Assessing Generative Artificial Intelligence Tools Usage Guidelines at the Top 200 Universities Worldwide

1. Introduction

Generative Artificial Intelligence (GenAI) has emerged as a groundbreaking facet of artificial intelligence (AI). It is characterized by its ability to analyze vast datasets and produce novel content across various media, including text, images, music, and code. This technology is not merely a reflection of existing data; instead, it employs advanced algorithms to generate new, innovative outputs in response to specific inputs or prompts [1]. Notable applications such as ChatGPT [2], Google Gemini [3], and DeepSeek [4], among many other GenAI tools, have become an integral part of today’s online experience, enhancing both search capabilities and content creation [5].

In academic research, the growing adoption of GenAI in online search engines and platforms is reshaping how we engage with digital content. Tools like Google AI Overviews, Google Scholar Labs, Scopus AI, and Clarivate’s Web of Science research assistant utilize AI to optimize the user experience. These platforms generate tailored search results based on user queries, making research and information retrieval more efficient and intuitive [6,7].

Furthermore, there is an increasing trend among university students and staff towards Edge AI solutions (desktop or mobile). These localized, offline applications help users better protect their data privacy and uphold copyright protections. By processing data closer to where it is generated or stored, Edge AI enables individuals and institutions to retain control over sensitive information while still leveraging advanced technologies. This shift reflects a growing preference for applications that prioritize privacy and security in an ever-evolving digital landscape [8,9].

The rise in the use of GenAI tools in higher education is transforming how students learn and enhance their educational experiences. However, these technologies raise concerns regarding their ethical, legal, and appropriate application. Therefore, it is crucial to establish clear, comprehensive, and principled guidelines for their use [10]. These guidelines provide a framework for educational institutions to develop institution-specific policies that promote consistent and ethical practices while encouraging innovation, with each institution tailoring these policies to its unique needs and context. Typically, specialized committees or administrative offices develop these guidelines and translate them into institutional policies, ensuring their effective interpretation and consistent implementation [11].

1.1. Objectives of the Study

The primary objectives of this study are as follows:

• To locate, assess, and provide a comprehensive overview of the publicly available guidelines on the websites of the top 200 universities worldwide.
• To compare our findings with those from our 2024 survey [12] for the 50 universities included in both surveys, and to observe changes over time.

Through these objectives, the study seeks to provide comprehensive insights into the evolution of GenAI guidelines in higher education.

1.2. Related Work

Ullah et al. [12] evaluated guidelines for the use of GenAI at the top 50 universities worldwide, as ranked by the Quacquarelli Symonds (QS) World University Rankings 2024. They used an expert-reviewed 24-item checklist divided into three categories: general information, usage instructions, and ethical/legal considerations. The study found that 82% of these institutions had published guidelines. All universities permitted the use of GenAI under specific conditions, such as requiring instructor approval and proper acknowledgment. A significant majority of the guidelines covered academic integrity (88%), data privacy (85%), and the verification of AI outputs (93%). However, only 20% of the survey guidelines provided explanations of AI algorithms, 32% documented specific prompts and outputs, and 34% included formal reporting mechanisms for misconduct. They concluded that while institutional responses are rapidly expanding, the current guidelines often serve as a “quick fix.” They argued for a comprehensive re-evaluation of the curriculum and for enhanced AI literacy to better prepare students and faculty for a future in which AI is deeply integrated into knowledge creation. They discussed literature published up to Q4 2024 on guidelines for the use of GenAI. Therefore, the subsequent sections will discuss selected studies published after this period.

Wang et al. [13] conducted a systematic literature review of 139 articles to analyze the “meso-level” application of GenAI in higher education classroom teaching. The findings highlighted a notable “disciplinary bias” in the adoption of GenAI, with a predominant focus on fields such as Engineering, Health and Medicine, and Language. In contrast, areas such as the basic sciences and humanities remain underrepresented and marginalized in this trend. They documented that GenAI primarily serves as a “New Subject” (virtual partner) or a “Direct Mediator” (instructional tool). However, its ability to support personalized data analysis and predict academic outcomes has not been fully explored. Many current integrations mainly target the Augmentation level of the SAMR (Substitution, Augmentation, Modification, and Redefinition) framework. However, the authors cautioned that merely substituting student thought processes can hinder critical thinking. Institutions need to progress towards the “Redefinition” level by promoting GenAI literacy. They should focus on creating data-driven, adaptive teaching strategies that emphasize real-world learning experiences.

Tong et al. [10] analyzed the contents of 28 journal articles published between 2020 and 2024 to evaluate higher education GenAI policies against a ten-element governance framework. Their findings indicated that most policies emphasized transparency (82%), privacy (71%), and fairness (71%). However, there was a critical lack of focus on technical dimensions, including interpretability (36%), explainability (32%), and robustness (18%). The study emphasized the necessity for institutions to adopt a collaborative, interdisciplinary governance model. This approach should prioritize AI literacy and technical oversight to create a responsible and equitable learning environment for students.

McDonald et al. [14] examined policy documents regarding the use of GenAI from 116 U.S. research universities. They used open coding to identify themes related to classroom integration, ethics, and privacy in the documents. The study found that most of these institutions (63%) actively encouraged the use of GenAI, with many providing resources such as sample syllabi (56%) and classroom activities (50%) to leverage the technology. Many universities discouraged the use of unreliable AI detection tools (44%). They found that guidance was predominantly focused on writing tasks, with STEM and coding applications remaining infrequent or vague. About half of the institutions addressed issues related to ethics and privacy, as well as Diversity, Equity, and Inclusion (DEI). However, these topics were often considered less important than practical implementation. The authors emphasized that adopting GenAI in educational institutions often requires significant changes to teaching methods.

An et al. [15] employed a mixed-methods approach to examine the GenAI guidelines of the top 50 U.S. universities. They manually collected 214 publicly available documents for analysis. The study identified four primary discourse topics: integration in learning and assessment, visual and multimodal media, security and ethical considerations, and academic integrity. Institutional sentiment was overwhelmingly positive across all types of universities. However, supportive language was used with faculty, while a more cautious, regulatory approach was adopted with students. Furthermore, 94% of the universities provided guidance specifically for faculty, while researchers (18%) and administrative staff (14%) received less attention. Additionally, the use of unreliable AI detection tools was discouraged. The authors concluded that institutions are placing a high priority on proactive pedagogical innovation, but they must develop flexible policies that address all stakeholders and adapt rapidly as technology evolves.

Taeihagh [16] highlighted the socio-technical risks associated with GenAI, including hallucinations, opacity, and the leakage of sensitive information. He also identified critical governance challenges such as Big Tech dominance and the “pacing problem,” in which regulation lags rapid technological innovation. The author critiqued existing technocratic solutions to governance, arguing that they are inadequate for addressing the complexities of GenAI, and highlighted the need for a fundamental shift toward proactive, adaptive, and participatory frameworks. The study concluded that sustainable governance must involve international cooperation and public engagement. Moreover, there is a need to address power imbalances and ensure the responsible, ethical integration of AI across all sectors of society.

Crompton et al. [11] utilized the Delphi technique and collective writing to establish a global consensus on the governance of GenAI in higher education. The study synthesized the perspectives of 35 international experts from 22 countries on six continents to identify the essential components of effective policy and practice. A significant majority of the panels (88%) advocated for a hybrid governance model that combines formal, authoritative policies for accountability with flexible, advisory guidelines to accommodate the rapid evolution of technology. The consultation resulted in an eight-part framework that addresses core pillars, including academic integrity, data privacy, equitable access, and GenAI literacy. The experts recommended establishing dedicated GenAI committees and conducting regular policy reviews to ensure institutional relevance. The study concluded that sustainable GenAI governance must function as an interconnected institutional ecosystem. It should transition from reactive prohibition to a proactive, evidence-based approach that balances technological innovation with ethical responsibility and pedagogical excellence.

Abir and Zhou [17] conducted a thematic analysis of publicly available GenAI guidelines from the top 50 Japanese universities. They examined how cultural and pedagogical values shaped institutional governance. Their findings indicated that Japanese institutions adopted a cautious, values-driven approach, aligning GenAI use with philosophies such as “all-round education” and intellectual resilience, which prioritized the learning process and critical thinking over technological efficiency or final outcomes. Policies emphasized faculty autonomy in classroom implementation and student responsibility for ethical use. The guidelines frequently highlighted risks, such as misinformation, privacy concerns, and Western-centric bias, through the strategic use of metaphors. The authors concluded that while Western institutions often favor flexibility and rapid innovation, the Japanese model offers a deliberate, ethics-grounded framework that balances technological adoption with a commitment to holistic student development.

Azevedo et al. [18] analyzed the GenAI guidelines for the faculties of Pennsylvania State University, the University of North Carolina, and the State University of New York. They found that universities often promote AI for its potential benefits but frequently overlook critical ethical and equity concerns. These concerns include algorithmic bias, risks to intellectual property, and the disproportionate burden on faculty to manage technological transitions.

Smith et al. [19] conducted consultative case studies at two Australian universities. They proposed a four-layer strategic framework: encompassing context, development, implementation, and review to guide the responsible integration of GenAI in academic research. The study identified critical governance needs, including mandatory disclosure of AI involvement, rigorous documentation for reproducibility, and adaptation of postgraduate supervision to safeguard critical thinking.

Pérez-Pérez et al. [20] conducted a systematic scoping review of 11 studies that were published between 2022 and 2026. Their objective was to identify and synthesize transparency mechanisms for the use of GenAI in higher education assessments. They found a significant difference between what is considered standard discourse and how things operate in practice. The study found that transparency was often vaguely defined, and institutions frequently relied on unverified self-disclosure or subjective instructor judgment rather than formal evaluative criteria.

Li et al. [21] conducted a cross-national qualitative content analysis of 124 GenAI policy documents from 110 leading universities in the United States, Japan, and China to identify distinct institutional governance patterns. Using an extended Technology Acceptance Model (TAM) as an analytical lens, the authors developed the University Policy Development Framework for GenAI (UPDF-GAI) to evaluate institutional responses across five key domains: perceived usefulness, perceived risk, facilitating conditions, social influence, and self-efficacy. Their findings revealed three divergent governance logics: U.S. universities adopted a decentralized, innovation-oriented approach prioritizing faculty autonomy and peer collaboration; Japanese institutions followed a centrally coordinated, risk-aware model focused on ethics; and Chinese universities exhibited a highly centralized, institutionally directed structure that emphasized national strategic alignment. The study concluded that GenAI governance is inextricably linked to national socio-political structures and advocated the UPDF-GAI as a flexible diagnostic tool to help institutions balance technological innovation with ethical responsibility.

A study by Alba, Mcilwain, and An [22] examined how top U.S. universities are managing Generative AI by analyzing 422 policy documents. They found that most schools have strict rules against using AI for coursework and tight controls on data privacy. While teachers have some freedom in how they use AI in the classroom, the overall policies are mostly controlled by the university administration. The size of the university or its funding doesn’t seem to change these policies much. The researchers believe that universities need to move away from strict rules and adopt more flexible, community-based approaches to effectively integrate technology in education.

Ofosu-Asare [23] conducted a review of 108 research papers to examine the ethical issues surrounding educational Generative AI and identified significant gaps in governance across five key areas: protecting privacy, reducing algorithmic bias, ensuring transparency, holding individuals accountable, and maintaining academic integrity. To address these concerns, the author developed a framework that integrates concepts from four ethical approaches: deontological ethics, which emphasizes the protection of rights such as privacy and fairness; utilitarianism, which aims to enhance learning for everyone, particularly the vulnerable; virtue ethics, which focuses on honesty and critical thinking; and care ethics, which prioritizes relationships and well-being. The study concluded that Generative AI should serve to enhance, rather than replace, fundamental educational values to ensure student dignity and fairness in education.

2. Materials and Methods

We compiled a list of the top 200 universities according to the 2026 Quacquarelli Symonds (QS) University Rankings. The QS World University Rankings 2026 were chosen for this study because they are well-known and respected in evaluating universities worldwide. These rankings consider factors such as academic reputation, employer reputation, citations, and international student involvement. This makes them a good way to identify universities that are diverse and research-focused, where the use of GenAI and related policies is likely to be more developed and impactful. The QS ranking includes a diverse range of universities in its top 200 list. By concentrating on these institutions, the study becomes more manageable and emphasizes those that play a significant role in policymaking. These universities often establish guidelines that other schools may adopt.

From April 1 to April 20, 2026, we searched the public websites of these universities for GenAI guidelines to evaluate their availability. The search term utilized in this study was ‘guidelines or policies for the use of generative artificial intelligence (GenAI) tools,’ specifically targeting university websites to gather relevant information and insights. If a paragraph or more is available on the university’s website discussing the use of GenAI tools, that information is included in this study for evaluation. This study focuses on measuring coverage rather than the quality or depth of the content. Only universities that made no mention of GenAI tools on their websites were excluded from the study. Most universities had an English version of their website; however, if an English version was unavailable, Google Translate was employed to translate the non-English guidelines and policies into English. If the university had multiple guidelines from different departments, they were all evaluated together. This assessment utilized the same evaluation tool/checklist [12] (Appendix A) that was previously developed, validated, and published in our December 2024 survey of the top 50 universities worldwide. The instrument used was a simple binary measure that assessed only the availability of specific items in the guidelines. As a result, the first author conducted the survey without a second rater, as the second rater would likely give the same scores. Our 2024 survey [12] and Abir and Shigeta’s 2025 survey [24] of the top 50 Japanese universities effectively utilized the same checklist in a similar manner. We assessed the presence of all 24 checklist items, and the results are reported as frequencies and percentages. The list of 200 universities, along with links to the identified GenAI guidelines, is included in Appendix B.

3. Results

The following summary synthesizes findings from our survey of 200 university websites, conducted between 1 April 2026 and 20 April 2026, which assessed the availability and content of guidelines these institutions published on the use of GenAI tools.

3.1. General Availability of Guidelines

Of the 200 university websites examined, a significant majority, 180 universities (90%), had publicly available guidelines for the use of GenAI tools. Conversely, 20 universities (10%) did not provide any such guidance on their public-facing online platforms (Table 1). Whilst it is possible that some of these institutions maintain guidelines on private intranets, such documents were inaccessible for the purposes of this survey, and their existence could not be verified. This group includes high-ranking institutions such as Université PSL, Paris, France (Rank 28) and Zhejiang University, Hangzhou, China (Rank 49). Geographically, these institutions represent a diverse range of countries, including China, Japan, France, Canada, Argentina, Korea, Germany, Italy, Chile, Spain, Indonesia, and Saudi Arabia.

3.2. Trends in General Information and Currency

Release/update date: This is provided by 69% of universities. Among the 180 universities with published guidelines, there is a clear trend towards keeping information up to date. In 2026 alone, 33 guidelines were issued or updated, followed by 49 updated in 2025, 26 in 2024, and 16 in 2023 (Table 2). The most common general information items identified are detailed below:

Examples of GenAI tools: These are present in 84% of the guidelines reviewed (152 institutions). A total of 102 GenAI tools were mentioned in the surveyed guidelines. The most frequently cited tools were ChatGPT (134 mentions), Microsoft Copilot (76 mentions), and Google Gemini/Bard (72 mentions). Other tools included Claude (30 mentions), DALL-E (20 mentions), and Midjourney (18 mentions). Google Notebook LM was mentioned 17 times, Grammarly 14 times, Microsoft Bing 12 times, and Zoom AI Companions nine times. Adobe Firefly, Perplexity, and Elicit were each mentioned eight times. The Chinese model, DeepSeek, was indicated in the guidelines of seven universities. The remaining 89 GenAI tools were mentioned fewer than seven times. Many universities recommend using only university-licensed GenAI tools for educational, research, and administrative purposes. Notably, 37 institutions mentioned Microsoft Copilot, while seven mentioned Google Gemini as a licensed GenAI tool available at the universities. A smaller number highlighted other GenAI tools. The focus on licensed tools stems from their improved security features. These GenAI tools do not use personal information for learning and training when users log in with university credentials.

Guideline issuing authority: The issuing authority was identified in 119 (66%) of the guidelines. In most instances (46 universities), these guidelines were developed by academic bodies or centers focused on teaching, learning, and innovation within the institutions. The names of these bodies varied, including titles such as the Center for Teaching and Learning, the Center for Teaching, Learning and Innovation, the Teaching and Learning Hub, and the Center for Excellence in Teaching, Learning, and Innovation, among others. At 24 universities, the Information Technology and Communication departments issued the guidelines. At 17 universities, dedicated AI units or skills centers focused on blended, online, or digital learning took on this responsibility. Furthermore, in 15 universities, the guidelines were released by the office of the President, Vice President, Vice Provost, or Deputy Vice Chancellor for Academic and Educational Affairs. Other departmental contributions included guidelines issued by the library at seven universities, by research offices at five universities, by departments focused on quality assurance, enhancement, or support at three universities, and by the marketing and communication office and the strategic development office, each at one university.

Universities, in addition to the guidelines set by their main governing bodies, have various departments that establish their own specific guidelines. For example, libraries often create Library Guides (LibGuides) to acknowledge and reference generative AI (GenAI) tools. The Information Technology department provides guidelines related to data security, privacy, and the accessibility of GenAI tools. The research department focuses on the application of AI in academic research, while the examination department addresses assessment issues concerning the use of GenAI. Moreover, various departments, disciplines, or fields have developed their own distinct guidelines.

Technical explanations in the surveyed guideline documents are rare; only 19% of guidelines explain how generative AI algorithms work. Furthermore, 54% of the institutions in our sample provide specific contact information for further guidance, and 56% of them clearly define the objectives and scope of their policies (Table 2).

3.3. Instructional Guidance and Practical Application

A consistent trend across all 180 institutions is the explicit permission to use GenAI tools, with 100% of guidelines permitting their use under certain conditions (Table 3). However, the depth of instructional detail varies significantly:

Limitations and Suitability: 89% of guidelines outline instances where GenAI is unsuitable.

Classroom Integration: 83% provide strategies for use in assessments.

Tool Identification: 81% require users to provide details of the specific GenAI tool used (e.g., name, version, and date).

Accountability: 79% require instructor approval to use the tool.

Usage Context: 77% specify the domains for GenAI utilization.

There is a noticeable lack of transparency regarding the practical process of using AI. Only 38% of guidelines require users to document the specific prompts provided to the tool. Furthermore, only 27% offer guidance on adapting or transforming GenAI output into final work (Table 3).

3.4. Ethical and Legal Considerations

Universities have prioritized academic conduct over legal compliance, with the primary focus directed towards academic integrity, data privacy, and verification of generated outputs:

Academic Integrity: 93% of the guidelines address GenAI-related misconduct.

Data Privacy: 85% discuss data privacy and security.

Verification: 79% instruct users to evaluate and verify GenAI-generated content.

While citation is a pillar of academic work, only 69% of guidelines include specific instructions for referencing GenAI outputs. Legal compliance is mentioned in about half of the documents (56%). Interestingly, due to the rise of detection software, 67% of universities provided instructions on using AI detection tools. The least common feature is a formal reporting mechanism for GenAI-related issues, found in only 18% of the surveyed documents (Table 4).

3.5. Comparison of the 2024 and 2026 Surveys

All 50 universities in our 2024 assessment [12] are also among the 200 universities surveyed in the current study. This presents an opportunity to examine how the policies of these 50 universities (a subgroup within the 200) have changed since 2024, including how many have been updated and what new material has been added.

In our initial review, only 41 of the 50 universities had publicly available guidelines. However, our 2026 survey indicates that 48 universities now have GenAI guidelines available on their websites. A comparison of specific items is presented in Table 5, which contrasts the findings from the 2024 and 2026 surveys. Overall, the latest guidelines from this subgroup of 50 universities provide greater coverage across almost all surveyed dimensions than the corresponding guidelines from the same universities that we surveyed in 2024.

4. Discussion

Our findings show that universities worldwide (180 of the 200 top-ranked universities surveyed) recognize the need for GenAI guidelines and AI literacy to ensure the appropriate legal and ethical use of these tools in academic research, teaching, and learning. A comparison with our 2024 data for the subset of 50 universities re-evaluated in 2026 shows progress in guideline coverage, reflecting institutions’ rapid responses to the evolving AI landscape.

Justification of the 24-item checklist: The checklist used in this study was developed by the authors (M.U.) in our earlier 2024 survey [12] and was reviewed for face validity by colleagues with expertise in educational technology and academic integrity. We acknowledge that it is not derived from a single, externally validated psychometric instrument. However, its items map directly onto the eight-part governance framework proposed by Crompton et al. (2026) [11], covering academic integrity, data privacy, equitable access, and GenAI literacy, and onto the five domains of the University Policy Development Framework for GenAI (UPDF-GAI) reported by Li et al. (2026) [21]. We therefore argue that the checklist has strong construct alignment with the most recent global consensus frameworks, even though it was purpose-built for rapid policy content analysis rather than for attitudinal measurement. Our study found that, although most top-tier universities have responded to the GenAI trend by establishing institutional frameworks, the focus remains heavily centered on academic integrity and general tool awareness. A secondary tier of instructional detail, including prompt documentation, output adaptation, and legal compliance, remains unaddressed by more than half of the institutions surveyed.

An analysis of 24 checklist items across various categories shows that leading universities worldwide prioritize essential governance aspects. They place a strong emphasis on academic integrity (93%), tool limitations (89%), and data privacy (85%). In contrast, secondary technical and operational aspects, such as algorithmic explanations (19%), details regarding the adoption and utilization of GenAI output (27%), and mechanisms for reporting misconduct (18%), are largely overlooked. Additionally, cross-national studies in the broader literature indicate that regional governance patterns vary with socio-political structures. For example, the United States typically adopts decentralized, innovation-oriented policies, while Japan tends to follow risk-aware, values-driven frameworks. China, on the other hand, employs centralized, state-aligned models [17,21].

Only a small number of the reviewed guidelines offered insights into GenAI algorithms, including how AI systems function and their inherently probabilistic nature. This nature underlies many limitations of these systems, including hallucinations, incomplete responses, and non-reproducible outputs. Furthermore, there were limited instructions on documenting prompts and outputs, using AI detection tools, and reporting misconduct. These findings suggest that while universities have made progress in developing guidance for GenAI, many frameworks still fall short in supporting the practical and responsible application of these technologies.

Taylor and LaCroix [25] examined the tension between institutional promotion of GenAI and the enforcement of academic integrity standards in higher education. They argued that whether the use of GenAI constitutes misconduct depends on a prior question about the university’s purpose and function. Drawing on historical and rhetorical analyses of mission statements from leading university groups, including the Canadian U15, the UK Russell Group, and the US Ivy League, the authors identified a disconnect between institutional ideals of innovation and excellence and the practical implementation of GenAI policies. They argued that this inconsistency creates moral and institutional ambiguity, whereby students may be held accountable for behaviors shaped by broader institutional practices. The study concluded that universities cannot credibly enforce academic integrity standards in the age of GenAI without first ensuring coherence between their stated missions, pedagogical approaches, and adoption of emerging technologies.

While AI detection tools may seem like a solution for some universities seeking to police academic integrity, their reliability in academic settings has repeatedly been shown to be limited [26]. Indiana University’s Kelley School of Business, for example, recently updated its AI Playbook [27], clearly stating that AI detection tools are not approved for use by faculty. These include tools such as GPTZero, Turnitin AI Detection, and Originality AI. The rationale behind this decision is that such tools can produce false positives and false negatives, raise privacy concerns, and may disproportionately affect short-form writing and multilingual students. Similarly, an increasing number of universities are discouraging or discontinuing the use of AI detection tools due to concerns about their reliability, transparency, and potential impact on students. For example, the University of Waterloo discontinued Turnitin’s AI detection functionality [28], while Vanderbilt University disabled its AI detector [29] and the University of Greenwich chose not to implement it [30].

Instead of focusing on AI detection tools, ideal guidelines should help students and academic staff develop the critical AI literacy needed to use GenAI responsibly and effectively. They should recognize that users need to critically evaluate AI-generated text and editing suggestions rather than accept them without scrutiny. When GenAI outputs are reviewed by users with the necessary awareness and skills, these tools can become powerful aids for refining text and improving efficiency. However, when users lacking such awareness accept outputs uncritically, there is a risk that inaccurate, misleading, or inappropriate content may be incorporated into academic work, potentially misrepresenting the author’s intended message and affecting the author’s academic or professional credibility. The ability to critically evaluate GenAI outputs should therefore be considered an essential lifelong competency that universities actively foster.

Guidelines should also emphasize the limitations of GenAI tools and clarify where they cannot and should not replace human expertise, critical thinking, subject knowledge, and scholarly judgment in teaching, learning, and assessment. AI-generated content may be inaccurate, misleading, fabricated (“hallucinations”), or may include inappropriate use of copyrighted material; therefore, users remain responsible for any content they produce, publish, or share. Universities should support responsible exploration of GenAI by providing students and academic staff with inclusive access to a range of vetted tools in secure, appropriate environments, e.g., at the University of Oxford [31]. Ultimately, GenAI should complement rather than replace the development of evidence-based reasoning, disciplinary knowledge, and critical thinking skills that underpin higher education. Users should be encouraged to verify AI-generated outputs against reliable sources to assess accuracy and identify potential errors.

4.1. Comparison with the UNESCO 2025 Survey

Our findings align with, but also diverge from, the broader global picture painted by UNESCO’s 2025 survey of 400 higher education institutions across 90 countries [32]. While our study found that 90% of the world’s top 200 universities already have publicly available GenAI guidelines, UNESCO reports that only about 19% of institutions globally have formal policies, though two-thirds have or are developing guidance. This discrepancy underscores a sharp equity gap. Furthermore, while 90% of academics in the UNESCO sample use AI tools, over half feel uncertain about how to integrate them into pedagogy, and one in four institutions has already encountered ethical issues, including student overreliance, authorship disputes, and research bias. Table 6 contrasts our elite-university sample with the UNESCO global survey.

UNESCO’s survey identifies two starkly different policy strategies. The first is a regulatory approach, focused on detecting AI use and punishing misuse. The second is an iterative, emergent approach that consults faculty and students, makes AI literacy a mandatory component, and redesigns assessment systems. Our finding that many elite universities, including Indiana University, Vanderbilt University, and the University of Waterloo, are abandoning Turnitin AI detection tools [27,28,29], and our recommended emphasis on AI literacy over surveillance, lean heavily toward this second, emergent approach.

4.2. Study Limitations

The limitations of this study include the exclusive focus on publicly accessible GenAI guidelines from university websites. It is important to note that some universities may have additional guidelines available on their private intranet, which we were unable to access or verify for our survey. Consequently, any potential insights or guidelines present in those documents were not considered in our analysis. Moreover, our study focused on items from a 24-item checklist, thereby excluding any additional items that might have been relevant.

Our paper does not systematically analyze regional differences (e.g., whether European universities emphasize data privacy more than Asian ones). This is a deliberate limitation of our descriptive design: the 24-item checklist was applied uniformly to all 180 guidelines, and the sample sizes within individual regions (other than North America, Europe, and East Asia) are too small for robust statistical cross-tabulation, with Africa and Latin America markedly underrepresented in the QS top 200 and therefore in our sample. Future research should purposively sample universities by UNESCO region or World Bank income category to test whether privacy framing, academic integrity enforcement, or tool licensing vary systematically with geography or national data protection regimes.

Additionally, this study lacked inter-rater reliability checking. Ideally, two researchers should have independently coded all 180 guidelines against the 24-item checklist, or at least a second coder should have coded a subset; however, due to limited resources, only one author (M.U.) performed the coding. Furthermore, our sample is focused on elite, wealthy, research-intensive universities and is not fully representative of the broader global higher education landscape, including developing regions and nascent universities. Nevertheless, we provide below practical guidance on how universities in the Global South (two-thirds or more of the world’s institutions) can benefit from our study.

4.3. Implications for Resource-Constrained Universities

Our study focuses on the world’s top 200 universities/institutions that are, by definition, elite, well-resourced, and research-intensive. A legitimate question, therefore, is how this analysis could be useful for a new or relatively new, non-ranked, and resource-constrained university in a developing country. Here, we outline where our study delivers direct value and where it requires significant adaptation.

4.3.1. What these Universities can Use Directly

First, the 24-item checklist (Appendix A) is the most practical contribution of this paper. Even if a nascent university cannot address all 24 items, the checklist helps administrators answer the question: ‘What are we forgetting?’ They can begin with the five or six items that cost nothing, such as requiring students to name the tool they used, or stating that instructors must approve AI use in assessments. Second, our finding that 0% of top universities ban GenAI outright provides political cover for administrators facing pressure from local stakeholders (parents, ministry officials, or traditional faculty) to prohibit AI use. The global consensus is clear: conditional permission is the only realistic stance. Third, our warning about AI detection tools could save resource-constrained universities both money and reputational harm. We note that many elite universities, including Indiana University, Vanderbilt University, and the University of Waterloo, have abandoned Turnitin’s AI detector and similar tools because they produce false positives and disproportionately flag non-native English speakers [27,28,29]. For a nascent university with limited funds, this is critical: scarce resources should not be wasted on unreliable detection software. Fourth, the ‘gaps’ identified in our study, such as explaining how algorithms work (in order to better understand their limitations), documenting prompts, and reporting mechanisms, provide an essential list of what not to forget, even if some items cannot be implemented immediately.

4.3.2. Where Our Study Needs Heavy Adaptation

The surveyed universities frequently mention ‘university-licensed’ versions of Copilot and Gemini, and have dedicated teaching centers, legal teams, and IT security offices to draft and enforce these policies. A new or resource-constrained university may lack the budget for enterprise AI licenses, the staff time to form a ‘Center for Teaching and Learning,’ robust data-privacy legislation to reference, or reliable campus-wide Internet. Moreover, the top 200 QS universities assume a baseline of digital fluency among students and faculty. A nascent university in a developing country may need to start with far more basic questions, such as ‘Does our faculty know what a large language model is?’ In other words, basic AI literacy levels of staff and students must be assessed before policy implementation.

4.3.3. A Pragmatic Way Forward

If advising a new university in a developing country, we would recommend the following steps. First, adopt the 24-item checklist, but not the guidelines’ verbatim policy language. The 180 guideline URLs in Appendix B provide direct access to real policies from MIT, Oxford, NUS, and others; however, a university should not attempt to build an Oxford-level policy on a shoestring budget. Instead, our study should be used as a map of the terrain, with each institution building the path it can actually afford to maintain. The UNESCO 2025 survey [32] further justifies a lean, literacy-first, iterative policy approach. Second, use the 24 items to run a quick internal audit, marking which are ‘free’ and easy to include (academic integrity rules, citation requirements) versus ‘expensive’ (licensed tools, dedicated GenAI portals, legal compliance reviews). Third, start with three pillars: (a) AI use is permitted only when the instructor approves it, (b) students must declare what tool they used, and (c) the student is responsible for verifying accuracy. These three rules cover approximately 70% of institutional risk with zero technology spend. Fourth, invest in faculty literacy first, not student surveillance. One of our paper’s core arguments is that AI literacy beats AI detection. For a resource-constrained university, training 50 faculty members to understand hallucinations and prompt engineering is cheaper and more effective than purchasing AI detection software licenses.

4.3.4. Other Resources to Consult

In addition to our study and the UNESCO 2025 survey [32], nascent universities should consult UNESCO’s ‘Guidance for Generative AI in Education and Research’ (2023, updated 2026) [33], which is explicitly designed for ministries and institutions in the Global South. National or regional ministries of education in developing countries have increasingly issued AI-in-education frameworks that carry legal weight; this paper does not. Practical, non-paywalled implementation guides are also available from Jisc (UK) and Educause (US). The references within our paper, particularly Crompton et al. (2026) [11], Li et al. (2026) [21], and McDonald et al. (2025) [14]—offer theoretically robust governance frameworks and cross-national comparisons. Finally, the ‘Kelley AI Playbook’ [27] provides a concrete faculty-facing guide from Indiana University that can be adapted to local needs.

4.4. Anticipated Development

The priorities expected for the use of GenAI in higher education will continue to evolve over time. A significant development is the transition from strict, punitive rules towards more flexible guidelines tailored to specific academic disciplines [11,13,22]. Additionally, there will likely be a greater focus on fostering AI literacy and providing instructional support for educators to effectively integrate AI into their teaching practices [10,13,14]. We can also expect ongoing shifts in institutional stances regarding AI detection tools and origin-tracking mechanisms, favoring evidence-based, thoughtful policymaking over automated surveillance [14,15,26]. Furthermore, it will be essential to align institutional policies with external regulations while promoting equity, data privacy, and fair access for all students [18,23,32]. Together, these developments aim to establish a balanced, sustainable approach to the incorporation of AI across higher education.

5. Conclusions

In summary, well-conceived guidelines for GenAI use are crucial for navigating the complexities of higher education as AI technologies become increasingly embedded in teaching, learning, and academic practice. Our research revealed that universities worldwide are actively adopting GenAI guidelines to ensure the ethical and lawful use of GenAI in higher education.

Future guidelines should adopt a more comprehensive and adaptive approach by continually evaluating and reflecting on the use of GenAI as technologies and use cases evolve. Institutions should strengthen their understanding of how AI affects academic integrity standards, recognize that new challenges will continue to emerge, and adopt policies accordingly. They should also address equity considerations related to the adoption and use of GenAI tools, remain open to feedback from their communities, and commit to regular reviews to ensure inclusive access for students and staff. Ongoing training, clear guidance, and institutional support will be essential to encourage the safe, responsible, and equitable use of GenAI across higher education.