The research team surveyed 150 lecturers drawn from multiple Cambodian higher education institutions using a 55-item online questionnaire administered through Google Forms. Cambodia in 2025 had 194 higher education institutions spread across 25 provinces and the capital, employing 22,214 higher education personnel, of whom less than 12 percent held PhD degrees. The respondents skewed heavily male, with only 11.33 percent female, and 82 percent held master’s degrees while 16.67 percent held doctorates. About 27 percent earned their highest degrees overseas, teaching experience ranged from under five years to over two decades, and roughly half taught part-time. The questionnaire covered demographic information, experiences with AI tools, perceptions of AI use, perceived challenges, perceived opportunities, and recommendations, with 38 Likert-scale items yielding an overall Cronbach’s alpha of 0.93, indicating excellent internal consistency.

The descriptive results reveal a striking pattern of tool adoption. The three most frequently used AI tools were Google Translate at 70.7 percent, Grammarly at 66 percent, and ChatGPT at 62 percent, followed by Quillbot at 33.3 percent, Gemini at 32 percent, and Copilot at 20 percent. Only 7.3 percent of participants reported never having used any AI tool for teaching and learning purposes. The reasons for use were overwhelmingly pedagogical: 60.3 percent used AI to create teaching materials, 58.9 percent to generate or brainstorm teaching activities, 54.1 percent to translate languages, and 51.4 percent to understand and simplify technical terms, concepts, or theories. Nearly half used AI to design quizzes and homework, and about 40 percent used it to develop course outlines or syllabi.

Overall, the lecturers held positive perceptions of AI use in higher education, with a mean score of 3.69 on a five-point scale. They agreed that using AI tools is a wise choice both now and in the future, that AI is good for learning and teaching purposes, and that AI has potential roles in supporting academic writing, personalizing higher education, and automating administrative tasks. Their agreement levels dropped notably when it came to trusting AI to offer fair evaluation of student performance, with a mean of 3.25, and to beliefs that AI use might contribute to memory loss, with a mean of 3.17. This gap between general optimism and specific skepticism about evaluation and cognitive effects is one of the study’s most telling findings.

When asked about challenges, the lecturers identified several serious concerns. They reported high agreement with the lack of human interaction, limited trust in AI systems, lack of proper training and technical support, and technical issues related to limited or unreliable internet access and the complexity of AI systems. They also agreed that AI use could breach academic integrity, add expenses, raise data security and privacy issues, and foster overreliance among lecturers and students. Moderate agreement was recorded for concerns about negative effects on critical thinking and problem-solving skills, inaccurate or false AI outputs, and biased learning assessments. These concerns closely mirror those reported in prior international literature, suggesting that the challenges Cambodian lecturers face are not isolated but reflect globally recognized issues surrounding AI adoption in higher education.

The lecturers strongly agreed with the opportunities AI offers, with an overall mean of 3.91. Their highest agreement levels concerned saving time and energy, supporting teaching and learning processes, improving the quality of teaching and learning in the digital age, developing materials for teaching purposes, and helping to simplify and explain terminologies, concepts, and theories. They also agreed that AI could support academic research design and development, encourage innovation in learning assessments, and increase professional development opportunities. The authors suggest that the emphasis on saving time and energy may stem from the fact that most Cambodian lecturers hold part-time teaching status while engaging in full-time employment elsewhere, making AI tools a way to enhance productivity while reducing academic workloads.

The inferential analyses using the Mann–Whitney U test revealed statistically significant differences in perceptions and perceived challenges across demographic, educational, and professional characteristics. Female respondents reported higher agreement than male respondents that AI supports academic writing, with means of 4.31 versus 3.82. Respondents who graduated from foreign institutions expressed greater concern about the ethical implications of AI and about the reliability and accuracy of AI than those who graduated from Cambodian institutions. Respondents from public universities reported significantly greater concern about the lack of human interaction associated with AI use than respondents from private universities. Full-time lecturers reported higher levels of concern about the lack of training and technical support for effective AI adoption, as well as technical issues related to AI implementation, than part-time lecturers. No other two-group comparisons or Kruskal–Wallis tests with more than two groups yielded statistically significant differences.

These differences suggest that individual experiences and professional contexts shape how lecturers perceive the benefits, risks, and challenges of AI use. The authors note that female lecturers’ more favorable perceptions of AI for academic writing may be associated with differences in perceived usefulness and technology-related experiences, consistent with technology acceptance research suggesting that gender can influence technology perceptions and use. The higher ethical awareness reported by graduates of foreign institutions may reflect greater exposure to different academic cultures and international discussions of academic integrity and responsible AI use. The higher perceived challenges among full-time lecturers may relate to their broader responsibilities in teaching, assessment, research, and administration, which provide greater exposure to the practical and ethical implications of AI. The findings suggest that a uniform approach to AI policy and training may not adequately address lecturers’ diverse needs and concerns.

When asked for recommendations on effectively integrating AI into higher education, the lecturers selected their top five from ten options. The most frequently selected was offering sufficient training and technical support to both lecturers and students, chosen by 86 percent of participants. This was followed by choosing AI tools that meet educational needs and purposes at 68.7 percent, establishing clear guidelines, protocols, or policies for using AI tools in teaching and learning at 54 percent, balancing AI use and human interaction at 52.7 percent, and investing more in improving digital infrastructure, tools, and resources of higher education institutions at 52.7 percent. The least frequently selected recommendations were the need to review and revise assessment policy and practices to cover AI use, establish an institutional committee to address AI concerns, and address the digital divide among students and lecturers.

The study makes a contextual contribution to the growing body of literature on AI use in education within the Global South, providing empirically valuable insights into the opportunities and constraints shaping AI integration in resource-constrained higher education systems. The authors emphasize several implications, including the need for AI training provision covering practical applications such as AI-assisted lesson planning, assessment design, academic writing support, research assistance, prompt engineering, and ethical issues such as plagiarism, data privacy, and responsible AI use. They also call for enhanced support structures including access to digital devices, licensed AI platforms, technical support services, and stable internet connectivity, as well as the establishment of AI guidelines or policies to govern how AI can be responsibly, ethically, and effectively used. The authors acknowledge limitations, including the use of convenience sampling, the relatively small sample size of 150, the underrepresentation of female participants, and the self-reported nature of the data, and they suggest that future research employ a variety of data collection methods, include larger and more representative samples, and utilize mixed-methods approaches to ensure triangulation of data.

Subject of Research: Cambodian university lecturers' experiences and perceptions of AI use in higher education

Article Title: A survey study of Cambodian university lecturers’ experiences and perceptions of AI use

Article References: Heng, K., Sok, S., Sol, K., & Sor, R. (2026). A survey study of Cambodian university lecturers’ experiences and perceptions of AI use. Discover Artificial Intelligence, 6(1), Article 1241. https://doi.org/10.1007/s44163-026-02313-w

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02313-w

Keywords: artificial intelligence, higher education, Cambodia, university lecturers, ChatGPT, Google Translate, Grammarly, survey research, academic integrity, digital divide, teacher training, Global South

The study, conducted by Sayed M. Ismail, Faris Allehyani, Khaled Ahmed Abdel-Al Ibrahim, and Mohamad Ahmad Saleem Khasawneh, recruited 60 undergraduate English majors aged 20 to 29 at a Saudi university, all screened as intermediate-level learners using the Oxford Quick Placement Test. Participants came from two intact sections of the same academic writing course taught by the same experienced instructor. One section of 30 students received twelve weeks of process-based writing instruction supplemented with Grammarly during drafting and revision; the other 30 completed an identical syllabus, task sequence, and revision routine without any AI-generated feedback. Crucially, the researchers standardized nearly everything else: the same ten argumentative writing tasks, the same three-draft cycle per task, the same deadlines, testing conditions, and assessment rubrics.

The theoretical scaffolding behind the experiment is as interesting as the results. Rather than treating Grammarly as a generic digital add-on, the authors framed automated feedback through three interlocking lenses: self-determination theory, which holds that learners persist when tools support competence, autonomy, and meaningful participation; social-cognitive and self-regulated learning theory, which emphasizes goal setting, monitoring, and strategic adjustment; and feedback theory, which insists that feedback only has value when learners understand it, judge its relevance, and convert it into revision action. In this model, Grammarly’s suggestions act as revision cues that can trigger a self-regulatory sequence: a learner notices recurring article errors, sets a goal to reduce them, checks later drafts, and reflects on whether the strategy worked.

The quantitative results were striking across all four measured outcomes. Using linear mixed-effects models that accounted for repeated measurements within each learner, the team found that the Grammarly-supported group improved significantly more than the control group from pre-test to post-test. Writing quality, scored on a 20-point rubric covering grammatical and lexical accuracy, fluency and coherence, and text structure, rose by 5.42 points in the experimental group versus 1.47 points in the control group, a differential gain of 3.95 points with a large standardized effect size of 1.77. Motivation climbed 13.50 points against 4.40, feedback utilization rose 4.44 points against 0.93, and self-regulated learning strategies jumped 13.90 points against 2.23. All four time-by-condition interactions were statistically significant.

Behavioral evidence from draft-to-draft comparisons reinforced the test scores. Trained raters, working independently and achieving strong inter-rater agreement, scored how accurately, completely, deeply, and consistently learners incorporated available feedback into successive drafts. In the Grammarly group, 77 percent reached the highest rubric level for accurate incorporation of feedback without distorting intended meaning, compared with 43 percent of controls. Eighty percent of the AI-supported students addressed all feedback points versus 60 percent of controls, 70 percent made revisions that went beyond surface-level correction versus 50 percent, and 85 percent maintained improvements across later drafts versus 65 percent. The pattern suggests that immediate automated cues did not merely decorate the texts; they changed what students actually did between drafts.

Semi-structured interviews with the experimental group added texture and caution in equal measure. Participants reported that Grammarly was easy to access and use outside class, and that it sharpened their awareness of grammar, punctuation, sentence structure, and word choice. Immediate feedback supported revision while writing rather than after the fact. But the students were far from uncritical. Some found suggestions overly general, contextually inappropriate, or inadequate for complex sentence structures, and they consistently wanted more help with content development and argument organization than any sentence-level tool can provide. Notably, they credited their teacher with helping them decide when to accept, modify, or reject automated suggestions, underscoring that writer agency remained central to the intervention’s design.

That design deliberately constrained what Grammarly could do. Students used the platform’s grammar, spelling, punctuation, clarity, concision, word-choice, and style functions through its web editor and Microsoft Word integration, but generative rewriting, text-generation prompts, plagiarism detection, and citation generation were switched off. Before the intervention began, the experimental group received two 45-minute orientation sessions teaching them to evaluate suggestions against their intended meaning rather than accept them automatically. The researchers were explicit that the treatment targeted automated written feedback and learner-initiated revision, not AI-generated composition, a distinction that matters as institutions grapple with how generative AI reshapes academic writing.

The authors are equally candid about the limits of their evidence. Because participants belonged to pre-existing class sections, individual randomization was impossible, and instructional condition was perfectly aligned with class membership, meaning class-level confounding cannot be ruled out. The experimental group actually started with slightly lower scores on all four outcomes, which the longitudinal models incorporated but which cannot substitute for randomization. Novelty is another live concern: learners knew they were using a new AI tool, and prior research has shown that perceived novelty can inflate motivational responses. The motivation and self-regulation findings rest on self-report questionnaires, so the researchers interpret them alongside the harder performance and behavioral data, and they caution that no delayed post-test was administered to check whether gains persisted.

Despite those caveats, the convergence across evidence streams is what gives the study its force. Writing quality and feedback utilization, measured by independent raters on matched prompts without platform assistance, moved in the same direction as the self-reported motivation and self-regulation gains, and the interview accounts of immediate revision and contextual judgment complemented the numbers rather than contradicting them. The study was not designed to test whether feedback use statistically mediates writing improvement, so the proposed pathway from automated cues to self-regulation to better texts remains a theoretically informed interpretation rather than a demonstrated causal chain. Future work, the authors argue, should employ randomized designs, larger and more diverse samples, delayed assessments, and explicit mediation models, while comparing Grammarly with other AI writing tools.

The practical takeaway resists both hype and dismissal. Automated feedback, embedded in a disciplined process-writing routine with teacher guidance, appears to give second-language writers more opportunities to notice problems, revise strategically, and experience visible progress, which in turn may feed motivation and self-regulation. But the same students who benefited most also flagged the tool’s blind spots at the level of argument, audience, and organization, the dimensions where human instruction remains irreplaceable. As universities worldwide debate the role of AI in the writing classroom, this Saudi experiment suggests a middle path: treat platforms like Grammarly as supplementary revision resources within structured instruction, teach learners to judge suggestions critically, and keep teachers firmly in the loop for everything that lies beyond the sentence.

Subject of Research: Effects of Grammarly-supported AI writing instruction on EFL learners' writing quality, motivation, feedback utilization, and self-regulated learning strategies

Article Title: Grammarly-supported AI writing instruction in a Saudi EFL context: Motivation, writing quality, feedback utilization, and self-regulated learning strategies

Article References: Ismail, S. M., Allehyani, F., Ahmed Abdel-Al Ibrahim, K., & Khasawneh, M. A. S. (2026). Grammarly-supported AI writing instruction in a Saudi EFL context: Motivation, writing quality, feedback utilization, and self-regulated learning strategies. Heliyon, 12(15), Article e45557. https://doi.org/10.1016/j.heliyon.2026.e45557

Image Credits: AI Generated

DOI: 10.1016/j.heliyon.2026.e45557

Keywords: Grammarly, AI feedback, EFL writing, self-regulated learning, writing motivation, automated writing evaluation, Saudi Arabia, process writing, feedback utilization, quasi-experimental study, second language acquisition, Heliyon