Exploring the Utility of AI-Assisted Session Notes

Ava Cloghessy, College of the Holy Cross

Introduction

Throughout the Spring 2024 semester, I experimented with ChatGPT to produce AI-assisted session notes for the Writer’s Workshop at the College of the Holy Cross. While session notes follow a standard format, the process of writing them is tedious and can be draining, especially at the end of a three to four hour shift. Given that session notes are an important resource for students, I partnered with my Writing Center Associate Director to explore session notes as a use case for generative AI. Although many may be uncomfortable with integrating AI, I found it enabled me to approach session notes through a recursive lens instead of a static one. Through this experience, I developed a process that not only can help consultants write more detailed notes in a shorter time frame, but also reflect on their practice.

In what follows, I provide an overview of how session notes function as a resource for writers and as a professional development tool. Next, I outline the experiments I conducted and the steps I take to generate an AI-assisted session note. By documenting my process, I aim to demonstrate that while it would fundamentally change how session notes are evaluated, it also is a valuable training opportunity for tutors to become more rhetorically adept users of AI.

The Function of Session Notes

As researchers like Bugdal, Reardon, and Deans (2016) describe, session notes are an essential record of consultations for writing centers that use them. While writing centers utilize different formats for session notes, they frequently represent the work that happens in tutorials to both staff and the larger campus community. At my writing center, which is located at a small Jesuit liberal arts college, undergraduate writing consultants meet with writers for 45 minutes and write a session note afterward that provides a brief summary and revision plan for the client. Session notes may also include tailored resources for students. They are sent to writers using WCOnline. Writers can choose to have their note sent to a professor. At my center, session notes are mandatory to write and consultants receive training on how to write them. Since the notes at my center are addressed to the student and are not solely for internal record keeping or reflection purposes, tutors are trained to write them in the second person. 

As a genre, session notes have largely been studied through rhetorical and audience-oriented perspectives. However, recent quantitative research within the past few years reveals their importance as a tool for tutor professional development. In this literature, session notes function as a way for writing center staff to understand how consultants incorporate the language and practices learned from their training by examining most used words and the context surrounding them (Giaimo et al., 2018; Giaimo & Turner, 2019). By analyzing session note data, Writing Centers can identify trends in tutor-writer communication such as conflict management and collaboration strategies, which can inform future professional development conversations (Dadugblor, 2021; Driscoll & Yim, 2024). Through corpus analysis across writing center institutions, Modey et al. (2021) identified areas for new research and tutor training opportunities to make their writing centers more responsive to writers’ needs. Collectively, these findings are relevant to my research as AI-assisted session notes rely on corpuses, data-driven analysis, and provide opportunities for tutors to critically reflect on their tutoring praxis.  

Session notes are an appropriate use case for AI assistance because the genre follows a predictable structure. While consultants develop their own session note writing style, they tend to use formulaic phrases and make similar rhetorical moves to streamline their production at scale. Since large language models like ChatGPT are probabilistic (predicting words through algorithms without understanding their meaning) (Bender et al., 2021), they perform best with short, routine tasks such as session notes. 

I believe AI will benefit tutor practice by increasing tutor productivity and enabling tutors to reflect on the quality of their session notes. Since ChatGPT can provide an accurate session summary and rough draft of the session note in a matter of seconds, tutors can focus their time on expanding on the feedback they provided during the session. Automating part of the process allows tutors to quickly produce session summaries, which can allow tutors to focus more time with writers in tutorials or have a much-needed break to reflect on their practice in between sessions. 

While session notes can be automated, the tutors that will see the greatest benefits are those that have already been trained on the rhetorical purposes of session notes (Bermingham, 2023a; Bermingham, 2023b). Experienced consultants are more familiar with the ideal finished product and therefore can effectively evaluate ChatGPT’s outputs and avoid over reliance on AI. Some studies have even found AI to have a positive impact on writing self-efficacy, which suggests that integrating AI in a humanistic way could increase the levels of engagement in session note writing. (Washington, 2023; Ranade & Eyman, 2024). Given that writing centers use session notes for different rhetorical situations, it is important to emphasize that integrating AI might do more harm than good. For instance, if the purpose of a writing center’s session note is purely intended for tutor self-reflection, then AI-assistance would be far less beneficial as AI is incapable of doing this kind of analysis and critical thinking.

How to Generate Session Notes with AI Accurately and Efficiently

My process for developing an AI-assisted session note consists of four steps:

• • Recording the session to produce an auto-generated transcript
  • Pasting a prompt containing a specific structure to follow into ChatGPT
  • Attaching the transcript, a corpus of my session note, and a session note how-to guide 
  • Re-prompting ChatGPT if necessary and revising

It is important to note that I did not develop this procedure alone. During the semester, I was the only consultant experimenting with AI assisted session notes, but I did talk to other consultants about the strategies they have developed to write session notes. The overwhelming majority mentioned that they use a consistent format that reads as a friendly email in the second person. When I approached consultants about their thoughts in integrating ChatGPT into their process, some expressed hesitancy as they felt they could write the notes on their own without issue. However, others were curious to see what the technology’s limitations and uses are. 

Throughout the process, I gathered staff input by presenting them with two notes from the same session. One of the notes was AI-assisted while the other was not (see Appendices A and B). Notably, when I asked which note they preferred and why, there were varied responses and it was difficult for people to identify the AI-assisted note. While these mixed responses could be due to the fact that consultants and full-time Writing Center staff members have different levels of familiarity with session notes, it also suggests that there is more openness to AI-assisted notes than previously anticipated. Further formal testing is necessary to yield a more decisive conclusion. The following sections describe some of the conceptual challenges I faced while creating this process.

Capturing the Session 

The first hurdle I overcame in designing my process was figuring out how to efficiently transmit enough details from the session for ChatGPT to generate an accurate summary. Typically, I write a bulleted list during sessions to remember the conversation. While I considered manually entering my notes into ChatGPT, this seemed time-consuming. Considering that generative AI is exceptional at summarizing large datasets within a matter of seconds (Chubb et al., 2022), I asked writers if I could record their sessions to generate automatic transcripts. I used Otter.AI as it allowed me to record using my phone and generate a transcript in real-time, but using transcripts from Zoom is another option.

By including the transcripts, ChatGPT incorporated specific keywords and phrases from the conversation and included information I would have otherwise forgotten to mention. For instance, in one session, a writer used the phrase “content analysis,” which I was unfamiliar with and would not have remembered to include in the session note. The note AI generated from this session; however, included reference to this method: “We explored the complexities of your methodological approach, especially focusing on your survey design and content analysis strategy” (see Appendix B). 

Takeaways: Requiring tutors to record each session to produce a transcript allows ChatGPT to produce a more accurate summary, but it is an intrusive step. With tools like Zoom offering transcription during meetings, it may be easier to integrate AI during virtual sessions with writers. It is ultimately still up to the tutor to ensure that the words ChatGPT pulls from the transcript reflect the overall themes from the session and are not just minor details.

Constructing and Revising the Prompt 

Although AI is well-equipped to generate short, formulaic genres, like session notes, consultants still need to consider prompt engineering strategies to receive a viable output. Without sufficient guidelines, ChatGPT is more likely to hallucinate and add irrelevant or inappropriate details. I based my prompt engineering strategy on advice offered by Deans et al. (2023). In developing my prompts for ChatGPT, I gave the model the identity of a writing consultant, outlined the structure a note should follow, and included phrases it can use as signposting language. 

In my initial experiments, I produced more detailed outputs when my prompts were structured and included models of example notes to follow. These prompts were short and defined the session’s length, a word count for the note, and the audience. They also included guidance that the revision plan for the student should be bulleted (I typically write mine in a bulleted format, and knew ChatGPT could replicate this). 

While the outcome was better than I anticipated, the note was still not ready to be sent out. ChatGPT’s note struggled to balance the different audiences of the student and the professor. The syntax was overly formal and incorporated lofty praise, which consultants are instructed to avoid. To correct this, I explicitly added a line prompting ChatGPT to “avoid evaluative language” and use a “friendly but professional tone.” 

I refined my process through several attempts of prompting and re-prompting. To track my results, I created a prompt archive that other consultants can view and add iterations of successful prompts. One of the additions I made to my prompts is limiting the revision plan to the three most relevant bullet points with a sentence of explanation for each one. The prompt I use most often is reproduced in Appendix A. 

Takeaways: Creating a prompt that addresses the rhetorical situation of a writing center’s session note requires trial and error; it is a recursive process. Despite the time and effort required to experiment with different prompts, it resulted in productive conversations with other consultants about what the ‘model session note’ should look like. Asking ChatGPT to experiment with different structures and audiences is a useful exercise for tutors to reflect on the quality of their own session notes, even if they do not adopt ChatGPT into their process.

Aligning and Customizing the AI Tool 

Although I was producing session notes that accurately documented my sessions, there were still some flaws: the notes conformed with my center’s guidelines but did so inconsistently and they still didn’t ‘sound like’ they were written by me. To train ChatGPT on the genre of the session note, our writing center’s director suggested providing the session note guidelines document for consultants to the AI via OpenAI’s custom GPTs. This feature allows users to complete specialized requests by inputting data and establishing parameters, thus creating a separate knowledge base for the model. At the time of writing, Custom GPTs are only accessible through the premium version of Chat GPT 4.0.  

To address the voice concern, I compiled all of my archived session notes from WCOnline into one document. For my next AI-assisted note, I attached the session transcript and my personal archive of notes. I also gave explicit instructions to emulate my style and tone. These changes brought significant improvement as ChatGPT began borrowing phrases like “going forward,” and “best of luck” that I typically incorporate. I then added these phrases to my prompt to ensure consistency and provide signposts for the model to follow.  

Takeaways: While some may feel uncomfortable training generative AI to ‘sound’ more like themselves, it is a necessary step in order to produce notes that are consistent in tone. ChatGPT performs best when it can draw from a large dataset and clear parameters. Without taking this step, each note ChatGPT generates will vary greatly in tone and the tutor will have to spend more time prompting and editing the first output. Although training ChatGPT to sound like the tutor’s personal writing style does result in upfront costs in time, it eventually results in time savings as the Custom GPT becomes more familiar with the parameters and session note dataset. 

Consultant Revisions 

All of the AI-assisted session notes I sent to students and professors were edited by me in some way. I sometimes prompt AI to make general revisions to structure and tone. However, for sentence-level concerns, I make my own changes. 

While inputting session transcripts led to accurate session notes, ChatGPT generally cannot distinguish the student’s major concerns nor what the consultant believes is most relevant. As a result, AI may generate a lengthy summary or include information that might be redundant. In these instances, the consultant can easily cut out these sections. 

Another aspect of session notes that consultants cannot rely on AI for is including resources and strategies. Although including links to sources or activities is not mandatory at our center, doing so makes the session note a more effective resource for writers. Since the process of creating AI-assisted notes takes less time, it allows the consultant to include resources they otherwise would not.

Takeaways: AI cannot not write entirely for the consultant. It’s helpful to think of it as a collaborative feedback loop between consultant and AI. The AI-generated session note acts as a “first draft” that the writer then uses as a guide and makes revisions to make it more personalized. The portion of the session note that requires the most human intervention is the feedback portion, as ChatGPT is not capable of generating original and specific feedback that is accurate.

More Than Just a Time-Saving Hack

I initially began this project expecting to help my writing center use generative AI to produce higher quality notes in a shorter time frame. ChatGPT helped me accomplish this; however, I found several other unexpected benefits. Most significantly, this process allowed my center to reimagine what session notes are for. Sharing my experience with staff and consultants sparked an open dialogue about how session notes are not merely static records, but dynamic resources. 

As anticipated, I wrote session notes in a shorter time frame with AI assistance though the training of the AI and the editing of notes added up-front time to the writing process. As a result, this process can allow consultants to spend more time with writers during the session. It often takes me fifteen minutes to write a session note. I find that newer consultants have difficulty finishing them in that timeframe. While developing a process for AI-assisted session notes was a time investment at first, I can now generate a usable note with ChatGPT in five to ten minutes. It may seem like a small amount of time-savings, but this allows consultants to spend longer with clients in sessions without running behind schedule. The ability to spend more time with clients does not necessarily mean more labor for tutors. Rather, it allows tutors to have more breathing room during sessions to experiment with different strategies and allow writers more time to develop ideas independently during sessions. At a larger scale, this time savings matters. My center typically posts 120 hours per week, and saving ten minutes each session amounts to 20 fewer hours spent on session notes each week. 

In addition to efficiency, AI-assisted session notes are more detailed than the ones I wrote alone. My AI-assisted notes averaged 212 words, while my own notes averaged 165 words. Although length does not always correlate with better quality as ChatGPT has a tendency to be wordy, it does produce descriptive summaries that consultants often are unable to provide. Attaching transcripts of sessions into ChatGPT was especially helpful in creating a detailed summary while maintaining accuracy. Summarization is one of the most time-consuming aspects of session note-writing.  By allowing ChatGPT to generate an initial draft note using the transcript, consultants can focus more attention on the revision plan and include resources that may be useful to the writer. As a result, AI serves the purpose of offloading the summary aspects of the note, so the consultant can spend more time focusing on providing feedback and resources. In my experience, I find I have more time to give detailed advice to writers that I may not have been able to mention during the session. While the summary is not the most useful part of the note for writers, it allows writing center administrators to get a more detailed understanding of what tutors are discussing with writers during sessions. 

Furthermore, AI-assisted session notes positively impact consultant development by increasing their AI literacy. Before undertaking this project, I was a consultant with little experience using AI tools. I was previously hesitant to use ChatGPT because I lacked solid understanding of how to use it ethically. Through trial and error and learning about prompt engineering, I have a stronger grasp of what AI can offer consultants and the writers they work with. Ensuring that consultants develop AI literacy skills will better prepare them for working with writers already using it in their writing process as they can model best practices.

While my experience using ChatGPT in my session note process is positive, expanding this practice raises some ethical concerns. First, ChatGPT’s tendency to flatten and standardize language is a threat to linguistic justice. However, coding custom GPTs and revising initial outputs, helped mitigate the risk of losing consultants’ personal voice. Nevertheless, the only way to truly preserve personal voice is for tutors to make their own edits to make the feedback more genuine and opinionated. Assessing AI-assisted session notes will also be challenging as it is impossible to determine which words were put there by ChatGPT and which were added by the tutor. As a result, it would be more difficult to measure tutor professional development and draw meaning from trends in AI-assisted session notes. Maintaining privacy is another concern that is difficult to guarantee. OpenAI collects data that users input into ChatGPT to improve its models. Thus, consultants must be cautious of the kinds of information they provide ChatGPT and receive a writer’s informed consent before attaching conversations to prompts. In addition to costs for premium subscriptions, there are concerns about AI’s impact on the climate. A single generative AI input takes about four to five times more energy than the average search engine request (Keller et al., 2024). AI is here to stay, but writing centers will have to decide whether it is worth these costs.

When comparing the AI-assisted session note and the non-AI assisted session note, a few key rhetorical differences emerge. At first glance, it may be difficult to identify which note is which; however, the non-AI assisted session note uses a more complex syntax and has greater sentence variety. The AI-assisted note has a more descriptive and encouraging introduction, but the feedback section is vague. The note written completely by me begins by explaining the writer’s assignment and the class she is in, whereas the AI-assisted session note focused more on what was actually discussed in the session. Despite my instructions to give feedback using a bulleted list, the section’s tone is more formal and wordy in the AI-assisted note. For instance, the phrases “effectively capture the relationship” and “ensuring a diverse representation that aligns with your research objectives,” are verbose and unnecessary for the writer. Ultimately, it is up to the tutor to customize this section and delete pieces of feedback that are irrelevant and add in more tailored advice.

Before fully implementing AI-assisted session notes into a writing center’s praxis, conversations with key stakeholders are necessary to ensure trust. While my writing center has had thoughtful conversations about what makes a session note an effective resource and how AI can support this goal, this practice cannot expand to other consultants without gaining student and faculty support. A question that still remains to be answered is how students and faculty would feel about the Writing Center’s integrity if they read a session note with a disclaimer that ChatGPT was used in the writing process. It is best practice to be transparent; however, how writers and professors will perceive consultants who use AI remains unknown. Consultants will need to ensure that using AI does not make offering feedback any less intentional (Giordano et al., 2024). Building trust will require collaboration between writing centers, faculty, and students. 

My process is by no means the only way to implement AI-assisted session notes. The method described in this article can be altered to meet the varying needs of other writing centers. I   also have explored using other AI tools like Claude to create session notes that do not have the option of programming a chatbot and received usable results. ChatGPT can be prompted to generate session note templates that consultants can adjust to meet their internal purposes or clients’ needs. Even if this process is not implemented as part of a writing center’s routine procedures, it can initiate conversations among consultants about how AI tools are developing the ability to automate certain tasks in their role.

Incorporating AI-assisted notes into writing centers is not a replacement for consultant labor. Instead, the goal is to help consultants “level up” the quality of feedback they provide writers. We now have technology that can automate the session note writing process and produce more comprehensive outputs. While more experimentation is needed to better understand how integrating AI would fundamentally change how session notes are evaluated, tutors can benefit from expanding their AI literacy and use ChatGPT as a training tool. Rather than being fearful of generative AI, writing centers should consider how introducing it can redefine what session notes do, resulting in better resources for writers.

References 

Bender, E., McMillan-Major, A., Shmitchell, S., & Gebru, T. (2021). On the dangers of stochastic parrots: Can language models be too big? FAccT ’21: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency. https://doi.org/10.1145/3442188.3445922

Bermingham, C. (2023a, August 30). Productive and ethical: Guiding student writers in a GenAI world (Part 1 of 2). Canadian Writing Centres Association/ revue Canadienne des centres de rédaction (CWCA/RCCR) Blog. https://cwcaaccr.com/2023/08/30/productive-and-ethical-guiding-student-writers-in-a-genai-world-part-1-of-2/. 

Bermingham, C. (2023b, October 5). GenAI and the writing process: guiding student writers in a GenAI world (Part 2 of 2). Canadian Writing Centres Association/ revue Canadienne des centres de rédaction (CWCA/RCCR) Blog.https://cwcaaccr.com/2023/10/05/genai-and-the-writing-process-guiding-student-writers-in-a-genai-world-part-2-of-2/?fbclid=IwAR24VeL1SqfZK9mq2zQix9U4XlmnyJQy–9yA1utaBHynxnLNnmw8q9JpI.

Bugdal, M., Reardon, K., & Deans, T. (2016). Summing up the session: A study of student, faculty, and tutor attitudes toward tutor notes. The Writing Center Journal, 35(3), 13–36. http://www.jstor.org/stable/43965689.

Chubb, J., Cowling, P., & Reed, D. (2022). Speeding up to keep up: exploring the use of AI in the research process. AI & Society, 37(4), 1439–1457. https://doi.org/10.1007/s00146-021-01259-0. 

Coffey, L. (2023, October 31) Students outrunning faculty in AI use. Inside Higher Ed. https://www.insidehighered.com/news/tech-innovation/artificial-intelligence/2023/10/31/most-students-outrunning-faculty-ai-use

Deans, T., Solod, A., & Praver, N. (2023, August 1). AI in the writing Center: small steps and scenarios. Another Word: From the Writing Center at the University of Wisconsin-Madison. https://dept.writing.wisc.edu/blog/ai-wc/

Dadugblor, S. (2021). Collaboration and conflict in writing center session notes. Praxis: A Writing Center Journal, 18(2). http://www.praxisuwc.com/182-dadugblor

Driscoll, D., & Yim, A. (2024). The loss of we: An empirical investigation of synchronous and asynchronous tutoring experience before and during the pandemic. In S. Rice, J. Katunich, & N. Lape (Eds.). The Post-Pandemic Writing Center: A WLN Digital Edited Collection. Pressbooks. https://ship.pressbooks.pub/thepostpandemicwritingcenter/chapter/the-loss-of-we-an-empirical-investigation-of-synchronous-and-asynchronous-tutoring-experience-before-and-during-the-pandemic/

Giaimo, G., Cheatle, J., Hastings, K., & Modey, C. (2018). It’s all in the notes: What session notes can tell us about the work of writing centers. Journal of Writing Analytics, 2, 225-256. DOI: 10/377514/JWA-J.2018.2.1.09 

Giaimo, G., & Turner, S. (2019). Session notes as a professionalization tool for writing center staff: Conducting discourse analysis to determine training efficacy and tutor growth. Journal of Writing Research, 11(1), 131-162. DOI: https://doi.org/10.17239/jowr-2019.11.01.05

Giordano, J., Hendrickson, J., Mckoy, T., Mina, L., Perdue, S. W., Ruttenberg, J., Wang, Z., Ward, J., & William, J. (2024, April). Generative AI and policy development: Guidance from the MLA-CCCC Task Force. https://aiandwriting.hcommons.org/working-paper-2/

Keller, J. B., Donoghoe, M., & Perry, A. M. (2024, January 24). The US must balance climate justice challenges in the era of Artificial Intelligence. The Brookings Institution. https://www.brookings.edu/articles/the-us-must-balance-climate-justice-challenges-in-the-era-of-artificial-intelligence/

Modey, C., Cheatle, J., & Giaimo, G. (2021). Session notes: Preliminary results from a cross-institutional survey. Praxis: A Writing Center Journal, 18(3), 52-74. http://www.praxisuwc.com/183-modey-et-al. 

Ranade, N., & Eyman, D. (2024). Introduction: Composing with generative AI. Computers and Composition, 71. DOI: https://doi.org/10.1016/j.compcom.2024.102834

Washington, J. (2023).  The Impact of Generative Artificial Intelligence on Writer’s Self-Efficacy: A Critical Literature Review. SSRN Electronic Journal. https://ssrn.com/abstract=4538043.

Appendix A. AI Assisted Note

Hi [Student’s Name],
Today’s session at the Writer’s Workshop was a deep dive into your research paper on the impact of social media on identity perception among female demographics. We explored the complexities of your methodological approach, especially focusing on your survey design and content analysis strategy. Your initiative to analyze the influence of key influencers on platforms like Instagram and TikTok was highly insightful. We also discussed the potential challenges posed by the personalized algorithms of these platforms and how they might affect your research outcomes.

Going forward, you plan to:

• • Refine the methods section to provide a clear and detailed explanation of your research strategies, particularly focusing on your content analysis approach.
    • https://writingcenter.unc.edu/tips-and-tools/scientific-reports/: This resource has some useful information for how to draft a methods section if you want more information.
  • Finalize the selection of influencers for your content analysis, ensuring a diverse representation that aligns with your research objectives.
  • Develop and refine your survey questions to effectively capture the relationship between social media engagement and identity perception among your target demographic.

Thank you for sharing your research. Good luck with the next steps in your project, and I look forward to seeing how it evolves!
Best,
Ava

Appendix B. Non AI-Assisted Note

Hi [Student’s Name]
Thank you for coming to the Writer’s Workshop this afternoon! It was great working with you and learning more about the research you are doing for your class on Writing for the Common Good. Today, you came in with a draft of your research paper, which is an abbreviated version of what you will be submitting in May. After sharing with me the topic you chose, we spent our session discussing the methods section and how you could draft potential survey questions. I appreciated that your Methods section clearly outlined your research plan and included some data to explain why your project is relevant. Please see below the next steps you plan to take.

Revision Plan:

• • Adding to Methods Section: We talked about how it may be useful to provide further justification of why you are looking at Instagram and TikTok (your independent variables) and how they will affect your dependent variables (survey responses). 
    • https://writingcenter.unc.edu/tips-and-tools/scientific-reports/: This resource has some useful information for how to draft a methods section if you want more information.
  • Drafting Survey Questions: While you don’t need to have your questions tailored exactly how you would ask them to participants in your study, sketching out the key things you are looking to ask would be helpful.

Thank you for coming in today and best of luck with the rest of your research! 
Best, 
Ava

Appendix C. ChatGPT Session Note Prompt Example

Write a session note for [Name] based on the attached transcript. The session note will be sent to the instructor, but the student is the primary audience. Write the session note in second person using a friendly but professional tone. It should start with a summary paragraph of about 150 words. The summary should provide a one-sentence overview of the project that we worked on followed by 3-4 sentences describing what we did/discussed in the session. The session note should avoid evaluative language. I want this to be followed by a section that starts, “Going forward, you plan to:” followed by short (1-3 sentences) bulleted descriptions of what the writer should do next.

The end of the note should thank them for coming to the Workshop and wish them good luck with the rest of their assignment.

Write the session note in second person using a friendly but professional tone. Write the session note using the attached session notes written by Ava as models. Try to approximate Ava’s style and tone.

Use the model session note (attached) as a guide for structuring the note.