AI in Public Libraries: Field Guide

Use this guide yourself. Read it, work through the decision tree, talk it over with your board. You don't need a consultant to figure out if AI is right for your library. You know your constraints and your community better than anyone. Try the framework first. If you get stuck on a specific vendor decision or need help negotiating terms, call us then.

67% of libraries are exploring AI. 7% have actually implemented it. 93% are still trying to figure out if it's worth the risk.

This guide exists because the discourse gives you two options: engage or refuse. Neither one helps you Monday morning. What you actually need is a decision framework based on your constraints (budget, staff, capacity), clarity on what works and what doesn\'t, and honest language about what\'s actually at stake when you deploy AI in front of patrons.

It\'s not theoretical. It\'s grounded in what\'s happening right now (January 2026): OverDrive\'s monopoly, Boundless shutting down, six states passing AI laws, documented liability precedents, and the patterns I watched play out across four industries and eighteen years.

What's in the guide

Decision tree

Seven questions. If you can\'t answer yes to all of them, you're not ready. No ambiguity.

Your situation

Small library vs. medium vs. large. Different constraints, different moves. Pick yours.

What's actually happening

OverDrive owns 90% of the market. Boundless shut down. There are almost no real competitors. This matters.

Board conversations that work

Your board WILL ask "why aren\'t we doing AI?" Here\'s the actual language that works. Not the polite version.

What librarians get wrong

Six mistakes I watch libraries make. How to avoid them.

Equity: Who gets hurt?

When AI fails, it doesn\'t fail equally. Who\'s most vulnerable and what you actually need to test before deployment.

Contract red flags

Specific language to watch for. These aren't theoretical. I watched these happen.

What works vs. what doesn't

Patterns that actually succeed. Patterns that fail. Which one you're doing.

The core problem

The AI discourse gives you two options: engage or refuse. Neither one helps you Monday morning.

The "embrace AI" crowd will never have to answer to a patron harmed by hallucinated medical advice. The "refuse AI" crowd never had to explain to their board why they still don\'t have solutions. Both positions assume clear thinking is possible. Eighteen years watching this pattern tells me it\'s not.

What you need: A decision tree. Clarity on what actually works. Honest language about stakes. Scripts that match your voice, not the polite version. Contract red flags you can actually use. Equity frameworks that name who gets hurt. Models that work.

This guide has all of that.

Who this is for

Directors who need to decide this quarter whether to renew, adopt, or refuse an AI vendor contract
Librarians who see problems but don't have decision authority and need language that actually works to raise concerns
Small library directors being pressured by their board and needing governance scripts that match their actual voice
Technology staff who need legal language that protects the library, not the vendor
Consortium leadership who want to negotiate on behalf of members instead of leaving each library trapped alone
Anyone who watched OverDrive go from helpful to extractive and sees the same pattern starting with AI vendors

What makes this different

It\'s not theoretical. Every framework is grounded in what\'s actually happening now (January 2026): Baker & Taylor\'s collapse, Connecticut\'s ebook law, six states enacting AI chatbot laws, documented liability precedents, real vendor practices.

It\'s not academic. No jargon. No citation-heavy paragraphs. No "some argue that." Just: here\'s the problem, here\'s what others did about it, here\'s what you should decide.

It respects your time. You can read the one-page summary and get to decision. You can read the full guidance if you need depth. Either way, you're not wasting time on philosophy when you need answers.

It names the stakes. Not just technology risk. Equity risk. Privacy risk. Liability risk. Vendor lock-in risk. Budget risk. The sections don\'t pretend these are equal; they\'re weighted by actual impact.

The 15-minute version

67% of libraries are exploring AI; only 7% have actually implemented it. You're not behind. This is typical.

What works: AI augmenting human judgment ("copilot"). After-hours chatbots with clear limits and human backup. Staff productivity tools with human review. Starting with specific problems, not "we should have AI."

What doesn't: AI replacing human judgment ("autopilot"). Deploying without testing. Trusting vendor accuracy claims. Multi-year contracts without exit clauses. Ignoring where patron data goes.

Before you deploy anything: Name the specific problem. Calculate total cost. Verify insurance coverage. Consult legal counsel. Review vendor contracts. Define success metrics.

Smart moves: Start with internal staff tools before patron-facing. Plan for 2-3 year cycles, not 5-10 year horizons. Avoid multi-year vendor commitments. Preserve entry-level positions. Document and share what you learn.

Download the guide

The complete guide is 35+ pages of decision trees, case studies, policies, legal analysis, and governance scripts.

Download TXT Download DOCX

Updated January 2026. Freely distributable. No login required. Share with your team, paste into your wiki, adapt for your library.

What to do after you read it

Step 1: Read the one-page decision tree. Answer the questions. This takes 15 minutes and tells you whether you're ready to adopt AI or whether other groundwork should come first.

Step 2: If you're moving forward, read the case studies relevant to your library size and situation. See what worked and what didn\'t.

Step 3: Read the contract red flags section. Run your current vendor agreements through that checklist. The results are often eye-opening.

Step 4: If you need board approval, use the governance scripts. They're written to work with actual boards asking actual questions.

Step 5: If you're deploying something, use the model policies and adapt them to your library\'s context. Don\'t write from scratch. The policies exist because they\'ve worked for other libraries in your situation.

Version history

January 2026: Initial release. Based on Clarivate Pulse 2025 survey, OCLC research, ACRL competencies, Connecticut legislation, Baker & Taylor collapse documentation, and documented legal precedents through December 2025.