, J., \u0026amp; Cordero-Castillo, A. (2025). Integration of generative artificial intelligence in higher education: Best practices. \u003cem\u003eEducation Sciences, 15\u003c/em\u003e(1), 32. https://doi.org/10.3390/educsci15010032 \u003c/li\u003e\n\u003cli\u003eEmran, A. Q., Mohammed, M. N., Saeed, H., Keir, M. Y. A., Alani, Z. N., \u0026amp; Ibrahim, F. M. (2024, January). Paraphrasing ChatGPT answers as a tool to enhance university students\u0026rsquo; academic writing skills. In \u003cem\u003e2024 ASU International Conference in Emerging Technologies for Sustainability and Intelligent Systems (ICETSIS)\u003c/em\u003e (pp. 501-505). IEEE.\u003c/li\u003e\n\u003cli\u003eFetters, M. D., Curry, L. A., \u0026amp; Creswell, J. W. (2013). Achieving integration in mixed methods designs: Principles and practices. \u003cem\u003eHealth Services Research, 48\u003c/em\u003e(6 Pt 2), 2134\u0026ndash;2156. https://doi.org/10.1111/1475-6773.12117 \u003c/li\u003e\n\u003cli\u003eFlower, L., \u0026amp; Hayes, J. R. (1981). A cognitive process theory of writing. \u003cem\u003eCollege Composition and Communication, 32\u003c/em\u003e(4), 365\u0026ndash;387. https://doi.org/10.2307/356600\u003c/li\u003e\n\u003cli\u003eFuna, A. A., \u0026amp; Gabay, R. A. E. (2025a). \u003cem\u003eExploring perspectives toward artificial intelligence integration in science education: A cross-generational study\u003c/em\u003e [Preprint]. SSRN. https://doi.org/10.2139/ssrn.5240513\u003c/li\u003e\n\u003cli\u003eFuna, A. A., \u0026amp; Gabay, R. A. E. (2025b). Policy guidelines and recommendations on AI use in teaching and learning: A meta-synthesis study. \u003cem\u003eSocial Sciences \u0026amp; Humanities Open\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e, 101221.\u003c/li\u003e\n\u003cli\u003eFuna, A. A., Gabay, R. A. E., Esdicul, K. L., \u0026amp; Prudente, M. S. (2023). Secondary teachers\u0026rsquo; and students\u0026rsquo; perceptions of distance education in science: Focus on learner-centered, action-oriented, and transformative learning. \u003cem\u003eDalat University Journal of Science, 13\u003c/em\u003e(3), 156\u0026ndash;181. https://doi.org/10.37569/DalatUniversity.13.3.1108(2023) \u003c/li\u003e\n\u003cli\u003eFuna, A. A., Ricafort, J. D., Jetomo, F. G. J., \u0026amp; Lasala Jr, N. L. (2024). Effectiveness of brain-based learning toward improving students\u0026rsquo; conceptual understanding: a meta-analysis. \u003cem\u003eInternational Journal of Instruction\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(1), 361-380.\u003c/li\u003e\n\u003cli\u003eFuna, A. A., Roleda, L. S., \u0026amp; Prudente, M. S. (2024). Integrated Science, Technology, Engineering, and Mathematics\u0026mdash;Problem-Based Learning\u0026mdash;Education for Sustainable Development (I-STEM-PBL-ESD) framework. In D. Luberto, G. Sharma, N. Capraro, \u0026amp; R. Yadav (Eds.), \u003cem\u003eA Diversity of Pathways Through Science Education\u003c/em\u003e. Springer. https://doi.org/10.1007/978-981-97-2607-3_9\u003c/li\u003e\n\u003cli\u003eGoodfellow, I. J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., ... \u0026amp; Bengio, Y. (2014). Generative adversarial nets. \u003cem\u003eAdvances in neural information processing systems\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eHan, Y. (2025). Beyond the algorithm: Reconciling generative AI and human agency in academic writing education. \u003cem\u003eInternational Journal of Learning and Teaching, 11\u003c/em\u003e(1), 39\u0026ndash;42. https://doi.org/10.18178/ijlt.11.1.39-42 \u003c/li\u003e\n\u003cli\u003e\u003cem\u003eHarzing, A.W. (2007) Publish or Perish, available from \u003c/em\u003e\u003cem\u003ehttps://harzing.com/resources/publish-or-perish\u003c/em\u003e\u003c/li\u003e\n\u003cli\u003eHuh, S. (2023). Emergence of the metaverse and ChatGPT in journal publishing after the COVID-19 pandemic. \u003cem\u003eScience Editing\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(1), 1-4.\u003c/li\u003e\n\u003cli\u003eHysaj, A., Dean, B. A., \u0026amp; Freeman, M. (2025). Exploring the purposes and uses of generative artificial intelligence tools in academic writing for multicultural students. \u003cem\u003eHigher Education Research \u0026amp; Development\u003c/em\u003e. Advance online publication. https://doi.org/10.1080/07294360.2025.2488862 \u003c/li\u003e\n\u003cli\u003eJensen, T. W., \u0026amp; Jensen, S. W. (2025). AI literacy in the context of working with sources: Pitfalls and possibilities of generative AI models in academic writing. \u003cem\u003eJournal of Academic Writing, 15\u003c/em\u003e(S2), 1\u0026ndash;12. https://doi.org/10.18552/joaw.v15iS2.1224 \u003c/li\u003e\n\u003cli\u003eJohnston, H., Eaton, M., Henry, I., Deeley, E.-M., \u0026amp; Parsons, B. N. (2025). Discovering how students use generative artificial intelligence tools for academic writing purposes. \u003cem\u003eJournal of Learning Development in Higher Education\u003c/em\u003e, (34). \u003c/li\u003e\n\u003cli\u003eJohnston, H., Wells, R. F., Shanks, E. M., Boey, T., \u0026amp; Parsons, B. N. (2024). Student perspectives on the use of generative artificial intelligence technologies in higher education. \u003cem\u003eInternational Journal for Educational Integrity, 20\u003c/em\u003e, Article 2. https://doi.org/10.1007/s40979-024-00149-4 \u003c/li\u003e\n\u003cli\u003eKarahan Adalı, G., \u0026amp; Bilgili, A. (2025). Generative AI in higher education: Students\u0026rsquo; perspectives on adoption, ethical concerns, and academic impact. \u003cem\u003eActa Infologica, 9\u003c/em\u003e(1), 147\u0026ndash;166. https://doi.org/10.26650/acin.1670197\u003c/li\u003e\n\u003cli\u003eKasneci, E., Sessler, K., K\u0026uuml;chemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., G\u0026uuml;nnemann, S., H\u0026uuml;llermeier, E., Peters, J., \u0026amp; Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. \u003cem\u003eLearning and Individual Differences, 103,\u003c/em\u003e 102274. https://doi.org/10.1016/j.lindif.2023.102274\u003c/li\u003e\n\u003cli\u003eKhalifa, M., \u0026amp; Albadawy, M. (2024). Using artificial intelligence in academic writing and research: An essential productivity tool. \u003cem\u003eComputer Methods and Programs in Biomedicine Update\u003c/em\u003e, \u003cem\u003e5\u003c/em\u003e, 100145.\u003c/li\u003e\n\u003cli\u003eKim, J., Lee, S.-S., Detrick, R., Wang, J., \u0026amp; Li, N. (2025). Students\u0026ndash;generative AI interaction patterns and its impact on academic writing. \u003cem\u003eJournal of Computing in Higher Education\u003c/em\u003e. Advance online publication. https://doi.org/10.1007/s12528-025-09444-6 \u003c/li\u003e\n\u003cli\u003eKim, J., Yu, S., Detrick, R., \u0026amp; Li, N. (2025). Exploring students\u0026rsquo; perspectives on generative AI-assisted academic writing. \u003cem\u003eEducation and Information Technologies, 30\u003c/em\u003e, 1265\u0026ndash;1300. https://doi.org/10.1007/s10639-024-12878-7 \u003c/li\u003e\n\u003cli\u003eKofinas, A. K., Tsay, C.-H., \u0026amp; Pike, D. (2025). The impact of generative AI on academic integrity of authentic assessments within a higher education context. \u003cem\u003eBritish Journal of Educational Technology, 00\u003c/em\u003e, 1\u0026ndash;28. https://doi.org/10.1111/bjet.13585 \u003c/li\u003e\n\u003cli\u003eLevac, D., Colquhoun, H., \u0026amp; O\u0026rsquo;Brien, K. K. (2010). Scoping studies: Advancing the methodology. \u003cem\u003eImplementation Science, 5\u003c/em\u003e(69). https://doi.org/10.1186/1748-5908-5-69\u003c/li\u003e\n\u003cli\u003eMaphoto, K. B., Sevnarayan, K., Mohale, N. E., Suliman, Z., Ntsopi, T. J., \u0026amp; Mokoena, D. (2024). Advancing students\u0026rsquo; academic excellence in distance education: Exploring the potential of generative AI integration to improve academic writing skills. \u003cem\u003eOpen Praxis, 16\u003c/em\u003e(2), 142\u0026ndash;159. https://doi.org/10.55982/openpraxis.16.2.649 \u003c/li\u003e\n\u003cli\u003eMediana Jr., N. L., Funa, A. A., \u0026amp; Dio, R. V. (2025). Effectiveness of inquiry-based learning (IbL) on improving students\u0026rsquo; conceptual understanding in science and mathematics: A meta-analysis. \u003cem\u003eInternational Journal of Education in Mathematics, Science and Technology, 13\u003c/em\u003e(2), 532\u0026ndash;552. https://doi.org/10.46328/ijemst.4769\u003c/li\u003e\n\u003cli\u003eMeyer, J. G., Urbanowicz, R. J., Martin, P. C., O\u0026rsquo;Connor, K., Li, R., Peng, P. C., ... \u0026amp; Moore, J. H. (2023). ChatGPT and large language models in academia: opportunities and challenges. \u003cem\u003eBioData mining\u003c/em\u003e, \u003cem\u003e16\u003c/em\u003e(1), 20.\u003c/li\u003e\n\u003cli\u003eMo, Z., \u0026amp; Crosthwaite, P. (2025). Exploring the affordances of generative AI large language models for stance and engagement in academic writing. \u003cem\u003eJournal of English for Academic Purposes, 75\u003c/em\u003e, Article 101499. https://doi.org/10.1016/j.jeap.2025.101499\u003c/li\u003e\n\u003cli\u003eMoorhouse, B. L., Wan, Y., Wu, C., Wu, M., \u0026amp; Ho, T. Y. (2025). Generative AI tools and empowerment in L2 academic writing. \u003cem\u003eSystem, 133\u003c/em\u003e, Article 103779. https://doi.org/10.1016/j.system.2025.103779 \u003c/li\u003e\n\u003cli\u003eNelson, A. S., Santamar\u0026iacute;a, P. V., Javens, J. S., \u0026amp; Ricaurte, M. (2025). Students\u0026rsquo; perceptions of generative artificial intelligence (GenAI) use in academic writing in English as a foreign language. \u003cem\u003eEducation Sciences, 15\u003c/em\u003e(5), 611. https://doi.org/10.3390/educsci15050611 \u003c/li\u003e\n\u003cli\u003eNguyen, A., Hong, Y., Dang, B., \u0026amp; Huang, X. (2024). Human\u0026ndash;AI collaboration patterns in AI-assisted academic writing. \u003cem\u003eStudies in Higher Education, 49\u003c/em\u003e(5), 847\u0026ndash;864. https://doi.org/10.1080/03075079.2024.2323593 \u003c/li\u003e\n\u003cli\u003eOpenAI. (2022, November 30). \u003cem\u003eIntroducing ChatGPT.\u003c/em\u003e https://openai.com/blog/chatgpt\u003c/li\u003e\n\u003cli\u003ePeters, M. D. J., Marnie, C., Tricco, A. C., Pollock, D., Munn, Z., Alexander, L., McInerney, P., Godfrey, C. M., \u0026amp; Khalil, H. (2020). Updated methodological guidance for the conduct of scoping reviews. \u003cem\u003eJBI Evidence Synthesis, 18\u003c/em\u003e(10), 2119\u0026ndash;2126. https://doi.org/10.11124/JBIES-20-00167 \u003c/li\u003e\n\u003cli\u003eRamallosa, M., Funa, A. A., Geron, C. A., Ibardaloza, R. P., \u0026amp; Prudente, M. S. (2022). Meta-analysis on the effectiveness of argument-based learning on students\u0026rsquo; conceptual understanding. In \u003cem\u003eProceedings of the 2022 13th International Conference on E-Education, E-Business, E-Management, and E-Learning (IC4E \u0026rsquo;22)\u003c/em\u003e (pp. 128\u0026ndash;133). ACM. https://doi.org/10.1145/3502485.3502495 \u003c/li\u003e\n\u003cli\u003eRodafinos, A. (2025). The integration of generative AI tools in academic writing: Implications for student research. \u003cem\u003eSocial Education Research, 6\u003c/em\u003e(2). https://doi.org/10.37256/ser.6220256517\u003c/li\u003e\n\u003cli\u003eStanford, J. (2025, March 17). How disruptive is AI in students\u0026rsquo; academic writing in higher education? \u003cem\u003eStudies in Technology Enhanced Learning, 4\u003c/em\u003e(1). https://doi.org/10.21428/8c225f6e.0390b853 \u003c/li\u003e\n\u003cli\u003eTricco, A. C., Lillie, E., Zarin, W., O\u0026rsquo;Brien, K. K., Colquhoun, H., Levac, D., Moher, D., Peters, M. D. J., Horsley, T., Weeks, L., Hempel, S., Akl, E. A., Chang, C., McGowan, J., Stewart, L., Hartling, L., Aldcroft, A., Wilson, M. G., Garritty, C., \u0026hellip; Straus, S. E. (2018). PRISMA extension for scoping reviews (PRISMA‑ScR): Checklist and explanation. \u003cem\u003eAnnals of Internal Medicine, 169\u003c/em\u003e(7), 467\u0026ndash;473. https://doi.org/10.7326/M18-0850\u003c/li\u003e\n\u003cli\u003evan Niekerk, J., Delport, P. M. J., \u0026amp; Sutherland, I. (2025). Addressing the use of generative AI in academic writing. \u003cem\u003eComputers and Education: Artificial Intelligence, 8\u003c/em\u003e, Article 100342. https://doi.org/10.1016/j.caeai.2024.100342 \u003c/li\u003e\n\u003cli\u003eVygotsky, L. S. (1978). \u003cem\u003eMind in society: The development of higher psychological processes.\u003c/em\u003e Harvard University Press.\u003c/li\u003e\n\u003cli\u003eWang, L., \u0026amp; Ren, B. (2024). Enhancing academic writing in a linguistics course with generative AI: An empirical study in a higher education institution in Hong Kong. \u003cem\u003eEducation Sciences, 14\u003c/em\u003e, 1329. https://doi.org/10.3390/educsci14121329 \u003c/li\u003e\n\u003cli\u003eWatermeyer, R., Phipps, L., Lanclos, D., \u0026amp; Knight, C. (2024). Generative AI and the automating of academia. \u003cem\u003ePostdigital Science and Education, 6\u003c/em\u003e, 446\u0026ndash;466. https://doi.org/10.1007/s42438-023-00440-6 \u003c/li\u003e\n\u003cli\u003eYusuf, A., Pervin, N., \u0026amp; Rom\u0026aacute;n-Gonz\u0026aacute;lez, M. (2024). Generative AI and the future of higher education: A threat to academic integrity or reformation? Evidence from multicultural perspectives. \u003cem\u003eInternational Journal of Educational Technology in Higher Education, 21\u003c/em\u003e, Article 21. https://doi.org/10.1186/s41239-024-00453-6 \u003c/li\u003e\n\u003cli\u003eZamorano, C. (2025). Enhancing academic writing in English language education through generative AI integration. \u003cem\u003eResearch Studies in English Language Teaching and Learning, 3\u003c/em\u003e(3), 424\u0026ndash;447. https://doi.org/10.62583/rseltl.v3i3.87 \u003c/li\u003e\n\u003cli\u003eZhang, J. S., Yoon, C., Williams, D. K. A., \u0026amp; Pinkas, A. (2024). Exploring the usage of ChatGPT among medical students in the United States. \u003cem\u003eJournal of Medical Education and Curricular Development, 11,\u003c/em\u003e 1\u0026ndash;7. https://doi.org/10.1177/23821205241264695\u003c/li\u003e\n\u003cli\u003eZhao, X., Cox, A., \u0026amp; Chen, X. (2025). The use of generative AI by students with disabilities in higher education. \u003cem\u003eThe Internet and Higher Education, 66\u003c/em\u003e, Article 101014. https://doi.org/10.1016/j.iheduc.2025.101014\u003c/li\u003e\n\u003cli\u003eZizka, L. (2025). \u0026ldquo;It looks good enough\u0026rdquo;: Recognizing the quality of generative AI output in academic writing tasks in higher education. \u003cem\u003eJournal of Hospitality \u0026amp; Tourism Education\u003c/em\u003e. Advance online publication. https://doi.org/10.1080/10963758.2025.2496663\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Academic writing, ChatGPT, GAI, higher education, large language models, scoping review","lastPublishedDoi":"10.21203/rs.3.rs-7440784/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7440784/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eGenerative artificial intelligence (GenAI) is reshaping academic writing in higher education faster than institutions can develop evidence-informed guidance, leaving practice ahead of proof. To clarify what is happening and where benefits and risks cluster, the researchers conducted a scoping review structured by a Population\u0026ndash;Concept\u0026ndash;Context (PCC) frame and aligned with PRISMA-ScR procedures. Peer-reviewed, English-language empirical studies published from 2024 through Q2 2025 in higher-education settings were included, and findings were synthesized via convergent integration that juxtaposed quantitative distributions with qualitative themes. A total of 25 studies met criteria. Across populations and contexts, GenAI was most often positioned as assistive scaffolding across the planning-to-revision span of writing; reported benefits concentrated on organization, fluency, efficiency, and language support (notably for multilingual writers). Recurrent risks included hallucinations and unreliable or fabricated citations, inconsistent disclosure or attribution, and overreliance when use was unscaffolded; the limited reliability of AI-detection tools complicated integrity judgments. Context shaped practice: clearer policies and better access supported more constructive use, while the evidence base skews toward English-medium, well-resourced institutions and relies heavily on short-term or proxy outcomes. By integrating counts and themes within a PCC frame, this review offers an up-to-date evidence map that distinguishes where benefits reliably cluster (process-level supports) and where risks persist (source work and attribution), while surfacing salient gaps (faculty/postgraduate cohorts and Global South contexts). Overall, the pattern supports an assistive, not substitutive stance in which GenAI complements\u0026mdash;rather than replaces\u0026mdash;human judgment in argument construction, source interrogation, and synthesis.\u003c/p\u003e","manuscriptTitle":"Generative Artificial Intelligence (GenAI) for Academic Writing in Higher Education: A Scoping Review of Applications, Challenges, and Implications","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-26 05:46:45","doi":"10.21203/rs.3.rs-7440784/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"33e7b171-aee6-4180-beaf-6a0a4b95aee1","owner":[],"postedDate":"August 26th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":53606021,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2025-08-26T05:46:45+00:00","versionOfRecord":[],"versionCreatedAt":"2025-08-26 05:46:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7440784","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7440784","identity":"rs-7440784","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}