ChatGPT as a Tool for Library Research – Some Notes on Why I’m Now Hesitant

Nick Pavlovski revisits his 2024 article and optimism.

Posted on June 1, 2026 in Blog Posts

Some of you will remember my previous post from 2024. It was intended to assist academic library staff, especially librarians who used to be known as “reference librarians”—what I still call myself, even though my workplace title is different. At the end of 2023, after a year of self-directed professional development into generative AI tools and their possible use by reference librarians, I presented my findings to an Australian community of academic STEM Librarians. I received plenty of positive feedback and rewrote my presentation into that blog post. In the two-and-a-bit years since, I’ve been trying to keep a finger on the pulse of what is a rapidly widening artery and attempting to find valid uses for ChatGPT and its ilk in my core tasks of finding scholarly information to assist undergraduates, graduates, Ph.D. candidates, and researching academics.

As I’ve done so, I found myself becoming frustrated by the unrelenting and invasive hype about AI, usually most strenuously pushed by AI companies themselves. At the same time, I wondered if there were any opposing voices from inside our profession—I never seemed to hear them. In staying abreast of developments, I found that news articles were beginning to report some negative features and side effects of GenAI usage. Finally, during a training course in 2025, I came across the dissenters I had long hoped existed. So, now I am currently resisting using AI if I do not consider it an improvement to existing tools and processes, and I wish to share what has motivated me to do so.

Where are the dissenters?

Since ChatGPT was released in November 2022, interest in GenAI tools has grown exponentially, and, along with that interest, there have been many voices urging experimentation and adoption. While there will always be technophiles who advise gaining familiarity and then adopting the latest gadget, it seemed to me that voices in the Library and Information Management field were quicker than usual to uncritically experiment and then recommend adoption. The tertiary education sector did have to respond quickly, if only to initially try to prevent student cheating and academic misconduct. In 2023 and 2024, early adopters and high-profile authorities provided helpful and thoughtful advice and recommendations to tertiary academics and academic librarians.

Since then, as database providers have added AI tools to their products and companies such as Nvidia and OpenAI have discussed the benefits of their tools, it has felt like pro-GenAI commentary has swollen from a wave to a tsunami. This left me wondering: How much of this headlong rush into AI is a genuine recognition of the benefits it will bring to library patrons and library staff … and how much of it is raw FOMO (fear of missing out), coupled with a desperate desire to retain relevant in the eyes of other professions and industries?

I lamented not finding opposing voices in the news feeds, emails, and Substacks I receive or subscribe to. Genuine critics seemed absent from the Library and Information Management field. This sat very badly with me, as librarians are usually very strong at evaluating information and making comprehensive assessments of technologies. Why was GenAI going unchallenged? This became a niggling question that popped up and demanded an answer whenever I read the latest announcement about an AI tool or widget being integrated into a database or similar academic product.

I enrolled in some refresher professional development last year about AI and libraries, and in the latter half of that self-paced course, suddenly, there they were—the opposing voices I sought. Dan McQuillan’s “The role of the University is to resist AI” was exactly what I had been seeking: a powerful deconstruction and denunciation of contemporary AI and the social and political forces that insist it is essential. It highlighted the decline in AI users’ critical thinking ability and promoted the incredible value of human imagination. I devoured Violet B. Fox’s “A Librarian Against AI; or, I Think AI Should Leave”. “AI is very bad, actually: A manifesto” by Julie Setele is a single-page zine and mirrors Violet B. Fox. These were the voices I had long sought, and they eloquently and provocatively argued the same points that had germinated in my mind and identified additional points I hadn’t considered but that are equally valuable. They helped bolster my determination to maintain my now slowed pace toward AI adoption in my reference librarian work. I was not completely against AI wholesale, but like Charles Logan with his “Luddite Praxis,” I wanted “a more heterogeneous approach to generative AI, one that makes space for learning with, about, and against.”

---

🔥 Stay up-to-date with LibTech Insights by signing up for our free newsletter. Just one weekly email with our new blog posts, top tech news stories, and other bonus content. Check out some posts from our archive:

• Free and Modified: A Look at Alma Inventory Software
• How to Run an Edit-a-thon for Your Library’s Staff
• Beyond the Job Title: Automation Librarian

📣 Do you teach AI literacy? We want to hear from you. Check out our call for papers, “AI Literacy Across Disciplines.”

---

Brain drain

Criticism without proof or supportive data cannot support such an ethos for long in today’s tertiary education workplaces. Evidence-based practice became important in the 1990s and has permeated everywhere since—for good reason. Looking for evidence or data to support my mindset required daily monitoring and digesting of my existing listservs and feeds, but also publications monitoring tertiary education as a whole and observations made in newspapers and generalist magazines.

As observations and trends began to appear in these forms of literature, one emergent theme is that students are using AI tools and producing excellent work, but those students then couldn’t explain how AI’s decision-making process resulted in the arguments or logic in their work, leading their college instructors to worry about those students’ inability to think critically for themselves. This is causing tertiary education institutions to change their methods of assessment.

MIT Media Lab went further than just looking at qualitative feedback and freehand comments. They used electroencephalography (EEG) to record participants’ brain activity when writing academic essays, and the participants were split into three groups: LLM use permitted, search engine use permitted, and neither permitted. LLM users showed weaker neural connectivity, under-engagement of alpha and beta networks, and couldn’t quote from their own essays they had written only minutes before. Nicholas Carr asked in 2008, “Is Google making us stupid?” Instructors and researchers are now rightly asking, “Is AI making our students stupid?”

The workers have nothing to lose but their chains

A second theme has also emerged: worker fatigue and possible burnout. I sat bolt upright when no less than the Harvard Business Review in March this year reported that workers using AI realized the tools had allowed them to add more tasks to their day when they should have been on nonwork breaks such as lunch. Workers further reported that AI use was blurring the boundaries between work and nonwork. Work overall was increasing, not decreasing! The CSIRO here in Australia has also reported similar effects.

Within our own profession, academic librarian burnout is not a rare or isolated problem: there are books, articles, and blog posts about it. Greyson Pasiak synthesized this and other issues they had with AI to produce a strong dissenting voice about AI, LLMs, and their place in academic library work—yet another voice that deserves more recognition in the library world. They report on how the concept of Information Fatigue, first discussed in the 16th century, has spawned what is called “AI Fatigue.” This Fatigue is exacerbated by the constant FOMO in academia, wonderfully summarized as “get on board or get left behind.” Greyson then observes how job creep as a result of GenAI has affected their own work, and concludes by calling for tertiary education institutions to pause and consider the emotional and mental costs these tools are inflicting on their workers.

Recolonizing information?

A largely ignored third theme is that AI tools reinforce colonialism. The datasets that AI tools are trained on often perpetuate white colonist knowledge and attitudes due to their sources, part of which relies on out-of-copyright content, historical information, and biased eyewitness accounts from white colonial authors. First Peoples’ knowledge in these AI tools datasets may therefore be very distorted, if present at all.

This was borne out in the Learning and Teaching Virtual Workshop for Australasian STEM/Engineering Librarians, held on December 3, 2025. Alissa Hackett, a proud Maori person and STEM library colleague from across the ditch in New Zealand, ran a frighteningly elegant demonstration of how LLMs perpetuate the biased information and misinformation of white colonizers. Titled “Whose knowledge gets retold? A live experiment investigating digital colonisation in GenAI tools,” Alissa’s demonstration involved asking workshop participants to interrogate an LLM regarding Maori origin stories about how a set of mountains in New Zealand came to be named. Follow-up prompts did not help clarify the lack of information, or misinformation, that the LLMs returned in their outputs.

Participants at the Workshop expressed their amazement and concern during an open discussion at the conclusion of Alissa’s session, again at the wrap-up at the end of the workshop, and yet again in post-workshop feedback. This means that the academic librarians and academics who rely on these tools may miss vital information and vital context concerning First Peoples, which, in turn, may affect the success of research grant proposals or the successful establishment of partnerships with commercial groups.

“Bah! Humbug!” – Well, not quite: A conclusion

There are bound to be more reports on how the use of GenAI and LLMs is affecting student learning and librarianship at tertiary education institutions. The fact that there is a small but growing set of articles like the one published in the Harvard Business Review proving that the AI hype pushed by OpenAI, Nvidia, and their ilk is not actually bearing the workplace fruit they insist it should is reassuring and will hopefully grow.

Beyond the predicted and actual effects on productivity, what academic librarians should strongly consider when deciding whether to implement GenAI in their work is the effects of intensified workloads and AI Fatigue, both of which accelerate librarian burnout. They should also consider the moral and ethical issues regarding how AI companies deliberately ignored copyright (which we librarians have to be so careful to respect) in order to “move fast and break stuff” to train their products (there are plenty of articles in the literature about this), and the contradiction of academic librarians working diligently to decolonize their collections, only to have GenAI and LLMs recolonizing our students and academics. Some AI companies may work to decolonize their training sets, but there are others that may not.

This leaves me exercising considerable caution in using GenAI and LLMs in my work as a reference librarian, and choosing very carefully how to instruct students to use it when beginning their searches for content for literature reviews or using it to review papers. “Make haste slowly” seems to sit best with me—at least for now.

Related Posts