1 of 14

AI for the Koha Side

AI isn’t only for our patrons – librarians can reap the benefits, too!

Learn how to integrate AI into your workflow and make AI work for you.

Practical AI for Everyday Library Work

2 of 14

AI Basics – ground rules

Working with AI responsibly: it’s auto-complete, not a mind. It can gather information and do some fact-checking (not always reliable). It can design; it can’t create.

Rule #1: PRIVACY

Never put patron PII (Personally Identifiable Information), patron circulation history, or financial spreadsheets into free public AI tools. Ask about an enterprise or locally-hosted account – data stays inside your organization.

Rule #2: REVIEW

Trust, but verify. Watch for hallucinations.

Run the CRAAP test: Currency, Relevance, Authority, Accuracy, Purpose. Start with low-stakes tasks – see if it actually saves time, then scale up your work.

A tale of cheese…

Think of AI as a passionate, occasionally unreliable digital intern: fast, enthusiastic, and sometimes confidently wrong. You are the one who can and should decide what is in the end result.

Let’s go over some basics of what AI is and is not. What is can and cannot do.

(read slide)

Does everyone remember Clippy? It would recognize the document you were building and give guidelines/structure? He didn’t write the content but did provide a guide.

Think of AI as Clippy on steroids, or Clippy as the father of AI – including the very excited and happy personality.

So let’s cover some ground rules of using AI:

Rule# 1 – Privacy!

Rule #2 – Review!

CRAAP test

My brother’s story: Cheese for a church young men camping trip for 1 week. 225 men and teenage boys. My brother Evan was on the food committee: plan, purchase supplies, prep/cook, serve, cleanup. Used ChatGPT to estimate how much to purchase. Spaghetti: noodles, sauce, cheese. Noodles and sauce estimates were great – only a small amount left over. Cheese – not so much: 3 cases from Costco = 36 canisters of parmesan cheese. Same with Sour Cream, but caught at time of purchase, and only had 2-3 tubs left over.

3 of 14

Where Do I Start?

Public/Free Tools

Fast, free, great for brainstorming & drafting
Wide general knowledge across many domains (i.e. the collective internet brain)
Inputs may be used to train future models

Enterprise / Local Accounts

Request an enterprise-tier account for library use – costs money
Keeps data inside your organization
The right choice for anything patron- or finance-related

Free vs paid versions
Collaboration
# of tokens given per session
File type uploads

Most common AI tools available:

ChatGPT (OpenAI) | Claude (Anthropic) | Gemini (Google) | Perplxity.ai

Variables to consider:

Web browsing capabilities
Reasoning models
Deep research options
Privacy and data handling

4 of 14

Treat it like a reference interview: tell AI who you are and how you want to be helped.

Context

The situation or background

Role

Who AI should act as

Objective

What AI should do

Process

How it should think or work

Grounding

The data or evidence to use

Format

The shape of the output

Writing a Good Prompt

“We are a public library with 13 branches and a bookmobile. We want to send Notices using the Koha ILS version 25.11.06 to borrowers with overdue books via email. Please create HTML for a Notice design that is friendly, concise, includes our library logo…”

Include an uploaded file of your library’s logo.
Answer any further questions it asks about what you want in the Notices
Tell AI how to ask you questions: multiple choice, open ended, etc.

REMEMBER: AI IS AN ASSISTANT – NOT A REPLACEMENT. Trust, but Verify.

https://koha-community.org/manual/latest/en/html/index.html

https://koha-community.org/documentation/

5 of 14

AI for Cataloging

Problem: As librarians and data specialists, WE often provide the data that AI uses to give other people answers!! We can see if others have provided this information before us (WorldCat, z39.50, etc.), but often we are the ones doing the work, book by book.

SO HOW CAN AI HELP ME CATALOG?

Okay - now that we’ve gone through the basics, the rest of the slides will be a lot of ideas and information about using AI in your everyday work. Remember that you’ll have access to the recording and the slides, so you can go back and review - this is just to get the ideas juices flowing. Feel free to take any of these ideas and run with them, and definitely work with your colleagues to come up with ideas of your own for your library and the work you do.

Biggest misconception (from our mgmt. at least) is that we can use AI to catalog – just show the book to the computer and it’ll catalog it for us! Doesn’t work that way. (Read slide)

So how CAN AI help?

Summaries; talking points and design; subject headings; classifications; translations and transliterations

DeepL used for translating (high volume and structural fluency) – Latin-based, Japanese, Russian, Cyrillic, Bulgarian, Greek (web, app, mobile app, browser ext, upload

Google translate/gemini does better for Latin based languages but not other alphabets.

Non-Latin: Papago

6 of 14

Importing - If you want to stick to putting in your own data, but still want a faster cataloging option, you can create and import batches of records.

AI for Quicker Work

Circulation and other statistics – make them fun and presentable to your stakeholders. Ask for help explaining them in an impactful way.

Use those same practices when writing Policy and Grant Templates

Cleanup - Export files into .mrc or other files.

Identify incomplete or inconsistent Marc records
Suggest authority file matches for name/subject headings
Find duplicate records to merge
List of messy/inconsistent author or subject authorities

Find the problems and clean them up. Even ask how to clean them!

Cataloging FOR you??

Several companies are now working on creating AI programs that will catalog FOR us. Ex-Libris, MarcAI, and OCLC are some of them.

There are companies working to create AI programs that CAN create Marc records from scratch for us.

Marcai.cloud - have not tried but it accepts photos or PDFs from the book, does a web search with ISBN/title/author, or uses text-provided metadata to create the record.

If you aren’t comfortable using MarcEdit to create importable Marc records using an Excel spreadsheet, you can use AI. Define each column, put the information for each book in a single row, then ask Claude to create a .mrc importable record. You can do this for bib records or item records to import.

I also tried ChatGPT – it did not handle this task well. Try different options, use what works best.

Ask AI to help you build a presentation to your stakeholders. Feed it your stats, ask for cool graphics, ask for help with using the right vocabulary to appeal to specific audiences. Make you sound awesome!

And finally, clean up your records! (Read slide)

Possible example to discuss: Asked about changes we were thinking of making (mirror 830$a/$v into 490$a/$v automatically). Claude created the jQuery, then pointed out that if we did that, we would need to be sure only one instance of the 830 field was present in the record. Created a report for me to check it right then. Step by step in thoughts, questions, testing, code, etc. A whole conversation about what would happen if we used Marc in that specific way.

7 of 14

AI for Reports

Create SQL reports using plain language:

Use AI as a teacher for learning SQL. Have it walk you through each part of the report it creates explaining schema, data tables, etc.

You can also start building reports on your own and having AI check them for you and explain what may need to be fixed.

Fix or amend reports already created:

If you are in charge of creating reports let AI be your helper!

I had never worked with SQL before, and it became a major part of my job about 2 years ago.

I spent a lot of hours reviewing reports that already existed, glanced through some online tutorials, and even bought the SQL for Dummies book. I reviewed the data tables and how they connected.

I did learn a lot, and I’m glad to have that foundation of linking data tables and SQL formatting, but I have learned a lot more using AI as it creates and amends the reports I need.

I personally use Claude when creating reports – I just like it. I always tell it which program and version I am using for the initial question, then just write what I need.

It often tells me about each part of the report, why it formatted it that way, what tables it linked, etc.

If it doesn’t tell you and you want to know more about the report - ASK.

Sometimes we have a report but it’s missing some information, or I’m not sure it’ll work. So I paste what I have into Claude and ask for help.

8 of 14

AI for Internal Design

AI can be your very excited helper, and now your friendly IT expert! Let it mentor you in your quest for independence.

Autofill a Marc field or subfield with a given value when I click in that field

Show the Biblionumber in the Orders table.

Remember that AI can find all the documentation it needs for Koha online. (Aspen too!)

Use AI to help you design and customize your sites using jQuery, CSS, and more.

Along with reports, AI can be your general IT helper - just remember that it will be a very EXCITED helper, so trust, but verify.

AI is a great teacher for a learning new programs, data systems, and other tools. It will walk you through the backend structure, how to use it, and ideas for how to use it more in the future.

Reminder that Koha and Aspen have their manual and documentation online, so once you tell your AI assistant what program and version you are using, it can help you design and customize with the best of them.

You can do small customizations in your instance of Koha with the IntranetUser JS or CSS system preference.

Here are a couple of mine: Adding the biblionumber to Basket Orders table, and making a few fields in the Marc record populate when you click in the field.

TRY IT IN DEV FIRST if you can! Don’t erase anything you didn’t create!

9 of 14

A great use of AI is for Acquisitions and Collection analysis

(Claude, Gemini seem to work well for these)

AI for Analysis

You can also paste any report results into your AI tool and ask what it tells you about trends: circulation, acquisitions, collection imbalances, etc.

Type a natural-language prompt into your AI SQL assistant tool:

Have AI review your Acquisitions invoice analysis for subscriptions.

Review a packing list vs order list in Koha: what's missing?

10 of 14

AI for Staff Who Serve

Consultant:

Sounding board on how to do certain tasks
Think through workflow problems
Gather ideas from other libraries/websites
Ask suggestions for solutions on data privacy
Policy guidance

Intern for the dirty work:

Calendar coordination
Computer and other resource monitoring
Scribe is an AI tool that creates instructions as you work so you can pass them along to others

Design:

Ideas and coding for customizations using CSS, JavaScript, and HTML within Koha and/or Aspen
Creating presentations, emails, and displays with a given audience, tone, and result in mind
Ideas and creative tools for Programming and Events, and then Marketing them

Using AI to improve your staff’s ability to perform patron services is where you can really get creative.

For staff, use AI as a consultant, an intern that does the boring work, or a designer!

As a Consultant: be the one that comes to meetings with ideas and solutions!

As an intern to do the dirty work: let AI speed through the tasks that suck time from more important (or fun) work, then just review it for accuracy. If you ask it to work on staffing calendars or schedules, review it to be sure it didn’t make up any new staff that don’t actually work there!

Scribe is a tool you can set up to record you doing work, and it will create instructions for that work as you go to help teach others.

Again - VERIFY

As a designer: creation that may be just out of reach or will take extra time to do by hand. Again, use AI to teach you how to do this - say what you are trying to make and ask how to do so. It’ll walk you through the entire process - you just need to provide the unique information.

11 of 14

Automatically cross-references new book acquisitions with local school standards to draft reading lists.

AI for Direct Patron Help

Design a chatbot for common questions with answers based on the information from your library’s website.

Universities require students to participate in a self-guided tour to get to know the library: use AI to build this tour activity.

Create reader’s advisory lists, or design a tool that can generate them.

Design a feedback system, then use AI to analyze that feedback.

Ask for ideas on how to serve certain populations better: those from out of town, those who do not have English as a first language, those with specific disabilities, etc.

Create instructions for the printer, computers, and other equipment with clarity and brevity. Have it include pictures and icons.

Gather Summer Reading Program ideas for each specific age group/grade.

Cross reference circ, pub date, condition notes to flag candidates for weeding.

Analyze hold queues, ILL requests, and purchase suggestions to recommend titles/genres to purchase.

12 of 14

AI for Reference Help

The FEAR: Reference questions that may stump the best of us!

In one of Jane Austen’s books she talks about a book that was really published. What is the title of that book?

There’s a book that was published in the last 10 years or so that was an animal version of Shakespeare’s Hamlet. What book is it?

I want to know the copyright on a book. Its title is "Chautauqua County: A History" by Helen G. McMahon, and it was first published in 1958. Is it still in copyright?

Where is the deadliest volcano on Earth right now? Can I go there? What books will tell me more about them?

13 of 14

The Future of AI for Koha

KohaCon classes about integrating AI - check them out!

The next questions are:

Where do we go from here?

What do we want to see in Koha that AI can do?

How can AI be integrated directly into Koha and our regular processes?

My personal ideas / wish list for our library processes:

If an ISBN or OCLC number has been entered in the record, have a button with the 520 field that will find the summary for that item and import it in.

If the language of the catalog is set to English (or other language), and a foreign language title is detected in the 245$a field, have a button in the 242$a field to generate the English translation, and a button in the 246$a field for a transliteration of the title.

If the 542 field contains information related to the copyright status of a work. Based on U.S. Copyright laws, find and generate the copyright status of a book in this field.

14 of 14

Questions?