will AI replace librarians?

Twenty-five of the 30 tasks analysed show zero AI penetration. That's not a rounding error. It means the majority of what you do every day, from cataloging and classifying materials to training junior staff, directing programs, and evaluating the collection for weeding, is work that hasn't been touched by automation in any meaningful way.

You're the one who plans storytime, runs outreach for special populations, builds the newsletter, and writes the library's policies. Those aren't just tasks. They require knowing your community, your patrons, your institution's history, and your budget constraints. A chatbot doesn't know that the Tuesday afternoon crowd is mostly seniors who prefer large-print materials, or that your branch's wi-fi policy was changed after a specific incident two years ago.

And there's a physical dimension here that people forget. You troubleshoot the projector when it dies ten minutes before a program. You check materials in and out. You walk someone to the shelf. The Anthropic Economic Index places librarians in a relatively low-exposure category precisely because so much of the role is grounded in physical space, institutional knowledge, and real relationships with real people. That's not going away.

AI is genuinely bad at reference work when stakes are high. Tools like ChatGPT and Perplexity will confidently fabricate journal citations, invent book editions that don't exist, and misattribute quotes. For a patron doing casual research, that might not matter much. For a researcher, a student, or a legal professional using your library's reference services, a hallucinated source is a real problem, and the liability for that error sits with the institution, not the AI.

Cataloging is another area where AI still falls short. Automated metadata tools can pull basic MARC records, but they struggle with unusual formats, local collections, archival materials, and anything that requires contextual judgment about subject headings. The Library of Congress has been testing AI-assisted cataloging for years and still relies on trained catalogers to catch errors and handle edge cases.

Privacy is the third major gap. Libraries have a legally protected relationship with patrons' reading records. Routing reference queries through a commercial AI tool means sending patron data to a third-party server, which can conflict with state library confidentiality laws and ALA privacy guidelines. That's not a theoretical concern. Several library systems have pulled back on AI deployments specifically because legal counsel flagged the data-handling risk.

The four tasks where AI has genuinely high penetration are all in the reference and search space. When a patron walks up and asks a factual question, tools like the AI-assisted search built into ProQuest or EBSCO can surface relevant results faster than a manual database search. ChatGPT and similar tools can help with locating obscure information quickly, particularly for general-interest queries that don't require verified sourcing.

For developing and maintaining information access aids, tools like LibGuides with AI-assisted drafting features can speed up the creation of annotated bibliographies and subject guides. Some systems are also using AI to handle initial complaint routing, flagging patron feedback and sorting it before a staff member reviews it. That's a narrow use case, but it's real.

On the cataloging-adjacent side, tools like OCLC's automated metadata services can generate draft MARC records from ISBN data, saving time on straightforward new acquisitions. BiblioCommons has built AI-driven recommendation layers into discovery systems, so the catalog itself surfaces related titles without librarian input. These tools are doing real work. But they're handling the routine layer of these tasks, and they're working best when a trained librarian is checking the output before it goes live.

The biggest shift isn't what you're doing, it's what you're doing less of. Initial reference queries, especially the simple ones like 'do you have a book about X' or 'what are your hours', are increasingly handled by chat widgets on library websites before a patron even walks through the door. That frees you for the longer, more complex reference consultations that actually need a human.

You're also spending less time manually constructing basic database search strings. The search interfaces have gotten smarter. But you're spending more time verifying what those searches return, because patrons are also arriving with AI-generated bibliographies that need checking. The quality-control work has gone up, not down.

What hasn't changed at all: programming, community outreach, collection development decisions, staff supervision, policy work, and the physical operations of the library. Those parts of the job look almost identical to how they looked five years ago. The admin around reference has shifted. The core of the job hasn't.

The BLS projects 1.7% growth for librarians through 2034. That's slow, but it's growth, not decline. With 142,100 people currently employed and 13,500 annual openings, the job market is stable rather than shrinking. Most of those openings come from retirements and departures, not net new positions.

The AI exposure score here is 27%, which is low by the standards of knowledge-work roles. For comparison, the Anthropic Economic Index places occupations with heavy information-processing tasks significantly higher. Librarians land in the 'amplified' quadrant, meaning AI is a productivity tool here, not a replacement driver. The role has enough physical, relational, and institutional work to keep demand for humans steady.

What's worth watching is the public library funding environment. Budget pressures have nothing to do with AI and everything to do with local government finances. Some municipalities have explored replacing librarians with automated kiosks, but those experiments have mostly reversed after community pushback. Academic and special libraries are seeing more AI integration, but headcount has held roughly steady because the research support work has expanded as the clerical layer has shrunk.

The 27% exposure score is unlikely to rise dramatically in the next five years. The tasks that remain at zero penetration, community programming, staff training, collection development, policy-making, physical operations, aren't tasks that are waiting for a slightly better language model. They require presence, authority, and institutional accountability that AI can't hold.

The scenario where this changes significantly looks like a 10-plus year horizon and requires breakthroughs in robotics (for physical operations), trusted AI sourcing (for reference work that currently can't be verified), and some legal resolution of library privacy obligations. None of those are imminent. The more likely five-year picture is a role where AI handles a larger share of basic search and initial patron routing, and librarians spend a higher percentage of their time on programming, community partnerships, and complex research support.

Double down on the 25 tasks that AI hasn't touched. Collection development, community programming, and staff training are the core of this role's value, and they're where you should be building depth. If you're early in your career, get involved in program design and community outreach now, not just reference desk coverage. Those are the skills that will matter most.

Cataloging expertise is also underrated as a career investment. As libraries adopt more automated metadata tools, someone has to check the output, handle the exceptions, and manage the local collections that don't fit standard workflows. That person needs real cataloging knowledge, not just familiarity with OCLC. The Library of Congress's ongoing work on linked data and the BIBFRAME standard is the direction the field is heading, and getting comfortable with it now puts you ahead.

On the AI side, learn enough to be the person who evaluates which tools are appropriate for your library's legal and privacy environment. That's a real skill gap in most institutions right now. Librarians who can assess a vendor's data-handling practices against their state's library confidentiality law are genuinely useful to their administration. The American Library Association's Office for Intellectual Freedom has guidance on this, and engaging with it puts you in a position of expertise rather than reaction.

Finally, build your programming portfolio. Special services, outreach to underserved groups, and partnerships with schools, social services, and local organisations are where public libraries are justifying their budgets. If your name is attached to programs that people show up for, your position is harder to cut.