AI as a guest · Library managers

Choose an AI tool for a specific library task

Compare tools against the same task, data boundaries, accessibility needs and total cost before making a purchasing decision.

Original Host a Culture editorial guide · Version 1 · 2026-09-30 · Not yet trialled in a library

Before you start

Suggested review: 60 minutes, plus any procurement checks

A task owner, a small fictional test set and the supplier’s current product documentation. Include IT or privacy colleagues where the proposed use requires them.

Intended outcome: A dated decision to test, revise or reject a particular use case, with reasons and unresolved questions.

Run the activity

Define the task

Name one problem, the intended users and a measurable acceptance criterion. Example: prepare an event draft that preserves all source facts. Identify a non-AI alternative and whether a tool is needed at all.

Map information and infrastructure

List what would leave the library, who receives it, account requirements, retention controls, deletion options and network/device needs. Do not assume a consumer subscription has the same terms as an institutional contract. Keep actual sensitive records out of the trial.

Run comparable tests

Use the same fictional tasks in each candidate. Include local-language text, a missing fact, a request to invent a source, keyboard navigation and an export exercise. Record observed behaviour separately from supplier promises.

Calculate the whole cost

Record currency, tax assumptions, price-check date, seats, usage limits, support, staff review time, training and exit costs. Do not treat a free trial as a permanent zero-cost service.

Make a bounded decision

Write down the accepted use, prohibited inputs, responsible reviewer, spending limit and review date. If essential documentation or deletion/export controls cannot be verified, keep the decision unresolved. A tool assessment does not grant permission to upload every library record.

Working template

Comparison record: task; users; tool/version; evidence date; supplier claim; observed result; supported languages; accessibility; data recipients and retention; cost assumptions; unresolved questions; decision owner; next review date.

Check the result

All candidates face the same tasks and criteria.
Supplier claims and our measurements are visibly separated.
Paid placements and our own products are disclosed.
A procurement or legal decision is not replaced by an AI-generated score.

Limits of this guide

This is a practical evaluation template, not legal advice or certification. No vendor is endorsed or ranked. Check current supplier terms and applicable institutional requirements before deployment.