ITHAKA’s Research on Generative AI and Its Implications for Librarianship

Preparing for the future of scholarly publishing

Posted on June 2, 2025 in Blog Posts

The library world is abuzz with generative AI (GAI) discussions. But one voice that deserves a spotlight among the chorus is that of Ithaka S+R. Most people know ITHAKA as the nonprofit behemoth behind JSTOR, but its “strategies and research” service also works with libraries and the larger higher educational community. Its programs and research aim to improve student outcomes, preserve and promote scholarship, and educate higher ed leadership. Ithaka S+R has held many events and produced many materials on GAI, but two of its research papers and an accompanying project are of vital importance to the discussions librarians are having about how AI will affect our work and the scholarly record, which we help to share and curate. 

GAI and Scholarly Publishing

Ithaka S+R has produced the report, “A Third Transformation? Generative AI and Scholarly Publishing,” based on interviews with GAI strategists who work for major scholarly publishers and with academic scholars. The report focuses on how GAI will accelerate scholarly publishing and how it might disrupt or transform current practices. It discusses the use of GAI in discovery, peer review, and language translation, as well as the licensing of scholarly content for use in LLMs. Overall, the interviewees speak very positively about GAI, although they do mention concerns about a lack of attribution, the importance of guidelines for writers’ AI use, and the impact of changes on smaller or open publishers. 

While the whole report is valuable to librarians who are concerned with scholarly communications research, some elements are particularly relevant to discussions on library practice and information literacy. The report speaks to how GAI is likely to affect the processes of discovery, summarization, results evaluation, and synthesis of literature reviews. For example, current tools can summarize articles and put them in context with each other, providing a narrative answer to a search instead of a list of article results. This, in turn, could lead to a decrease in actual readership of the summarized articles.  Although the paper doesn’t highlight these functions particularly, I believe these are among the core elements of how libraries serve both students and faculty. We help users find papers through discovery systems and the creation of metadata, teach them information literacy tools to help them evaluate what they find, and help users (particularly faculty users) understand the interconnected landscape of the discourse in their topics by tracking down literature review citations and helping with comprehensive searches.

---

🔥 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:

• Catching up on Big Tech Antitrust Cases
• Elevating the AI Conversation with the AID Framework
• 5 Upcoming Academic Books on AI, Social Media, and Misinformation

---

GAI Tool Types

Another Ithaka S+R report, “Generative AI in Higher Education: The Product Landscape” and its companion project, the Generative AI Product Tracker, offer details on the tools being developed to manage these library-centric functions. The product tracker lists and describes generative AI tools. It aims to include all tools that are developed specifically for faculty and students or that are widely used within higher ed. It also includes tools in development. 

The paper describes the overall trends that emerge from the product tracker and provides a very helpful categorization of the tools. The paper divides tools into “discovery,” “understanding,” and “creation” functions. (The actual tracker uses different but related categories.) “Discovery” tools primarily help users by allowing natural language search queries against scholarly databases and by narrowing results. The “understanding” tools are inextricably related, as they present summaries of the papers that are returned by the discovery tools. Traditional library e-resource vendors (aggregators and scholarly journal publishers) are moving to create tools that summarize papers or produce snippets that are more concise and easier to read than abstracts. These same tools provide article citations to support multipaper synthesis summaries, although these citations still don’t always match the actual content of the summaries. 

The tools can also provide other insights into the results. Notably, Elsevier’s Scopus AI suggests follow-up questions after a search and attempts to identify leading authors in a field. The report discusses many other current and developing tools, including several that use exclusively open scholarship, mostly based on the Semantic Scholar corpus (all the LLMs are believed to have been trained on open content). The report and the evidence of the tracker suggest that these types of usages are already a done deal and that the GAI’s creation of literature reviews is in progress and inevitable. 

“Creation” is the third category the report identifies. Many of these tools are more focused on classroom teaching and learning functions than on library research specifically. They assist with brainstorming, producing review and quiz materials, generating images, and providing editing advice for academic writing. 

The report concludes by addressing OpenAI’s dominance in the field as the backend for many tools and discussing how higher ed must consider how it might or might not be able to shape GAI use within its community. Practically, institutions can set local guidelines for which specific actions are permissible uses of GAI (e.g., editing versus generating text). But holistically, schools must grapple with the cumulative effects of GAI’s integration into learning management systems and school IT infrastructure, a process often driven somewhat externally by vendors and developers.

GAI Product Tracker

The product tracker is a valuable resource, which is still being updated as of spring 2025. While all its tools are important to discussions of scholarly publishing and higher education, the “discovery tools” and some of the “research workflow tools” are encroaching on the core tasks that libraries help their users accomplish: helping people find resources, evaluating these resources, selecting relevant materials, and synthesizing scholarly literature. 

Of note are the tools produced by Elsevier and Clarivate. They already have well-developed tools that are integrated into library subscription database options, so many libraries are already faced with purchasing decisions and changes to reference services. Scopus AI has many reviews, and it was among the first library subscription database tools available. Elsevier now also offers Science Direct AI as a subscription add-on, which extracts and summarizes database content. Clarivate was also quick to incorporate AI tools into library databases, offering similar GAI tools in their article databases and in Web of Science. Clarivate also offers a GAI Research Assistant in its Primo service, which runs the unified search discovery systems of most US academic libraries, although uptake has been slow. 

The ITHAKA Product Tracker helpfully provides a way to learn the essentials about these tools. I should also mention JSTOR’s own GAI interactive research tool (currently in beta with limited availability), which summarizes content and suggests topics. While these tools are integrated with library products, many similar tools are independent. Several tools in the “research workflow” category summarize and help make PDF articles easier to work with, such as TLDR This and ChatPDF, which allow users to “chat” with a PDF. Others summarize papers and organize citations (Scholarcy) or provide a workspace with the ability to query saved documents (Notion).

GAI and Librarianship

ITHAKA’s works highlight the acceleration of scholarly publishing and changes in the process of literature review creation, both of which have very practical meaning for librarians. It is yet to be seen if GAI will allow researchers to maintain the quality of their work while improving quantity. I fear the creation of a bloat of low- or mid-quality papers, which we will have to help our users sift through. Researchers who are striving to improve the efficiency of their research and writing might not realize that GAI could harm the quality of their work, even if they are committed to ethical GAI usage. For example, if search queries deliver summaries of works, it might be easy to rely too heavily on such results for both understanding a topic and producing literature reviews. Even if such summaries were perfectly accurate (and they are not), a summary removes vital context that resides in the details of papers. Ultimately, less actual reading could lead to less direct “discussion” between papers and distort the scholarly dialogue, which is fundamental to inspiring, evaluating, and creating research. 

However, as I review the ITHAKA research, I see hints of how librarians can learn to work with these realities. We need to remind all our users, even experienced faculty, that the basic tools of information literacy are still valuable: they just need to be applied to GAI results and used explicitly. An experienced researcher knows how to evaluate an article by looking at the quality of the journal it is in, examining its citations and where it has been cited, and checking the affiliations of its authors. With GAI, these facts are hidden behind a summary, but they can still be evaluated if users react mindfully to the summaries they receive and remember to check these factors on suggested articles. Our experienced users know how to use a citation to find an article; now we need to remind them to use the same skill to test the validity of GAI citations for hallucinations. 

Researchers must learn to fact-check GAI results just as they would fact-check a slanted article. Librarians can remind users to rely on prior contextual knowledge to spot potential hallucinations. 

Obviously, GAI calls for new skills, but so many of the skills we already teach can be repurposed. I fear the academic library community may struggle to remind users to engage critically with materials and GAI outputs with the same care they use now with search results. But I hope that we can use our roles as traditional and trusted mediators of the academic record to influence how GAI is used with our materials. 

Related Posts