Public Library AI Readiness Toolkit

Part 1: AI Vendor Questionnaire

Use this during sales calls or email the questions to vendors before you commit. You want written answers, not verbal assurances.

Vendor / Product name

Date of assessment

Evaluated by

1. Data privacy: Does this tool collect, store, or train on patron PII or search histories? Can you opt out of data sharing?

Yes No Unclear

Notes

2. Copyright & sources: Does the vendor disclose what the model was trained on? Does the tool respect copyright and intellectual property?

Yes No Unclear

Notes

3. Bias & transparency: Has the tool been audited for algorithmic bias? Does it clearly label AI-generated content so patrons aren't misled?

Yes No Unclear

Notes

4. Accessibility: Does the AI interface comply with WCAG accessibility standards so all patrons can use it?

Yes No Unclear

Notes

Part 2: AI Impact Assessment

Complete this before deploying any new AI tool to staff or the public. Document the risks now so you're not explaining them later.

Tool name

Intended use: What is this tool supposed to do?

Data used: What information does it access or require?

Potential risks: What could go wrong? (hallucinations, over-reliance, privacy leaks, etc.)

Mitigations: How will you keep it safe? (disclaimers, monthly staff testing, review process, etc.)

Benefits: Why are you using it? (accessibility, staff time savings, patron outcomes, etc.)

Approval decision

Approved for use Needs more review Rejected

Part 3: Staff AI Policy Agreement

Review these guidelines with staff before they use AI tools in their workflows. Check each box to acknowledge.

Staff acknowledgments

Protect patron privacy. I will not enter a patron's name, contact info, library record, or reference question into a public AI tool. Patron confidentiality applies to AI inputs.

Keep the human in the loop. AI is a tool, not a replacement for professional judgment. If I use AI to draft content, I will review and edit it for accuracy, tone, and local relevance before publishing.

Be transparent. If AI was substantially used to create a public-facing resource (image, summary, program description), I will note that AI was used.

Uphold equity. AI models can carry the biases of their training data. When using AI for collection development or programming ideas, I will actively review suggestions to ensure they don't exclude marginalized voices.

Employee name

Date