AIEd Knowledge BaseAIEd KB
On this page

Concept

Libraries

created 2026-11-05 · updated 2026-11-05

Libraries — the institutions that provide access to the scholarly and public record, and the operational systems behind that access: catalogues and discovery layers, reference and research services, collections, and the instruction programs that sit on top. In AI in education they matter twice over: as a setting where generative AI is being deployed in live services, and as the institution that carries information-literacy instruction while its own search layer is rebuilt underneath it.

Questions to Consider

  • A reference desk once answered the factual question. When a chatbot answers it in seconds, what is the service for — and what should the staff who ran it be doing instead?
  • The systematic reviews find AI adoption concentrated in reference and cataloguing while funding is the most-cited barrier. Is the binding constraint money, staff skill, or the absence of a decision about what the library is for?
  • Library systems see the whole population a course-level tool never does: borrowers, non-borrowers, community members. Does that make the library the right place to host public AI access, or the wrong place to concentrate surveillance?
  • AI literacy and information literacy are both taught by librarians and both lack a settled assessment. Should a library merge them into one program, or keep the frameworks distinct?

Introduction

The library is the part of the education system that has always been in the information-access business, which is why AI arrives there less as a novelty than as a change in the supply chain. Two literatures meet on this page. One studies libraries as adopters: what they are doing with AI, what stops them, and how the work is distributed across technical services, reference, collections, circulation, instruction and staff development. The other studies libraries as infrastructure: the place where students and the public meet search at all, and therefore where a shift from search to synthesis is first felt.

The page is deliberately distinct from Librarians, which covers the people and their professional role, and from Information and Media Literacy, which covers the construct they teach. This page is about the institution: its services, its systems, its staffing model and its public obligation.

How AI enters library operations

A systematic review of 29 studies published between 2020 and 2024 organizes AI adoption into six dimensions — technical services, reference and information services, administrative services and circulation, collection development, information literacy and user education, and professional development and collaboration — and finds the center of gravity in reference: AI for answering directional and ready-reference questions appears in 34.48% of the reviewed studies, with cataloguing and chat-transcribed reference next at 27.58% each (Ayinde, Ebiefung and Oladokun, 2026). The same review records what the adoption literature usually softens: funding is named as a critical barrier in 48.28% of studies, technical skill gaps and poor data quality in 58.62%, and fear of job loss in 13.79%.

The distribution matters because it describes a library automating the parts of its work that were already transactional — the directional question, the catalogue record, the chat transcript — while the judgment-heavy work remains human. That is consistent with the instructional evidence elsewhere in the knowledge base: an embedded librarian consulted when a recommender was least confident handled mostly source-evaluation questions, not lookup questions (Huang, 2026).

The search layer moves

The library's discovery system is no longer the only route to the record. In a survey of 236 US university students, search engines remained dominant for academic retrieval, but generative AI had become a significant source, with younger and international students more likely to use it and heavy users more satisfied with the answers (Lund et al., 2026). Behavioral data complicate the self-report: at one Chinese university, nine years of loan records covering 3.2 million transactions showed no level or trend change at the January 2023 generative-AI breakpoint, even though a parallel survey of 368 students found substitutive AI use weakly associated with reduced intention to keep using print (Huang et al., 2026). The honest reading is that displacement is real in stated preference and uneven in behavior, which is exactly the kind of gap a library is positioned to measure and a course is not.

Instruction, staffing and the research base

Libraries carry AI instruction before they have a curriculum for it. A systematic review of 32 studies found integration running through LibGuides, tutorials, workshops and revised modules, with 30 of the 32 studies published in 2024 or 2025 and 11 from the United States (Mirza and Arif, 2026). A scientometric map of 1,669 publications in the same period shows the field's center of gravity in a small set of library journals, with the Journal of Academic Librarianship both the most productive outlet and the central node of the journal co-citation network (Jabeen and Lux, 2026). Neither map shows practice changing; both show a literature growing faster than the evidence of effect.

Staff capacity is the recurring constraint in the survey work: 412 academic librarians across 38 Nigerian institutions were studied for their student-engagement strategies in information literacy and AI instruction, and staff development recurs across reviews as the intervention libraries ask for (Dada, 2026). The operational question is not whether to train staff but whether the training is tied to a service the library has decided to run.

Public access as an AI question

The library's oldest function — providing access that individuals cannot buy — has acquired a new object. The Public AI Libraries pilot places a shared AI workstation in public libraries, gives each patron a separate account, and charges participating libraries $1,000 per kiosk per year. Whether or not the model spreads, it names the equity argument precisely: in a digital divide that now runs through AI capability, the library is the institution that can supply shared, staffed access rather than a subscription per person.

What this means for practice

  • Decide which services the library is running before buying tools. The adoption literature is full of pilots without an owner; the reviews' most-cited barrier is funding, but the second is the absence of staff capacity matched to a service.
  • Keep a human escalation path in any AI service. The evidence supports AI for lookup and drafting, and human staff for judgment about sources; designs that erase the second to scale the first remove the part patrons cannot get elsewhere.
  • Instrument the search shift rather than surveying it. Loan and discovery logs already exist; a library that pairs them with instruction data will know more about AI's effect than a one-off student survey can show.
  • Do not bill AI literacy as information literacy. They share staff and frameworks but differ in what is assessed; conflating them produces a program that measures neither.

Connected Concepts

Connected Articles

Connected Resources

  • Public AI Libraries
    A 2025–2026 national pilot that puts a shared Public AI workstation in US public libraries — chat, image, music and video generation, a code IDE and an educational guide — with a private account for each patron.

Metadata

DisciplineLibrary And Information Science
LevelHigher Ed, Adult Learning
Intended audienceLibrarians, Administrators, Institutions
Technologies and techniquesGenerative AI, Large Language Models (LLMs), Recommender Systems and Learning Paths, Knowledge Graph
Institutions and policyAI Governance, Educational AI Policy
Equity, ethics, and responsible useDigital Divide, Equity
Foundations of AI in educationInformation and Media Literacy, AI Literacy
Page kindSynthesis
SourceHistoryEditChat

Embed this page

Copy the code below to embed this page in a learning management system or other website. The embedded version shows only the page content — it hides the site header, navigation, and footer.