As the integration of artificial intelligence (AI) rapidly advances, academic libraries are increasingly pivotal in supporting AI literacy among students and faculty (Mairn, 2024). This study provides a novel examination of how U.S. academic libraries, particularly those affiliated with the ARL and Oberlin Group, address generative AI through their library guides (hereafter referred to as LibGuides). Employing content analysis guided by the EDUCAUSE AI Literacy Framework, this study focuses on 70 recently created LibGuides dedicated to generative AI.

The originality of this study lies in the adaptation of the EDUCAUSE AI Literacy Framework, developed for Barnard College to guide AI literacy in educational settings (Hibbert et. al., 2024), within the context of academic library guides. This framework, which comprises four levels: understanding AI, use and apply AI, analyse and evaluate AI, and create AI, was reorganized to align more closely with the specific characteristics and objectives of LibGuides.

The present study holds strong implications for advancing AI literacy in higher education. By reorganizing and enhancing the EDUCAUSE AI literacy framework, it proposes a more comprehensive framework tailored specifically to the needs of higher education institutions. This adaptation addresses gaps identified in the original framework, offering a nuanced approach to AI literacy that reflects the unique challenges and opportunities faced by academic libraries.

Additionally, the study highlights on the evolving role of librarians in the higher education setting and their collaboration with faculty, AI experts, and other campus stakeholders in promoting AI literacy. Understanding how librarians and campus stakeholders may share responsibilities in disseminating AI knowledge provides valuable insights into how these roles can work together more effectively. This may pave the way for future research on optimizing collaboration between librarians’ and other campus members’ roles to address the broader impact of AI.

This study also identifies limitations in librarians’ current approach, such as the often-limited technical depth in their guides. These findings highlight the need for further research into librarians’ competencies and the future training required to meet the growing demands of AI literacy education. Addressing these limitations is crucial for enhancing the effectiveness of library resources and services while empowering librarians in their key role in advancing AI literacy.

Generative artificial intelligence (AI) creates new images, text, and software by learning from existing data without directly replicating it. Although AI development has been ongoing for years, the launch of ChatGPT in November 2022 captured global attention due to its ease of use through natural language interaction. While generative AI offers considerable convenience, it also raises critical concerns, including issues of transparency, accuracy, bias, intellectual property disputes, cybersecurity risks, and sustainability challenges (Gartner, 2023). The rapid proliferation of AI tools for writing, coding, translation, information retrieval, illustration, and presentation creation also poses challenges in higher education. Not all campus stakeholders are familiar with or have access to these tools (Henke, 2024), potentially exacerbating disparities in learning opportunities. Furthermore, while it is recognized for enhancing productivity, concerns have surfaced regarding its effectiveness in creative and intellectually demanding fields, as well as its potential to facilitate plagiarism, which has heightened academic integrity concerns (Mogavi et al., 2024). According to EDUCAUSE’s ‘AI blueprint report’, a survey of 910 higher education professionals conducted in late 2023, 78% of respondents believe generative AI will impact academic integrity and the design of assignments, assessments, and curricula. Additionally, 73% reported that their institutions have already implemented related strategies, and over half have started offering education and guidance on the subject (Robert, 2024).

Libraries have long played a vital role in supporting teaching and research in higher education, serving as key partners in the adoption of technology and information literacy education. In ‘Developing a library strategic response to artificial intelligence,’ Cox (2023) conducted a survey of 111 libraries in the UK and identified the perceived impacts of AI on various library functions. These include using AI to enhance metadata, improve discovery and retrieval processes, and implementing AI-driven tools, such as chatbots for user services. The paper outlines three key strategies for libraries:

using library AI capabilities to model responsible and explainable applications of descriptive AI, leveraging Librarians data competencies to enhance organizational AI capability, and promoting AI literacy to enhance organizational and societal AI capabilities.

AI literacy is emerging as a critical area of focus for libraries around the world, driven by the increasing integration of AI technologies into information services and academic environments. Many information literacy instructors have begun exploring ways to transition into teaching AI literacy (Hervieux & Wheatley, 2024). Additionally, numerous librarians have designed AI literacy courses based on the core principles of information literacy and shared their approaches (ACRL, 2024).

An increasing number of academic librarians are developing library guides tailored to specific subjects or courses, assisting users with database searches, and recommending relevant resources. These guides serve as valuable tools for teaching information literacy and can, at times, have a greater impact than traditional classroom instruction (Duffy et al., 2021). Ko and Chiu (2024), in their analysis of 28 library guides using the ACRL framework, also found that librarians effectively applied the core concepts of information literacy to the introduction of AI. Currently, these guides present a timely opportunity to help build, promote, and advance AI literacy. By building on their foundation in information literacy, librarians can provide essential support for higher education institutions as they navigate broader AI policy development and respond to the growing demands of the AI era.

The concept of AI literacy was first introduced in 2015, and scholars from various fields have continued to explore its definition. At its core, AI literacy centres on equipping individuals with the ability to critically assess, engage with, and make use of AI technologies, even if they do not have the skills to develop AI models themselves (Chan, 2024). According to a review by Lintner (2024), although many AI literacy scales have been developed, their reliability and validity still require further verification. These scales share common elements, particularly in their emphasis on a basic technical understanding of AI, awareness of its societal implications, and a focus on AI ethics.

Meanwhile, Talagala (2021) introduced the concept of context of AI application, suggesting that individuals from different fields and with varying levels of prior knowledge will require different aspects of AI literacy. Pinski and Benlian (2024) provide an overarching conceptual framework for AI literacy, synthesising findings from 68 academic papers. Their work critically examines the specificity of AI literacy to different user groups and distinguishes it from other technology literacies. The conceptual framework is structured around the categories of ‘Proficiency Dimensions’ and ‘Subject Areas’, which encompasses the content areas relevant to AI literacy: AI models; data for AI; AI interfaces; AI tools; humans, organizations, and society; and cross-area. The cross area includes various literacy skills, including ethical literacy, critical literacy, meta-literacy, and future literacy.

Ng et al. (2021) explored the concept of AI literacy and provided a foundation for defining, teaching, and evaluating it in educational contexts. Through reviewing 18 peer-reviewed articles published between 2016 and 2021, the authors identified four key aspects of AI literacy: ‘know and understand AI’, ‘apply AI,’ ‘evaluate and create AI,’ and ‘AI ethics’. The first three of these components were aligned with Bloom’s taxonomy. A few years later, Ng et al. (2024) developed an AI literacy questionnaire (AILQ) and proposed a model based on the ABCE (affective, behavioural, cognitive and ethical) to evaluate students’ literacy development. The authors’ framework lays the foundation for other frameworks, including the EDUCASE AI literacy framework, offering educators information on how to foster AI literacy in students.

In June 2024, EDUCAUSE published its AI literacy framework, offering a structured approach to understanding AI literacy in educational contexts. This framework is designed to guide institutions in developing AI education programs and is intended to meet individuals where they are scaffolding upon their current level of AI literacy. These levels are not strictly sequential; individuals can engage at any level based on their prior knowledge and goals (Hibbert et. al., 2024). The framework breaks down AI literacy into four levels:

The following are the research questions our study is pursuing:

RQ1. For librarians developing AI literacy LibGuides, how could the EDUCAUSE AI Literacy framework be modified to be optimally suitable for libraries’ user needs?

This study employs content analysis, guided by the EDUCAUSE AI literacy framework, to examine LibGuides created by academic libraries in the USA affiliated with the ARL and the Oberlin Group. The focus is exclusively on newly created guides specifically addressing generative AI. Guides in which librarians simply added information to existing content or those that are simply compilations of resources without explanatory text were excluded from the analysis, as they were deemed ineffective in conveying AI literacy. Furthermore, to avoid potential disciplinary biases, guides that introduced generative AI from a subject-specific angle were excluded.

By reviewing 114 ARL members in the United States and 83 members of the Oberlin group, the researchers identified 48 ARL libraries and 22 Oberlin libraries that created new LibGuides for AI, meeting the criteria for analysis. The content of these 70 guides was downloaded by the researchers, both having backgrounds in library and information science, between June 24 and June 30. The subsequent analysis was conducted between July and August 2024.

Among the 70 libraries analysed, several produced multiple guides addressing different aspects of AI, while others consolidated their information in a single guide. For this study, each library was treated as a unit of analysis to capture the overall message conveyed through its guides (details of the study samples are provided in Appendix 1).

The EDUCAUSE AI literacy framework consists of four levels, each linked to specific core competencies, key concepts, and reflection questions. However, the framework does not explain how these elements were developed. To address this limitation, the researchers reorganized the framework, elevating some key concepts and reflection questions to core competencies. Similar or overlapping concepts were consolidated, and multidimensional core competencies were split into sub-codes. Those not elevated or consolidated were categorized as sub-codes under their respective core competencies. Additionally, the researchers incorporated concepts from the LibGuides that were missing or underrepresented in the original framework, reflecting the LibGuides’ unique perspective, and enriching the coding schema for a more comprehensive analysis. Tables 2, 3, 4, and 5 show the core competency, sub-code, description, and source for Level 1, Level 2, Level 3, and Level 4, respectively (examples of coding details are provided in Appendix 2).

To validate the coding scheme, a pre-test was conducted on a sample of 10% of the guides (seven guides) to ensure intercoder reliability, yielding an agreement ratio at 83.97%. Throughout this process, MAXQDA and Excel were used as coding tools, and consensus on newly emerging codes was achieved through discussion.

In this section, we present the findings of our study based on the research questions outlined earlier.

This study examines 70 newly developed generative AI LibGuides from libraries affiliated with ARL and Oberlin group, analysed through the lens of the EDUCAUSE AI literacy framework. Our findings align with Talagala’s (2021) assertion that AI literacy must adapt to individuals’ roles and task contexts. Similarly, those responsible for providing AI literacy adapt their focus based on their training, background, and tasks. LibGuides include content absent from the EDUCAUSE framework, such as proper disclosure of AI usage, highlighting academic integrity concerns. Furthermore, LibGuides emphasize Level 1 (understand AI) and Level 3 (analyse and evaluate AI), focusing on helping users critically evaluate AI tools’ strengths and limitations.

However, many technical concepts featured in the EDUCAUSE framework, such as ‘diffusion models’ and ‘few-shot prompting’, were underrepresented in the LibGuides. Furthermore, prompt techniques and operational instructions for AI tools at Level 2 (use and apply AI) were less frequently addressed. The near absence of Level 4 (create AI) was particularly striking, suggesting that libraries focus more on resource discovery and ethical AI use, rather than advanced AI creation and programming.

These findings underscore the need for collaboration across higher education to foster comprehensive AI literacy, with various roles contributing their unique expertise. Hervieux and Wheatley’s (2024) AI literacy framework, grounded in Bloom’s taxonomy and shaped by a library science perspective, provides a concise alternative to the EDUCAUSE framework. By integrating their approach with the detailed competencies identified in this study, institutions could develop a more actionable and inclusive AI literacy framework that better addresses diverse user needs. Moreover, the detailed coding scheme from this study offers libraries a valuable tool for identifying underrepresented areas and fostering partnerships with other campus units to bridge these gaps effectively.

Our analysis also revealed that ARL libraries generally cover more components of the AI literacy framework than Oberlin group libraries, which may reflect differences in resources, institutional priorities, and librarian training. Addressing these disparities through more consistent training and knowledge-sharing initiatives could ensure a common standard of AI literacy support across academic libraries.

Lastly, this study excludes discipline-specific generative AI guides, which could offer valuable insights into discipline-based AI literacy needs. Friesen et al. (2023) examined publications by librarians across various disciplines and found that legal and health sciences librarians had the highest number of AI-related articles, with different disciplines focusing on distinct AI-related topics. Future research should explore whether different disciplines prioritize unique AI literacy competencies and how librarians in specific fields can better support those needs.