Industry Playbooks
AI knowledge work for librarians is transforming reference research, catalog metadata enhancement, subject guide development, and information literacy instruction — while raising professional questions about source verification, hallucination risk, and the expert judgment that remains distinctly human in library services.
Librarians have always been the people tasked with evaluating information quality — teaching patrons to assess sources, curating collections of authoritative materials, and providing guidance on how to find and evaluate credible information. They are the professionals most equipped to understand what AI systems do well, where they hallucinate, and why source verification matters.
Now those same professional tools are directly relevant to their own work. AI knowledge work for librarians is genuinely changing how librarians do research assistance, develop instructional content, enhance catalog metadata, and build subject guides. And librarians — who teach information literacy precisely because information quality is not guaranteed — are uniquely positioned to use AI with appropriate critical evaluation.
What AI does well:
Practical application: A business librarian gets a complex reference query about maritime salvage law in the 1970s — outside her core expertise. She asks Claude: "What are the primary sources for US maritime salvage law from the 1960s-1970s, and what databases or archives would be appropriate for researching this topic?"
AI identifies: Admiralty and Maritime Law Institute publications, WESTLAW maritime law databases, the Federal Maritime Commission historical records, and suggests checking the National Archives for relevant regulatory proceedings. The librarian uses this as a starting point, verifies the specific source suggestions against her professional knowledge, and uses several of the resources the AI identified as appropriate.
Critical caveat: AI reference research suggestions may hallucinate specific database names, coverage dates, or resource titles. "The Maritime Law Annual, published by the Federal Maritime Institute since 1965" may sound authoritative and be entirely invented. Verify specific resource names and coverage claims before recommending them to patrons.
What AI does well:
Practical application: A science librarian is developing a new subject guide on synthetic biology — a rapidly evolving area she's developing expertise in. She asks Claude: "Suggest a structure and resource categories for a subject guide on synthetic biology for graduate-level researchers. What database types, journal categories, preprint servers, regulatory sources, and grey literature sources should be included?"
AI returns a useful structure: databases (PubMed, Web of Science, CAS SciFinder), preprint servers (bioRxiv), regulatory sources (NIH, FDA for related regulations, European Medicines Agency), grey literature (NIH Biosafety Committee resources, iGEM community resources). The librarian uses this as a starting framework, adds the specific databases available at her institution, and verifies each resource is current and accessible.
Critical caveat: AI-generated subject guide content may include databases the library doesn't subscribe to, databases that don't exist, or resources that have changed significantly. Every AI-suggested resource needs verification before inclusion in a subject guide that patrons will use.
What AI does well:
Practical applications:
Important limits: AI-generated cataloging is a starting point, not a finished product. Professional catalogers apply controlled vocabulary (LCSH, MARC standards) with precision and institutional consistency that AI does not automatically provide. AI suggestions should always go through cataloger review before being added to the catalog.
What AI does well:
Practical application: A librarian is developing a new instruction module on evaluating AI-generated information — a topic that didn't exist in the curriculum 3 years ago. She asks Claude: "Suggest 3-5 active learning activities appropriate for a 50-minute instruction session on evaluating AI-generated information for undergraduate students. Include learning outcomes for each activity."
AI returns 5 activity ideas with learning outcomes. The librarian selects 2 that fit her audience and time constraints, revises them based on her knowledge of the specific course context, and has a draft session plan in 30 minutes rather than 90.
| Use Case | Tool | Notes |
|---|---|---|
| Reference research starting points | Claude | Verify all specific resource suggestions |
| Subject guide development | Claude + LibGuides | AI for structure and draft text; librarian for verification |
| Catalog metadata | Claude + cataloging workflow | As starting points for cataloger review only |
| Instructional design | Claude + Canvas / institution LMS | Draft activities and outcomes; librarian revision required |
| Current resource monitoring | WebSnips | Dated clips of current resource pages and professional literature |
WebSnips for AI-assisted librarian work: AI suggestions for library resources need current verification. A subject guide built from AI suggestions requires verifying that each suggested database is currently subscribed, covers the claimed content, and has the interface described. WebSnips clips of database product pages, coverage statements, and subject guides from peer institutions — with dates — provide the current verification material. For professional development on AI in libraries, WebSnips clips of professional association statements (ALA, ACRL), vendor AI announcements, and peer institution AI policy pages with dates build a current, dated professional intelligence file.
A social sciences librarian, Elena Rodriguez, integrates AI into her work at an urban university library:
Reference encounter — complex social policy query:
A doctoral student needs comparative policy research on public housing demolition and displacement across US cities from 1995-2015.
Elena asks Claude: "What databases and primary source collections would be appropriate for comparative policy research on US public housing demolition from 1995-2015? What grey literature sources would be most relevant?"
Claude suggests: JSTOR, Sociological Abstracts, Policy File (ProQuest), HUD Annual Reports, National Housing Law Project publications, Urban Institute reports, and local housing authority annual reports.
Elena verifies: her library has access to JSTOR and Sociological Abstracts; Policy File is included in the ProQuest subscription. HUD reports are freely available at hud.gov. National Housing Law Project is a real organization with freely available publications. Urban Institute has relevant reports that are freely available.
She supplements from her own knowledge: adds PolicyMap for geographic data, notes that CityLab/Bloomberg CityLab archives have relevant journalism, suggests the Displacement Research and Action Network.
Final reference response is more complete and faster than a purely manual approach — AI provided the initial structure; librarian expertise and verification provided accuracy and completeness.
Instruction development — SIFT framework session:
Elena is updating a one-shot information evaluation session to include explicit instruction on evaluating AI-generated information. She asks Claude: "I'm developing a 15-minute segment on evaluating AI-generated information to add to an existing SIFT method instruction session. Suggest 2 activities appropriate for an undergraduate audience. The existing session already covers Stop, Investigate the Source, Find Better Coverage, and Trace Claims."
AI suggests:
Elena likes activity 2 — it directly demonstrates hallucination in a hands-on way. She adapts it: uses a Claude-generated paragraph on a topic relevant to the class's research area, identifies specific hallucinated citations by testing them herself first, and builds the activity around the actual academic databases the students use for the class.
The AI suggestion gave her a direction she might not have arrived at as quickly; the librarian expertise made the activity accurate and appropriate.
The most valuable reference service is the conversation — clarifying what the patron actually needs (which is often different from what they asked for), calibrating the complexity of the response to the patron's knowledge level, and knowing when to say "this is more complex than a quick database search; you need a research consultation."
AI can suggest databases. It cannot read the patron's email and infer that what they're describing as a "literature review" is actually a systematic review that requires methodological guidance.
Teaching information literacy is not just showing students where to search. It's teaching the critical thinking habits — slowing down before sharing, investigating sources, tracing claims to their origin — that produce lifelong information evaluation skills. AI can help draft instruction materials. It cannot teach the habit of mind.
Collection development reflects institutional values, community needs, and professional ethics about representation and inclusion. These decisions involve balancing factors — historical underrepresentation in the collection, patron demand, intellectual freedom commitments, budget constraints — that require human professional judgment about institutional context and values.
Patron data and AI tools: Do not upload patron query information or patron records to AI tools. Patron confidentiality protections apply regardless of the tool used to process that information.
AI-generated catalog records: AI-generated cataloging metadata should go through professional cataloger review before entering the catalog. Incorrect subject headings, inaccurate descriptive information, or metadata that doesn't conform to institutional standards can harm discoverability and reflect poorly on the library's professional standards.
Institutional AI policies: Many libraries and universities are developing formal policies on AI use in library services — particularly around AI-generated content on library websites, AI in cataloging workflows, and AI tools used with patron data. Know and follow your institution's applicable policies.
Attribution of AI assistance: If AI-generated content is published on library websites or instructional materials without human review and editing, that should be disclosed. Content that has been reviewed, verified, and substantially edited by a librarian may not require AI attribution, but institutional policy varies.
Mistake 1: Using AI-suggested resources without verification. "Claude said [Database X] covers this topic from 1980" — verify before recommending to a patron. AI hallucination risk is highest for specific factual claims about database coverage dates, specific resource titles, and specific organizational details.
Mistake 2: AI catalog metadata without cataloger review. AI-generated subject headings and descriptions need professional cataloger review for controlled vocabulary compliance, institutional consistency, and descriptive accuracy. AI speeds cataloging; it doesn't replace cataloging judgment.
Mistake 3: AI instruction content without review for accuracy. An instruction module drafted by AI may include inaccurate information about specific databases, incorrect search syntax, or outdated information about library resources. The instruction context means errors reach multiple students. Every AI-drafted instructional content must be reviewed for accuracy before use.
Mistake 4: Patron data in AI tools. Any patron information — query content, patron name, patron contact information — should never be submitted to third-party AI tools. Library confidentiality obligations apply to AI-mediated processes the same as any other.
AI knowledge work for librarians offers genuine acceleration in the tasks that take time but reward expertise — reference research starting points, subject guide drafts, catalog metadata, instructional activity ideas. The librarians who benefit most are those who use AI for the initial scaffolding and then apply professional judgment to verify, refine, and make the AI-generated work accurate and patron-appropriate. The professional skills that librarians have always offered — source evaluation, subject expertise, patron understanding, ethical judgment — don't diminish because AI can generate a database list. They become more important, because someone needs to decide whether the database list is accurate.
More WebSnips articles that pair well with this topic.
Knowledge management for librarians is the practice of organizing professional knowledge — subject expertise, collection intelligence, patron research patterns, reference resources, and instructional materials — in systems that let library professionals deliver better research support and build on accumulated expertise.
Research workflows for librarians are the structured processes for reference research, collection assessment, subject guide development, information literacy curriculum design, and professional development — enabling library professionals to deliver consistent, high-quality research support.
A note-taking system for librarians must capture reference encounter insights, resource evaluations, instructional observations, subject expertise, and professional development learning — building the documented record that makes library professional knowledge transferable, current, and cumulative.