Category: Teaching & Learning Resources

UW-Green Bay’s libraries have an excellent student-facing webpage on how to acknowledge or formally cite the use of GAI. This blog is intended to supplement that resource with information more specific to instructors. Professors will be vital in helping students understand both the ethics and practicalities of transparency when employing GAI tools in our work. Please keep the following caveats in mind as you explore this resource.

• As with all things GAI, new developments are rapid and commonplace, which means everyone needs to be on the alert for changes.
• Instructors are the ones who decide their specific course policies on disclosing or citing GAI. The information below provides some options for formatting acknowledgments, but they are not exhaustive.
• Providing acknowledgment for the use of GAI may seem straightforward, but it is actually a very nuanced topic. Questions about copyright implications, whether AI can be considered an “author,” and the ethics of relationships between large AI entities and publishing houses are beyond the scope of this blog. Know, though, that such issues are being discussed.
• Please remember that it is not only important for students to acknowledge or cite the use of GAI. Instructors need to do so with their use of it, as well.

Acknowledgment or Citation of GAI

There is a difference between acknowledging the use of GAI with a simple statement at the end of a paper, requiring students to submit a full transcript of their GAI chat in an appendix, and providing a formal citation in APA, MLA, or Chicago styles.

• UWGB Libraries have some excellent acknowledgment examples on their page.
• UWM’s library page provides basic templates for citations intended to be consistent with APA, MLA, and Chicago styles.
• There are also lengthy blog explanations and detailed citation examples available directly from APA, MLA, and the Chicago Manual of Style.

Regardless of the specific format being used, the information likely to be required to acknowledge or cite GAI includes:

1. The name of the GAI tool (e.g., Copilot, ChatGPT)
   Microsoft Copilot, OpenAI’s ChatGPT 4.o (May 23, 2024 version), etc.
2. The specific use of the GAI tool
   “to correct grammar and reduce the length in one paragraph of a 15-page paper”
3. The precise prompts entered (initial and follow-up)
   “please reduce this paragraph by 50 words and correct grammatical errors”; follow-up prompt: “now cut 50 words from this revised version”
4. The specific output and how it was used (perhaps even a full transcript)
   “specific suggestions, some of which were followed, of words to cut and run-on sentences to revise”
5. The date the content was created
   August 13, 2024

Ultimately, instructors decide what format is best for their course based on their field of study, the nature and extent of GAI use permitted, and the purpose of the assignment. It is important to proactively provide specific information to students about assignments. Professors who are particularly interested in whether students are using GAI effectively may focus on the prompts used or even ask for the full transcript of a session. If, in a specific assignment, the instructor is more interested in students learning their discipline’s citation style, then they might ask for a formal citation using APA format. Although the decision is up to the professor, they should tell students in advance and strongly encourage them to have separate Word documents for each of their classes in which they save any GAI chats (including prompts and output) and their date. That way they have records to go back to; If they use Copilot with data protection, it does not save the content of sessions.

What Messages Might I Give to Students about Using, Disclosing, or Citing GAI?

Instructors should consider how they will apply this information about acknowledgments and citations in their own classes. CATL encourages you to do the following in your work with students.

1. Decide on a policy for acknowledging/citing GAI use for each course assignment and communicate it in your syllabus and any applicable handouts, Canvas pages, etc.
2. Reinforce for students that GAI makes mistakes. Students are ultimately responsible for the accuracy of the work they submit and for not using others’ intellectual property without proper acknowledgment. They should be encouraged to check on the actual existence of any sources cited by a GAI tool because they are sometimes “hallucinated,” not genuine.
3. Talk to students about the peer review and publication processes and what those mean for source credibility compared to the “scraping” process used to train GAI models.
4. Explain that GAI is not objective. It can contain bias. It has been created by humans and trained on data primarily produced by humans, which means it can reflect their very real biases.
5. Communicate that transparency in GAI use is critical. Instructors should be clear with their students about when and how they may use GAI to complete specific assignments. At the same time, one of the best ways instructors can share the importance of transparency and attribution is through modeling it themselves (e.g., an instructor disclosing that they used Copilot to create a case study for their course and modeling how to format the disclosure).
6. Remind students that even if the specific format varies, the information they are most likely to have to produce for a disclosure/acknowledgment or citation is: a) the name of the tool, b) the specific use of the tool, c) the prompts used, d) the output produced, and e) the date of use.
7. Finally, encourage students to copy and paste all GAI interaction information, including an entire chat history, into a Word document for your course and to save it for future reference. One advantage of Microsoft Copilot with data protections is that it does not retain chat histories. That’s wonderful from a security perspective, but it makes it impossible to re-create that information once a session has ended. They should also know that even GAI tools that save interactions and use them to train their model are unlikely to re-produce a session even if the same prompt is entered.