Librarians occupy a distinctive position in relation to AI research tools: the professional responsibility is to evaluate information quality and guide others in accessing reliable sources, while the daily work involves navigating an AI information environment that has non-trivial accuracy problems. The librarian who can articulate why AI-generated reference guidance is unreliable for specific question types — based on professional understanding of how information systems work — is better positioned to help patrons than the one who avoids AI entirely or accepts AI outputs uncritically.
The practical question for librarians using AI in their own professional work is the same question they teach patrons: what are the limits of this source, and what does it need to be verified against? The answer is different for different types of professional tasks.
Where AI Provides Genuine Leverage for Librarians
Research Guide and Pathfinder Drafting
AI is reliable for drafting the prose structure of research guides — the introductory sections, the organizational framework, the descriptive context around subject areas. Given a librarian's outline of the guide content, AI can draft readable, well-structured prose that the librarian then populates with verified source information.
High-reliability research guide uses:
- Drafting the introduction and subject context sections based on a professional description
- Generating the organizational framework for a guide topic the librarian is developing
- Improving the readability and clarity of guide prose
- Drafting the patron-facing instructions for using specific resources (after the librarian verifies the instructions against the actual interface)
The discipline: AI drafts the prose; the librarian provides and verifies all source information. A research guide populated with AI-generated database descriptions or AI-generated coverage claims is a guide that will mislead patrons. The source knowledge comes from professional expertise; AI contributes prose quality.
Instructional Content Development
AI can support the development of library instruction materials — lesson plans, activity designs, workshop outlines, and tutorial scripts.
High-reliability instructional uses:
- Drafting lesson plan structures based on learning objectives the librarian specifies
- Generating discussion questions for information literacy topics
- Drafting quiz questions for information literacy assessments (reviewed for accuracy)
- Scripting tutorial narration given a content outline
- Developing practice activity scenarios for database instruction
The discipline: AI-generated instructional content requires review for accuracy regarding specific library resources, current database interfaces, and institutional context. An AI that generates a database instruction tutorial may describe an interface that has changed or a feature that doesn't exist in the library's subscription tier.
Administrative and Professional Documentation
AI is reliable for administrative and professional documentation tasks where the content is provided by the librarian and AI contributes organization and prose quality.
High-reliability administrative uses:
- Drafting meeting agendas and minutes from notes
- Drafting policy documents from provided policy decisions and rationale
- Organizing and improving grant application prose given the program description
- Drafting collection development policy sections from provided priorities
- Summarizing committee reports or professional reading notes
For cataloging work, AI can suggest subject headings, classification numbers, and descriptive metadata as starting points — which the cataloger then verifies against authority files and applies professional judgment to.
The discipline: AI-generated metadata suggestions are not catalog records. Subject headings need to be verified against the Library of Congress Authority File; classification numbers need to be verified against the library's local classification practice; all metadata needs professional review. AI suggestions that are applied without verification introduce errors that degrade catalog quality.
Where AI Introduces Risk for Librarians
Database Coverage and Content Claims
AI will confidently describe specific databases — what they contain, their coverage dates, their content types — with claims that are unreliable for several reasons: database content changes, subscription tiers vary, coverage dates are frequently incorrect in AI training data, and specific databases that AI describes may have changed substantially or may not match the library's specific subscription.
The problem: a librarian who asks AI "what does ProQuest Historical Newspapers cover?" and uses the answer to guide a patron is providing guidance based on AI's training data about the database, which may be outdated, incomplete, or reflect a different subscription tier. The patron who follows that guidance and doesn't find what was promised loses trust in the library.
Rule: all database coverage information used in reference work or research guides comes from current, verified professional knowledge about the specific resources the library subscribes to. AI is not a reliable source for specific database coverage claims.
Catalog Record Accuracy
AI used to generate catalog record content — titles, dates, subject headings, notes — will produce plausible-looking but unreliable records. Cataloging requires primary source consultation (the item in hand), authority file verification, and professional judgment that AI cannot replicate.
The specific risk: an AI-generated MARC record might have the right format and plausible-looking content, but the bibliographic details may be wrong (incorrect publication dates, missing co-authors, incorrect subject headings that don't match authority files). A catalog populated with unverified AI-generated records degrades the catalog for all users.
Rule: catalog records are created through primary source consultation and authority file verification, not AI generation. AI can suggest starting points; catalogers verify everything against authoritative sources.
Patron Privacy and Reference Transactions
AI-assisted reference work requires careful attention to patron privacy. Using AI tools to assist with reference questions may involve sharing patron research interests or research questions with third-party systems.
The privacy concern: in jurisdictions with strong library confidentiality statutes, sharing patron research information with external AI systems may implicate patron privacy rights. The librarian who pastes a patron's detailed research question into an AI interface to get source suggestions may be sharing patron information with a commercial service in ways that conflict with library confidentiality obligations.
Rule: patron information should not be shared with external AI systems without understanding the institution's privacy obligations and the AI service's data handling practices. Research question topics can sometimes be generalized ("I'm helping someone research [topic]") without including identifying information about the patron.
Reference Answer Verification
AI answers to factual reference questions — "who wrote X?", "when did Y happen?", "what is the definition of Z?" — are plausible-sounding but unreliable. AI produces confident factual answers with a notable error rate, including for questions that seem straightforward.
The librarian's professional authority: the librarian who provides reference answers based on AI responses without verification is not practicing reference work — they're relaying AI output. The professional skill in reference work is source identification and verification, not just answer retrieval. AI-assisted reference that skips verification undermines the professional value of the reference service.
Rule: factual reference answers come from verified sources, not from AI. AI can suggest where to look; the answer comes from the authoritative source.
The Librarian's AI Evaluation Framework
The same source evaluation framework librarians teach to patrons applies to AI:
Authority: What does the AI system know about this topic, and how reliably? (AI's training data has an authority problem: it includes unreliable sources alongside reliable ones, with no clear provenance.)
Accuracy: How would I verify this claim? (For AI output, always know the verification source before relying on the claim.)
Currency: When does the AI's knowledge end? (AI training data cutoffs matter for database coverage, current resources, and evolving standards.)
Coverage: What types of information does AI handle reliably vs. not? (Prose generation and structure: high reliability. Specific factual claims and current database information: low reliability.)
Purpose: What is AI optimized to produce? (Plausible, well-formed text — not verified factual accuracy.)
This framework — which librarians teach daily — is also the framework for evaluating AI output in professional work.
A Worked Example: AI Used Well in Research Guide Development
Situation: a librarian is developing a new research guide on genealogical resources for the library's region.
Good use of AI:
Structural drafting: describe the research guide topic to AI and ask it to suggest an organizational structure — what sections a genealogical research guide typically includes. Use this as a starting framework, adapting it based on professional knowledge of what the library's specific collection and patron population need.
Prose drafting: for each section of the guide, provide AI with the actual content (the specific databases and resources, with their real coverage) and ask it to draft readable patron-facing prose. Verify the draft against the actual resource descriptions.
Instructional scenarios: ask AI to generate practice scenarios for a genealogical research instruction session — "give me five realistic genealogical research questions that a patron might bring to the library." Use these as raw materials for instruction activities, adapted for local relevance.
Where AI doesn't substitute for professional knowledge:
Database coverage: all coverage information in the guide comes from professional consultation of the actual resources — date ranges, geographic coverage, content types. Not from AI.
Source recommendations: the determination of which resources to include and how to prioritize them comes from the librarian's professional knowledge of the resources, not from AI suggestions about which databases exist.
Verification: every link in the guide is verified against the actual resource; every coverage claim is checked against the current database description.
Key Takeaways
-
AI drafts prose; librarians provide and verify source knowledge. Research guides and instructional materials benefit from AI's prose quality; all source information comes from professional expertise and is verified against current authoritative sources.
-
Database coverage claims from AI are unreliable. Database content changes, subscription tiers vary, and AI training data is not current. All coverage information used in reference work comes from professional knowledge of the library's specific subscriptions.
-
Patron privacy requires care in AI-assisted reference. Sharing patron research information with external AI systems may implicate library confidentiality obligations. Generalize patron questions when using AI assistance; never share identifying information.
-
Factual reference answers come from verified sources, not from AI. AI can suggest where to look; the answer comes from the authoritative source. Providing AI-generated factual answers without verification is not professional reference practice.
-
The source evaluation framework librarians teach applies to AI. Authority, accuracy, currency, coverage, purpose — the evaluation questions that apply to databases and web sources apply equally to AI output.
Conclusion
AI tools provide genuine leverage for librarians in prose-heavy professional tasks: research guide drafting (structure and prose, not source information), instructional content development (lesson plans, activities, tutorial scripts), administrative documentation, and metadata suggestions (as starting points for cataloger verification). They introduce professional risk for librarians in database coverage claims (training data is unreliable for current specific subscription coverage), catalog record accuracy (requires primary source consultation and authority file verification, not AI generation), patron privacy (sharing research questions with external systems may implicate confidentiality), and factual reference answers (AI generates plausible answers with a significant error rate). The evaluation framework — authority, accuracy, currency, coverage, purpose — is the framework librarians teach patrons, and it's the same framework that guides appropriate AI use in professional library work.
Related reading: Best Web Clipper Extensions.