Communicating Clear Guidance: Advice for Generative AI Policy Development in Higher Education

Technological Change in Higher Education

GenAI’s reach has expanded and become available in commonly used products, applications, and browser extensions. Microsoft Copilot incorporates GenAI into Word, PowerPoint, Teams, and Excel. Google offers Duet AI, which is available in Google Workspace. Duet AI includes GenAI for writing, and Google pitches the product as “AI as a real-time collaborator” (Pappu, 2023). In short, students will have GenAI built into traditional tools for writing.

Further, when considering the reach of GenAI, new incoming students will encounter GenAI as a normal tool they may have used during their primary school years. Psychologically, for new university students, as far as they are concerned and insofar as they will have experienced it, the college experience will have “always had” an AI component. This technology transition mimics the technological disruptions seen with digitized card catalogs, access to search engines for research, or the advent of slides or email as teaching and learning tools. Considering the advent of email, current students could not conceive of a college environment where professors required only in-person office hours and did not have an email. Still, some current faculty had email-free experiences as college students or junior faculty.

Policies have been used to manage technological change in the classroom and provide a point of consideration when developing policies and guidance about GenAI. For example, some faculty used syllabus policies that banned online encyclopedias in the 1990s and early 2000s. Faculty expressed concerns about academic integrity given the new technologies. Some banned Wikipedia and included that ban in their syllabi. At a conference in 2006, the founder of Wikipedia said he had “no sympathy” for students who got into trouble for using the online user-written encyclopedia (Young, 2006). In 2007, Middlebury College made national news by banning Wikipedia in the history department, citing Wikipedia’s inaccuracies despite other scholarly findings that Wikipedia was more accurate than traditional encyclopedias (Jaschik, 2007). Meanwhile, other faculty pointed out the importance of focusing on content to promote student success during the trend to ban Wikipedia (Maehre, 2009). The teacher-scholars who emphasized content focused on postgraduation outcomes rather than classroom-specific artifacts appropriate for assessment. Maehre (2009) argued:

If we consider students to be learners rather than performers, and if we value the learning of thinking skills rather than the carrying out of orders, we won’t be afraid to allow students to explore and to not always get the “right” answer. If we won’t do this, I’m afraid we’re not helping students develop into the increasingly-interactive world they’ll enter after graduation. They will embark on professional careers not just as readers or finders of information, but as collaborators in information creation. (p. 236)

Since that article’s publication, the conversation about Wikipedia has changed (Konieczny, 2016), but the discourse about Wikipedia and learning echoes the current language about GenAI as a teammate or collaborator. For most faculty teaching now, banning Wikipedia would be unlikely, and banning search engines is implausible. Some have compared banning calculators or Wikipedia to banning ChatGPT (Kelly, 2023). In an editorial, a communication researcher who studied Wikipedia calls ChatGPT this generation’s Wikipedia (Geiger, 2023). Similarly, ChatGPT may be this decade’s Wikipedia for faculty.

The anxiety and frustration about students’ Wikipedia use mirrors ongoing sentiment about students’ use of GenAI. In 2023, a majority of surveyed college officials “agreed” or “strongly agreed” that GenAI posed “a threat to higher education” (Chronicle of Higher Education, 2023, p. 4). However, of those respondents, only 32% had created a policy, and about half had no conversations or meetings with faculty or students about GenAI (Chronicle of Higher Education, 2023, p. 19). The negative sentiment and lack of direction have led to reports of confusion and stress. Some university’s restrictive policies reportedly “panicked” students, who wanted more information on using GenAI tools appropriately (Williams, 2024). Lacking a policy or guidelines can create greater confusion.

Responses to AI and Policy Development

Higher education reactions have ranged from not having a written policy about GenAI (Zimmerman, 2023) to allowing students to help write the policy (Brown, 2023). Many universities have posted policies online. In Grammarly’s Fall 2023 “Back to School” webinar on AI policies, attendees marked this shift of GenAI’s presence in academia with a poll about what attendees’ schools were doing in response to the growing popularity of GenAI tools. Of the 313 attendees, 67% (209) reported that in the 2022-2023 school year, their school did not have a formal GenAI policy guiding faculty and students on how to use it in academic settings. Conversely, 10% (32) responded their institution had such a policy, while 23% (72) were unsure of their school’s policy. However, when asked about the 2023-2024 school year, of the 318 attendees polled, 45% (142) reported their school did not have a formal GenAI policy, 29% (94) said their school did, and 26% (82) did not know (Moore & Maxwell, 2023). As of 2023, most universities did not have policies (Xiao et al., 2023). Universities with policies follow an interesting pattern in policy adoption. Xiao et al. (2023) found that “of the universities with ChatGPT policies, approximately 67% embraced ChatGPT in teaching and learning, more than twice the number of universities that banned it” (p. 2).

Specific to the field of business communication, faculty views on policies varied. In a 2023 survey, most business communication faculty preferred that policy be developed “primarily” at the school, university, or department level; less than 20% preferred that policies be developed by instructors (Cardon et al., 2023). In that survey, about 12% of faculty wished for a ban; the researchers found those faculty had “strong feelings” and saw “AI-assisted writing as incompatible with the development of writing skills and critical thinking” (Cardon et al., 2023, p. 273).

The policies identify how administrators and faculty perceive and frame GenAI. Luo (2024) argues that policies about GenAI “shape how problems are understood by prioritising certain representations over others” (p. 2). In an analysis of 20 university policies, the problems identified included academic misconduct, a lack of policy, weak assessment design, training, equity, and issues with technology such as hallucinations (Luo, 2024).

Privacy and access are also factors in policy development. Clear policies are needed for student use and data protection (Dwivedi et al., 2023). For example, our university has taken the position that given FERPA, student work cannot be put into an external GenAI tool including ChatGPT and external plagiarism detectors. Further, the policies of the tools may limit who can access and use the tool. Age limitations are present. ChatGPT requires users to have parent permission if under the age of 18 years (OpenAI, n.d.). Microsoft Copilot for higher education is limited to student users of age 18 years or older (Welcome to Copilot in Windows, n.d.). About 2.1% of students in postsecondary institutions are under the age of 18 years (Hanson, 2024), but that number is about 20% at a 2-year college (The Hamilton Project, 2023). Given the age restrictions, faculty must consider who will be unable to use GenAI, particularly at the 2-year college level. Further, institutions must act to protect minors and promote equitable use, particularly if students can use GenAI but those under 18 years technically cannot. Policies must consider the changing legal and technological landscape to ensure all students have an equitable academic experience.

Policy language can influence equity and comprehension. Language use influencing compliance includes the use of the negative. Researchers studying codes and language that prohibited behavior identified issues with how respondents read the codes. George et al. (2014), for example, note, “This result is surprising as it indicates that respondents do not differentiate between ‘prohibition’, ‘should not’ and ‘will not’!” (p. 11). In short, George et al. found that saying “shall” rather than “should” did not make a significant difference. Further, they found that factors like age and race influence perceptions.

Finally, effective policy should operate appropriately for different dimensions: pedagogical, operational, and governance (Chan, 2023). In Chan’s (2023) AI policy framework, the following considerations are proposed for pedagogical policy: “Rethinking assessments and examinations; Developing student holistic competencies/generic skills; Preparing students for the AI-driven workplace; [and] Encouraging a balanced approach to AI adoption” (p. 20). Complicating the considerations, some universities have adopted GenAI tools for classroom use or teaching while adopting policies that ban GenAI to promote academic integrity (Xiao et al., 2023). This article focuses on the pedagogical dimension of policy.

Sentiment and Emotion Analysis

We conducted a sentiment analysis and emotion analysis of the policies to gain an overview of the sentiments and emotions embedded in them. The analyses were performed through Qualtrics Text IQ AI tool. These methods have been successfully used by researchers to examine qualitative data (e.g., Durand et al., 2023; Tiruye et al., 2023; Winkelmann & Games, 2021). Text IQ uses textual analysis to determine a sentiment score for each policy that ranges from −2 (very negative) to 2 (very positive) (Qualtrics, n.d.). Additionally, Text IQ noted if policies contained mixed sentiments (i.e., including both positive and negative sentiments). The researchers also reviewed the sentiments associated with the policies to verify agreement.

Overwhelmingly, over 60% of the policies presented were either negative (30; 29.41%) or very negative (33; 32.35%) regarding AI. Less than 5% of policies had positive (3; 2.94%) or very positive (2; 1.96%) sentiments. Fourteen (13.73%) policies were neutral, while 20 (19.61%) were mixed in sentiment. The detected emotions (102) reflected the negative sentiment. The predominant emotion detected in the policies was anger (28; 27.45%), followed by frustration (15; 14.71%) and mixed emotions (15; 14.71%). Of the policies, 35 (34.31%) were detected as having no emotion. See Figure 1 for details.

OPEN IN VIEWER

Figure 1.

Detected emotions in policies about generative AI (N = 102).

Negative and strongly negative sentiments did not necessarily align with negative views of GenAI use. Intent and word choice sometimes revealed complicated positions. Some policies stressed the limitations of GenAI while permitting its use. This sometimes resulted in a very negative sentiment for a policy in which the faculty member may have felt positively about GenAI. For example, in a policy that allowed using GenAI for brainstorming and outlining, the policy stated, “the material generated by these programs may be inaccurate, incomplete, biased or otherwise problematic. Beware that use may also stifle your own independent thinking and creativity.” That policy was coded as very negative. Similarly, another policy that encouraged cited use explained that generated content could be “biased, offensive” and “unethical.” While the policy had positive intent, the sentiment was scored as very negative because of the negative words incorporated.

Similarly, a positive sentiment policy might limit or ban GenAI use. For example, one of the two policies coded as “very positive” banned using GenAI. The policy employed positive language to explain that “AI tools” could be used for spelling and grammar checks, and only for those applications. The policy avoided most negative words and sentence constructions and did not include discussions of violations and punishments. Instead, the policy encouraged students to ask the faculty member questions, with the following explanation: “Scholarly citing is not particularly intuitive, and part of my role is to help you learn the rules for intellectual attribution.” The policy promoted conversations and comprehension without using fear. Similarly, the other policy coded as very positive included limitations on GenAI use. The policy explained using bulleted items the types of assignments and use that were not allowed. However, it also explained ethical and appropriate use using positive language and clear explanations free of jargon. For example, the same policy stated, “When you use AI for an assignment, you should screenshot any queries and output from AI and submit it with your work. . . . If you are open with me about your uses of AI, I can help you make the most of it without penalizing you.” The policy did not allow all uses but promoted positive conversations.

Common Policy Themes

While the policies share many common phrases and word choices, the policies differ significantly. Some are around 100 words or less, while others gave more than 500 words of guidance, including examples. Some appear to repeat university policy or guidance; some authors note they differ from official guidance and other professors’ policies.

To take a deeper look at the policies themselves, the authors conducted a thematic analysis of the policies. The themes provided insight into how universities and professors defined uses and explained AI.

Write the Policy for Your Audience (Your Students)

First, faculty and institutions must have clear policies created with the audience in mind—that students can understand. Appendix A provides examples of language that might be used. Syllabi language can improve feelings of belonging and inclusion (Sunds et al., 2023). First impressions and language matter.

Set the Tone and Focus for the Course

Second, faculty and institutions must set an appropriate tone and use policies not merely as a tool for legal compliance but as an opportunity to promote appropriate and effective communication. Beyond writing an AI use policy, instructors should:

Finally, beyond the findings in this article, we recognize the potential for continued rapid technological changes and changing institutional policies. We may need to adjust our policies to consider Artificial General Intelligence (AGI) in a few years. Faculty should consider using a broad syllabus policy that narrow assignment instructions can support. Using assignments to clarify AI tool use rather than a broad policy can support teaching different applications and contexts for AI tool use. For example, faculty might want students not to use GenAI on one assignment but allow GenAI on another assignment. This matches corporate practices, which deserve future research for their sentiment and themes.