December 1, 2025

By Dr Rachel Schulkins, NAG Chair and Head of Subscriptions and Publisher Relations, University of Manchester.

Throughout my career in libraries, I’ve been driven by a fundamental question: how do we adopt, deploy, and benefit from emerging technologies to improve our services and strengthen our teams? Today, as generative AI rapidly evolves – reshaping knowledge, information and even our profession– that question feels more urgent, and more contentious, than ever.

Recently, during one of its Coffee and Chat sessions, the National Acquisition Group explored the role of AI in acquisitions workflows. The conversation revealed a wide spectrum of views. Many—perhaps most—expressed strong opposition, insisting that AI has no place in acquisitions or broader library workflows. Others raised vital ethical concerns, questioning whether libraries should contribute to the growth of commercial AI giants, and highlighting the environmental costs of AI in relation to sustainability goals.

What emerged was a clear tension between the societal role of libraries and their engagement with AI. Some questioned what might follow when the current AI bubble inevitably bursts. These are valid questions our sector continues to grapple with. The ethical dimensions—environmental sustainability, the dominance of commercial giants, and the ethics behind training data—deserve serious attention, not dismissal.

However, AI is advancing whether we engage or not. Publishers, vendors, and platforms are already shaping workflows and user experiences. If we, as professionals, disengage, we lose our voice in how these tools evolve. Yet, if we engage uncritically, we risk compromising our values.

I don’t believe this tension is easily resolved. But I do believe that informed engagement—testing, questioning, and understanding these tools—gives us the knowledge base to advocate effectively for responsible AI development.

Whether we’re ready or not, AI is being woven into the fabric of library services. Publishers are integrating AI into peer review processes.[1] Suppliers and platform providers are developing embedded AI tools.[2] EBSCO introduced AI Insights and Natural Language Search to analyse user needs and enhance research workflows; Kortext has developed AI-powered study tools to support student learning journeys; and Clarivate is exploring AI applications in both metadata enrichment and acquisitions management.[3] The recent partnership between Wiley and Anthropic signals yet another shift in how content is created, accessed, and consumed.[4]

In acquisitions specifically, AI capabilities proven in corporate procurement—price prediction, vendor selection, duplicate detection, budget forecasting—haven’t yet fully arrived in library systems.[5] Clarivate’s Alma AI Assistant for funds queries represents an early step, but most acquisitions work still relies on traditional methods. Yet the trajectory is clear: these tools are coming. The question isn’t whether AI will touch library acquisitions, it’s how we prepare now to shape that integration so it serves our professional values and user needs rather than simply replicating corporate priorities.

The Historical Echo

This isn’t the first time our profession has faced technological disruption. From print to eBooks, the rise of digital databases, the emergence of open access publishing, and the integration of learning technologies, have all challenged traditional knowledge routes. Each shift sparked concerns about authority, access, and expertise, and required librarians to develop new skills, rethink and reassert our role in guiding users through increasingly complex information landscapes.

As a Gen X’er, I’ve lived through it all—from massive box TVs to flat screens, from dial-up to broadband, landlines to mobiles and smart tech, and from the dire warnings about Wikipedia corrupting learners’ minds to today’s concerns about AI hallucinations. At a conference some time ago, someone said something that resonates deeply with how I see this moment: technology will always change and evolve, but the skills we require as users will remain the same—critical thinking and critical reading. Those are the skills we must continue to cultivate, both for ourselves and for our Library users.

Our Role as Librarians

As librarians working in a profession centred on information and knowledge, I believe that our role is to guide, explain, and equip. Whether discussing AI, databases, or the written word, our role is to help our users to assess information, question it, and interrogate it in ways that generate curiosity rather than passive consumption. In an age of misinformation, disinformation, and AI-generated content, these skills are more crucial than ever.

The NAG AI Working Group: Learning by Doing

These reflections, tempered by curiosity about what’s possible, led to the formation of the NAG’s AI Working Group. We wanted to understand how AI tools could support acquisitions work, improve efficiency and automate repetitive tasks so staff can focus expertise where it matters most.

The group is deliberately small and experimental at this stage—and we’re actively looking to expand it. Rather than waiting for perfect answers, we’re learning by doing: testing tools hands-on, identifying practical use cases and limitations, and gradually building shared knowledge that could benefit the wider NAG community.

Our scope is deliberately broad, focusing on areas where acquisitions staff spend significant time and where AI might offer support: data analysis, metadata enhancement, workflow documentation, staff training, donation evaluation, classification assignment, location allocation, and text accessibility.

As we continue this work, we’re planning to develop resources that could help others navigate similar decisions—prompt libraries for common tasks, best practice guidance rooted in what we’re learning, and ethical checklists to help institutions think through their own AI adoption carefully. If you’re working on AI questions in your own institution, or curious about what’s possible, we’d welcome more voices in this conversation. The more diverse the perspectives and institutions involved, the more useful our collective learning will be. For more details, please contact nag.office@nag.org.uk

One thing we are not trying to do is develop organisational AI policy- that remains an institutional responsibility. But we do hope our efforts might contribute some insights to those larger conversations.

Case Study

When I set up the working group, I had a specific challenge in mind from that list: classification number assignment. At my previous institution, we had an in-house author cutter system that required manual checking and cross-referencing to generate author classification numbers. I wanted to explore whether AI could streamline this process. What followed was a six-day intensive experiment that taught me more about AI’s potential and limitations than any theoretical discussion could have.

For those unfamiliar, cutter numbers are the codes that arrange books alphabetically by author or title within a subject classification. They’re essential for shelf organisation. The system used for this experiment was developed in-house by the Library and required manual consultation.

To automate the process, I started simply by sharing the cutter tables and instructions to AI platforms (Claude and ChatGPT) and asking them to generate numbers for test titles. However, the free versions hit limits quickly—conversations cut off mid-process, upload sizes proved too small, and the AI frequently hallucinated numbers that didn’t exist in our tables. In another, it inserted numbers that had no basis in the cutter table at all.

These failures were instructive; I had to refocus and define the core issue I was trying to solve.

The Breakthrough: Structure and Specificity

The project transformed when I shifted focus. Instead of asking AI to reproduce full Library of Congress classification numbers (which usually come pre-supplied in records anyway), I concentrated specifically on cutter number generation—the part that requires staff time.

By converting tables to JSON and focusing on specific tasks (monographs, literature, geographical works), results improved dramatically. Using GitHub Copilot, we developed a modular web application that semi-automates cutter number generation while preserving human oversight.

The tool is now live, with the code repository at GitHub for those interested.

Although it still requires refinement, and if I were still in post I would have taken it through full testing and implementation, the potential is already clear.

The process was crucial in drawing basic insights:  

1. Be precise,  focused and clear: Once I defined the logic and included edge cases, AI handled the task reliably. This shows potential for automating other structured library tasks.

2. Clean data is essential: Structured, clean data is valuable for success.

3. Semi-automation beats full automation. My tool doesn’t eliminate human oversight  but it handles the mechanical lookup while staff verify results and manage exceptions. This hybrid approach mirrors what’s realistic in acquisitions and with the limitations of AI.

4. AI can support learning. By explaining why it selected specific cutter numbers, the tool can help staff learn the rules while using them. This “teaching while automating” approach addresses a concern I hear frequently that AI will dilute skills and expertise. But can it do the opposite?

5. Free tools have hard limits. They are great for learning and experimentation, but not for production workflows with sensitive data. For serious use, institutional licenses are essential—which has budget implications for libraries considering AI integration.

6. Collaboration is key: At several points, I needed support from colleagues with IT expertise—especially when my own understanding didn’t match the technical demands of the task. Developing tailored AI tools requires cross-team collaboration, drawing on skills from across the library to ensure the tools are both functional and fit for purpose. AI can help, but it cannot replace the depth of expertise needed to work effectively with it.

AI tools can offer meaningful support in professional settings, but their limitations are just as important as their strengths. They often require human oversight, especially when dealing with context-specific or nuanced decisions. Successful implementation depends on ongoing refinement and maintenance, not just initial setup. Ultimately, while AI can streamline routine processes, complex judgment calls and strategic thinking remain firmly in the domain of human expertise.

Moving Forward: Fear, Engagement, and Understanding

So where does this leave us?

It’s right to be cautious about AI. The environmental toll of powering data centres, the concentration of control in a few tech giants, and the troubling labour and data practices behind AI development all raise serious ethical questions. These concerns aren’t abstract—they directly challenge values around fairness, sustainability, and intellectual rights, and deserve our scrutiny.

We have a responsibility to name the above issues, discuss them openly with our colleagues and users, and to advocate for better practices in the AI tools we do adopt. At the same time, we need to engage with AI, because it’s already shaping our professional and personal environments. Disengagement hands over all decision-making to vendors and publishers who may not share our values. And while we must remain critical, we should also acknowledge that AI can support a wide range of reading and learning styles, by enhancing accessibility, personalising content, and helping bridge the gap between print and digital formats.

In acquisitions specifically, we’re being asked to do more with less. Our workflows are growing more complex, budgets are tighter, and user expectations are shifting. AI tools can offer potential to support some tasks, but these possibilities must be approached thoughtfully. We need to ask: What tasks are appropriate for automation? How do we verify outputs? What are the implications for data privacy and professional standards? And what happens when AI recommendations conflict with our values — do we have the expertise to understand and override them?

Engaging critically means testing tools rigorously, documenting what works, sharing knowledge, and investing in staff training, not just in using AI, but in understanding how it works. It means preserving human judgment where values and community needs are at stake, and advocating for ethical development in our professional spaces. This isn’t just about meeting expectations—it’s about being informed professionals who can guide conversations with nuance and integrity.

My own project didn’t solve cataloguing or acquisitions challenges, but it gave me hands-on insight into what these tools can and cannot do. That knowledge—earned through experimentation, failure, iteration, and modest success—feels more valuable than any position of pure acceptance or rejection. As we navigate these challenges together, I’m hopeful that by engaging thoughtfully, questioning critically, and learning continuously, we can harness the benefits of AI while remaining true to our professional values. The technology will continue to evolve. Our commitment to ethical practice and critical thinking must evolve alongside it.

[1] Kousha, K., & Thelwall, M. (2024). Artificial intelligence to support publishing and peer review: A summary and review. Learned Publishing, 37(1), 4-12.

[2] Enis, Matt, AI on the Horizon, 2024.08.05; EBSCO Information Services. (2024, August 5);.

[3] Ex Libris. (2025, July 11). Artificial intelligence in Alma. Ex Libris Knowledge Centre; Inside Higher Ed. (2025, March 18). Publishers adopt AI tools to bolster research integrity; Kortext Premium launches: Generative AI in higher education. Kortext Blog; Library Technology Guides. (2025, February 6). AI Metadata Assistant headlines Alma’s February 2025 feature release; ProQuest. (2025, April 14). New Ebook Central AI tool helps users engage deeper with ebooks. ProQuest Blog.

[4] Wiley Partners with Anthropic to Accelerate Responsible AI Integration Across Scholarly Research. John Wiley & Sons, Inc. (2025, July 9).

[5] Art of Procurement. (2025, October). State of AI in procurement in 2025; Spendflo. (2025, October). AI and machine learning in procurement: A game changer for efficiency and cost savings.