Generative Artificial Intelligence in Higher Education: Evidence from an Analysis of Institutional Policies and Guidelines

1. INTRODUCTION

• ChatGPT’s uptake prompted faculty and administrators to respond to student use, particularly for text-based assignments and possible academic misconduct.Institutional reactions ranged from calls to ban ChatGPT to embracing it, while guidance for teachers and students proliferated.Source: §1 · p. 2
• The study examines R1 universities’ GenAI policies, the guidance they give faculty, and themes shared across institutions.The reported findings describe a range of guidance, from prohibition to endorsement, alongside a trend toward embracing GenAI.Source: §1 · p. 2
• Many institutions endorse pedagogical revision without addressing long-term implications, time investment, ownership, or privacy and ethics concerns.The introduction reports that fewer than half discuss privacy implications and ethics of GenAI use.Source: §1 · p. 2

2.1. Resistance to ChatGPT

• Some are staunchly opposed to LLM-based GenAI applications like ChatGPT, while ChatGPT detectors are unreliable, though recent work suggests the detection battle may continue.New detection modes, such as prompt engineering, are emerging.Source: §2.1 · p. 3

2.2 Cautious Embrace of ChatGPT

• Some embrace ChatGPT’s classroom possibilities, including roles as a “personal tutor,” “study buddy,” or “socratic opponent”; good teaching is mentioned, and many questions remain.Mollick and Mollick argue that ChatGPT can improve “transfer”.Source: §2.2 · pp. 3–4
• Even among those embracing GenAI, instructors may struggle to assess its classroom feasibility, and its benefits for teaching workload, long-term engagement, and knowledge building remain uncertain [9, 42].The literature also cautions that GenAI cannot repair structural problems or biases in education.Source: §2.2 · p. 4

2.3 Privacy Concerns

• Privacy concerns include uncertainty about whether entering student work into GenAI violates FERPA or school policies, and risks from sharing personal or sensitive data with third parties.Potential harms include data breaches and targeted advertising.Source: §2.3 · p. 5

2.4 Intellectual Property Concerns

• Intellectual-property concerns arise because GenAI content is trained on other sources, while students’ own essays may also be sent to GenAI when they seek prose improvements.The literature questions whether citing only ChatGPT addresses intellectual property as well as authenticity.Source: §2.4 · p. 5

2.5 Diversity, Equity and Inclusion (DEI)

• GenAI may expand access through writing support for non-native English speakers, but could widen gaps by advantaging students who can afford updated versions and reinforcing existing biases.The literature presents both equity potential and concern about unequal access.Source: §2.5 · p. 5

2.6 Research Considerations Related to ChatGPT

• In research, GenAI is used for brainstorming and writing, and may help non-English-speaking researchers publish, but raises concerns about intellectual integrity, authenticity, and credibility.The extent of its use in academic papers remains unclear, and publication policies vary between bans and transparency requirements.Source: §2.6 · p. 6
• A research consideration is that GenAI can produce entirely fake publications.Source: §2.6 · p. 6

2.7 GenAI and the Future of the Workforce

• GenAI has permanently impacted education and will continue to do so; universities should prepare students for workplaces where it will be used and teach its uses and ethical implications.Source: §2.7 · p. 6

2.8 Research Questions

• The study asks what guidance HEIs provide, whether they encourage or discourage GenAI use and how that appears in guidelines, and what ethical and privacy considerations they address.Source: §2.8 · p. 6

3. RESEARCH STUDY

• The researchers examined publicly available GenAI policies and guidance from R1 institutions, limiting the sample to support greater consistency in interpretation.R1 institutions were selected from the Carnegie list as universities with the highest research activity.Source: §3 · p. 7, §3.1 · p. 7

3.1 Data Collection

• Researchers searched university websites for classroom GenAI policies or mentions, checking Google when site information was absent; 14 institutions had none at search time, and 116 institutions were included.Source: §3.1 · p. 7
• The search ran from October 9 to November 26, 2023, used only publicly available material, and was not fully exhaustive.It could miss policies made public after collection or documents accessible only through institutional resources.Source: §3.1 · p. 7

3.2 Open Coding

• Researchers developed and refined an open codebook, then applied codes to the remaining documents in separate coding rounds.After independent coding of a random subset, two researchers reached near-perfect agreement; formal reliability measures were not used given the qualitative analysis.Source: §3.2 · pp. 7–8
• The initial analysis identified five main areas, followed by validation through subcodes that informed the findings.The subcodes are shown in Table 1 and elaborated in the findings.Source: §3.2 · p. 8

4.1 Overview

• Half of institutions supplied sample GenAI curricula or activities, while 44% discouraged GenAI detection tools.Sample syllabi were supplied by 65 institutions (56%), and 64 (55%) included syllabus statements for different use guidelines.Source: §4.1 · pp. 8–9
• Guidance included both integration and discouragement, with over half offering assignment-design advice to discourage student use and half urging reflection on teaching and evaluation.Some institutions provided both kinds of advice, allowing faculty to choose based on their intended approach.Source: §4.1 · p. 9
• Ethics guidance addressed DEI, privacy, third-party data sharing, and classroom discussion of GenAI ethics.The respective counts were 60 institutions (52%) for DEI, 66 (57%) for privacy, and 61 (53%) for classroom ethics discussions.Source: §4.1 · p. 9
• Activities focused mostly on writing, while code and STEM activities appeared in half the institutions’ guidance and were often vague.The passage reports code and STEM-related activities in 58 institutions (50%).Source: §4.1 · p. 9

Discourage GenAI detection

• Strategies for discouraging detection-tool use included contextualizing assignments, process-based evaluation, visible thinking, and scaffolding work in stages.Other listed approaches include using course materials and core concepts and incorporating personal details.Source: Discourage GenAI detection · p. 10, Assignments to discourage · p. 10

Assignments to discourage

• Suggestions for assignments range from having students critique GenAI output to testing assignments with GenAI and using low-stakes work.Source: Assignments to discourage · p. 10
• Some assignment guidance cautions against sharing sensitive or private information and notes legal implications such as FERPA and HIPAA.Source: Assignments to discourage · p. 10

4.2 Embracing GenAI

• Many institutions provide sample classroom activities and lesson-planning ideas while warning that GenAI detection tools are faulty and unreliable.Source: §4.2 · p. 11

4.2.1 GenAI curriculum and activities

• Activities invite students to refine prompts and critically assess GenAI-generated information, including its accuracy and weaknesses.Some guidance also uses GenAI to prompt brainstorming and deepen group discussion.Source: §4.2.1 · p. 12

4.2.2 Discourage use of GenAI detection tools

• Institutions caution that detection tools are unreliable and advise instructors to be thoughtful about using them in courses.Source: §4.2.2 · pp. 12–13

4.2.3 Lesson Planning

• Lesson-planning suggestions include using GenAI to design lectures, quizzes, projects, exams, and rubrics, with some guidance on tailored learning and feedback.Source: §4.2.3 · p. 13
• Students are encouraged or required to describe their GenAI use and cite it; 44 institutions (38%) provide formal citation guidelines, often based on APA.Source: §4.2.3 · p. 13

4.3 Discourage GenAI

• 31 institutions (27%) provide curriculum intended to discourage GenAI use.Source: §4.3 · p. 13

4.3.1 Assignments to discourage use of GenAI

• Suggested assignments connect work to course material, class discussions, lectures, readings, or students’ lived experience.Source: §4.3.1 · p. 14
• Other strategies include reflection, scaffolding, and questions requiring logical reasoning or abstract thought.Source: §4.3.1 · p. 14
• Some instructors are advised to use GenAI to test and make questions more difficult, while others use process-based evaluations such as oral exams and debates.Source: §4.3.1 · p. 14
• The University of Central Florida describes preventing GenAI use as futile and says it will not pursue conduct cases based only on detector evidence.Its guidance cites rapid technological improvement and unreliable detection methods.Source: §4.3.1 · pp. 14–15

• 69 institutions (60%) caution instructors about privacy concerns with GenAI.Source: §4.4 · p. 15
• Guidance advises instructors to caution students about personal or sensitive data and to review privacy, security, and legal implications before using GenAI.Source: §4.4 · p. 15
• The paper characterizes many privacy admonitions as vague and not student-focused.Source: §4.4 · p. 15
• Some institutions recommend voluntary use or an opt-out alternative, particularly when students may not want to create accounts or share data with AI systems.Source: §4.4 · p. 16
• Only 18% mention legal prohibitions on entering student data, despite guidance noting that submissions could violate FERPA unless the system is approved for such use.Source: §4.4 · p. 16

• The most frequent DEI concern is that biases in human-created training material may be reflected and reinforced in GenAI outputs.Source: §4.5 · pp. 16–17
• Guidance recognizes that GenAI may support accessible or customized learning, but students can face barriers such as cost, screen-reader access, and unequal infrastructure.Source: §4.5 · p. 17
• Some guidance notes a tension: GenAI may help non-native English speakers with language, while those relying on it may be more exposed to plagiarism accusations.Source: §4.5 · p. 17
• The paper also reports concerns that minority students, including nonusers, may face increased plagiarism accusations related to structural bias.Source: §4.5 · p. 17
• Some DEI policies do not make clear whether their discussion concerns GenAI specifically or AI technology more broadly.Source: §4.5 · p. 17

4.6 STEM-related use

• 58 institutions (50%) mentioned GenAI in STEM courses; 56 (48%) mentioned computer science, while engineering appeared at 7 institutions.Math and natural sciences were discussed less often.Source: §4.6 · p. 17
• STEM guidance was often superficial, framing GenAI around coding alongside writing or listing fields without specific STEM-related discussion.Some institutions associated STEM with non-ChatGPT tools, while others described content generation across fields.Source: §4.6 · p. 18
• Problem sets requiring explanations also depend on students generating language to learn, while institutions described GenAI’s learning impact as applying across disciplines.The discussion says coding and related STEM applications are not seen as a major threat or boon.Source: §4.6 · p. 18

4.7 Research

• Fewer than ten institutions mentioned GenAI use in research.Source: §4.7 · p. 18
• Some research guidance warned that institutions lacked clarity about platform privacy implications or contractual compliance, and urged caution about exposing primary research.Universities also cautioned against unintended release of intellectual property, copyrighted materials, or trade secrets.Source: §4.7 · p. 18
• Guidance also told faculty not to enter confidential, protected, or non-public research data into publicly available or vendor-enabled AI tools.Source: §4.7 · p. 18

4.8 Where do institutions fall along the spectrum of embracing (or not) GenAI?

• 31 universities (27%) were categorized as discouraging GenAI use, including 9 (7%) that provided little guidance.The categorization included policies emphasizing legal and intellectual dangers.Source: §4.8 · p. 18
• 73 universities (63%) encouraged GenAI use, and 48 (41%) supplied ample classroom guidance.Figure 2 arranges policies by encouragement or discouragement and by extensive or minimal guidance, using bubbles and Ns.Source: §4.8 · pp. 18–19, Figure 2 · p. 19
• Another 25 universities (22%) encouraged use through extensive guidance, such as syllabus statements, without necessarily providing sample curriculum.The authors defined encouragement to include ample classroom guidance or other extensive guidance.Source: §4.8 · p. 18

5. DISCUSSION

• Many institutions embrace GenAI while also offering strategies to limit its use, and some describe that embrace as pragmatic rather than pedagogically driven.Examples of pragmatic rationales include treating GenAI as a fact of modern life or questioning whether detection tools can be satisfactory.Source: §5 · p. 19
• Policies often make GenAI a central classroom presence, recommending it for group work, engagement, discussion, tutoring, feedback, lesson planning, grading, email, and individualized resources.The authors describe policies that model a classroom in which GenAI is the protagonist.Source: §5 · p. 19
• GenAI-related redesign does not necessarily originate in-class evaluations, reading, reflection, or oral exams; making AI central can instead foreground prompting and critique.The authors note that outputs are frequently not students’ own work, and that scholarly output may include commentary on how GenAI was used.Source: §5 · p. 20
• Some pro-GenAI guidance gives little attention to ethics and privacy and can contradict itself by recommending AI grading while warning against entering sensitive data.The authors also say that FERPA, intellectual property, and student privacy concerns appear to be an afterthought in some guidance.Source: §5 · p. 20
• The authors distinguish critically examining GenAI outputs from making GenAI a central classroom actor, warning that acknowledging it can be mistaken for experimenting with it.Suggested activities include comparing and fact-checking GenAI, while the authors question its positioning as a classroom “buddy” or “tutor.”Source: §5 · p. 20
• The authors find no reference to research-backed evidence for guidance, while normalization may impose time, ethical, and cognitive overhead around prompts and process data.They also raise concern about possible intellectual discouragement when students complete work and then ask GenAI to replicate it.Source: §5 · pp. 20–21
• Long-term effects on intellectual growth and pedagogy remain unclear, and the authors worry that dependence on GenAI for analysis and idea generation may weaken retrieval.Some guidance recommends retrieval-reinforcing activities, which the authors connect to a concern that retrieval skills may atrophy.Source: §5 · p. 21

6. CONCLUSIONS, LIMITATIONS, AND FUTURE WORK

• The analysis characterized institutions by whether they encouraged or discouraged GenAI and by how much guidance they offered.The identified themes included communication and curriculum guidance, lesson planning, privacy, and ethics.Source: §6 · p. 21
• Many institutions fervently embrace GenAI in ways that may burden faculty and students, with little regard for ethical concerns.The authors caution that encouragement may not produce the desired learning outcomes.Source: §6 · p. 21
• GenAI can be viewed as a catalyst for assessment and evaluation practices that are more ecologically valid and grounded in fairness, justice, and ethics.The authors say classroom use may require more thoughtful consideration of GenAI’s role.Source: §6 · p. 21
• The study captures one moment in time and could not include all guidance instructors encounter through their institutions, colleagues, and networks.Future work will explore higher-education contexts outside the US.Source: §6 · p. 21