Last week I attended the Charleston Library Conference. As I listened to the conversations, I was struck by how often the focus was on the generative AI chatbots writ large and how little I heard about efforts to understand how researchers and learners are engaging with AI technologies and tools or about investigations into changing information practices.

That absence stood out as I remembered back to when web-scale discovery layers came on the scene. The conversation then was not only about products. Lorcan Dempsey’s essay “In the Flow,” the Studying Students initiative at the University of Rochester Libraries, and numerous other efforts had helped focus our attention on library users and understanding the practices and needs of our particular campus communities. 

Libraries and publishers both need to understand how researchers and students use AI, but I’m concerned that both are focusing most of their attention elsewhere — on generative AI chatbots. Libraries want agency in the campus AI environment, but they do not run the enterprise AI for their campuses (and few would want to). It can be difficult to determine who is leading campus efforts and how to engage with them. Existing license terms can be a barrier to integration, and shrinking budgets further limit opportunities.  Publishers have parallel challenges. They are deciding on AI licensing terms, which AI features, if any, to add to their platforms, and whether they can develop a sustainable business model. Each sector’s focus is understandable, but taken together, these two priorities leave a gap.

Researchers and students are adapting quickly, without waiting for guidance from either of us. When libraries and publishers both organize their strategies around AI tools, no one is systematically studying the people using AI. This is not to suggest that librarians and publishers should ignore the AI tools. But engaging with the tools is not the work that will determine whether libraries and publishers succeed in serving researchers and learners.

I believe that libraries and publishers can claim agency in the AI age by once again investigating and deeply understanding how our users work. Below, I describe what that inquiry could look like and suggest next steps for libraries and publishers.

User-Centric AI Strategy

The discovery-layer era supplies a precedent for approaching the AI age. At that time, libraries were deeply engaged in investigations of users’ information work practices: how people searched, where they got stuck, and what they did when library tools failed them. Libraries used what they learned to decide how to configure platforms and evolve library services. 

Some publishers and platform providers conducted similar investigations into user work practices and user experience. For example, EBSCO’s research found that college students skimmed, scanned, and favored efficient search strategies over deep exploration of results. Findings such as these helped shape how products were designed. 

Libraries adopted tools when they matched information workflows, and adjusted them when they did not. The discovery tools mattered, but the findings about users were the focus, and decisions followed from them.

The same logic holds for AI.

For libraries, the question is where library work belongs in a research workflow that is increasingly mediated by AI. A library cannot answer that without knowing what researchers and students delegate to AI, how they verify the results, and what they expect from the library. Librarians developing their AI strategy without understanding their user communities risk designing around library staff preferences and AI vendors’ product plans rather than user needs. 

For publishers, the question is whether the AI features they build into their platforms fit how people want to work. Users may want to turn features on or off, to see how an answer was produced, or to bypass a feature entirely. Products designed without that knowledge risk solving a problem that users do not have, or imposing choices that users would rather make for themselves.

In both cases, the AI vendors cannot supply the answer. Their expertise is in their tools, not in our users’ practices.

Shared Inquiry Agenda

Libraries and publishers share an interest in the same questions, which means we can also cooperate in our investigations:

• Which tasks and decisions do researchers and students trust to AI?
• When do researchers and students intentionally keep AI out of a task or decision?
• When and how do users check AI output?
• What controls do users want over AI features, and what settings do they want to see as the defaults?
• What do users want from libraries and publishers, both with respect to AI and in areas where AI cannot help?

The methods for these investigations are familiar: interviews, observations, diary studies, and analysis of how people use existing services. Libraries have staff experienced in this work, and many publishers have usage data and user research teams. Neither sector sees the whole picture on its own. Libraries see campus-level practice and context, and publishers see patterns across institutions. Together, these perspectives would give a fuller picture of how people work with AI. 

This work has challenges, of course. User practices are changing rapidly, and methods that rely on self-reporting have known limitations, including lack of awareness and failure to report candidly.  It may also be challenging to identify the capacity needed to do this work.

Next Steps for Libraries and Publishers

Some libraries and publishers are already doing this work, and in such cases it deserves a wider audience for greater impact. But where they are not, engaging in it should become a priority.

For libraries, starting with the inquiry questions, consider how to:

• Study AI practice in your user community. Interview or observe a small, varied group of researchers and students about what they delegate to AI, how they verify results, and where they still turn to the library. Start small, but start. 
• Share the findings with those leading campus AI initiatives. Campus IT and research computing units are building AI infrastructure, but it can be unclear who is providing vision and leadership for the campus. Sharing insights about users offers a basis for integrating library resources and services into campus platforms.
• Use the findings to evaluate vendor features. For each AI feature a vendor offers, librarians should ask what it assumes about how people work, and check that assumption against what is observed in their own user community.
• Connect the research to staff development. Staff who serve users should learn from the findings, so staff training and development reflect user needs and practices rather than product announcements or personal preferences.

Publishers may be facing pressure to move quickly on AI, and releasing an AI feature can itself generate useful information. The risk is that the research follows the release instead of informing it. These steps could help:

• Ask users about AI features before release. Find out whether people want a feature on by default, whether and in what form they want to see how an answer was produced, and when they want to switch AI off. These preferences may vary across user groups, and exploring those nuances will strengthen decision-making.
• Treat user research as a shared resource. Publishers see patterns across institutions that a single library cannot see. Sharing aggregate findings with library partners (and vice versa) would let both sides see what has been observed and negotiate shared commitments.
• Explain design choices to library customers. When a platform adds AI features, share the evidence from users that supports the choice. Libraries are being asked to make decisions about these features, and that kind of information will enable them to do so in an informed and therefore confident way.
• Invite libraries into the research. Librarians know their campus communities, and they can help recruit participants and interpret findings in context.

Learning Alongside Our Users

The discovery era also required libraries to retrain staff to think about discovery and workflows differently. Publishers faced a parallel need to rethink how their products fit into those workflows. The AI transition asks both sectors to make the same adjustment, under greater ambiguity and at greater speed. Our users appear to be leaning into that ambiguity and adjusting their work patterns rapidly, and as a result library staff and product teams should be doing the same.

Staff training and product development cannot be built on certainty, because none exists yet. Approaches built on inquiry fit the moment better: taking time to try tools, a habit of recording what we observe, and a route for those observations to reach the people who design services and products. This is also a time of great experimentation, so we should avoid becoming prematurely fixed in our thinking and decisions.

The Path to Success

The measure of success for libraries and publishers is the same, and it has not changed. It is how well we serve our user communities. AI vendors will shape the tools we use, and we should pay attention to our relationships with them. But we should judge those tools and relationships only against what our users do and need.

Acknowledgments and Disclosures: I am grateful to colleagues and friends who reviewed drafts of this essay, provided feedback, and helped me clarify my thinking. I used Claude as a conversational interlocutor in developing this essay and Grammarly to polish the writing.