How I Use AI in a One-Person Library

From the desk of a solo school librarian · ~7 min read

I’m a school librarian, not a programmer. For most of my career, the genuinely useful free tools — an open-source catalog, a custom lookup tool, a tidy way to digitize records — sat just out of reach, because taking advantage of them seemed to require skills I didn’t have. That’s changed. An AI assistant has become the most useful thing on my desk, not because it does my job, but because it lets a non-technical librarian finally pick up tools that used to need a specialist. Here’s what that actually looks like, day to day.

Turning a photo of a book into catalog text

Cataloging often means typing out a table of contents, or the text off a cover, into the record. It’s slow, fiddly work. Now I photograph the page — I keep a cheap scanner on my desk, but a phone photo emailed to myself works just as well — and hand the image to the AI with a prompt that spells out exactly how I want the text formatted. It reads the image and returns clean, consistent text I can drop straight into the record.

These are the two prompts I actually keep on hand. The first turns a table of contents into a library-style contents note; the second tidies the text off a cover or jacket. Yours would encode your catalog’s rules — the point is that you write them down once and the AI applies them every time.

Prompt 1 · contents note from a table of contentsIn this chat, I will upload book table-of-contents photos. Please make library-style contents notes following these formatting rules:
– Use " -- " between main headings
– Use " ; " between secondary (sub-level) headings
– Use a period after a main heading before secondary headings (instead of a colon)
– Use sentence-style capitalization: capitalize only the first word and any proper nouns
– Omit page numbers
– For sub-sub-level items (like items within a Chapter Project or Chapter Tests), put the individual items in parentheses after the group title, separated by " ; "
– If the contents note would run more than 500 words, use only the levels needed to stay under that limit
– Drop unit/chapter designations and numbers (e.g. "Unit 1 Plants" becomes "Plants")
– If items list authors, include them with a "/" between the item and the author — as long as that keeps the note under 500 words

Prompt 2 · transcribing cover or jacket textTask: digitize the text from book-cover photos.
Formatting rules for any bulleted lists:
– Convert the list to paragraph form
– Separate items with semicolons
– Capitalize only the first item; all others lowercase
– End the final item with a period
– Combine separate paragraphs into one single paragraph
Ready for the first photo?

The one discipline that matters: always proofread the result against the book. AI transcription is good, not perfect, and a catalog record is a promise to your patrons.

Building small learning tools for students

A librarian is also a teacher, and AI is surprisingly good at turning a lesson idea into a working, interactive tool. One I use: a little citation-practice module where students build a citation piece by piece and get immediate feedback — the kind of thing that would once have required a web developer. I describe what I want in plain language and the AI writes it.

Example promptBuild a simple interactive web page where a student practices writing an MLA “Works Cited” entry for a book: they fill in author, title, publisher, and year, and it shows them the correctly formatted citation and points out any mistakes.

And here’s proof that the prompt doesn’t need to be polished. This is the actual, word-for-word message that started the biggest student project I’ve built this way — what grew into a whole suite of interactive skills modules, now live on my school library’s website:

The real prompt that started it allHey Claude, I'd like to construct online interactive modules that teach middle schoolers and high schoolers about Research/Searching; Citations; and Digital Citizenship (maybe based on the uploaded lesson plans) -- my idea is that a student could go to the library web page, click on a link, choose a category to learn about, go through a lesson (hopefully not boring!) and, in 3-5 lessons, learn about these areas. What do you think? Is this possible? What would we need to make it happen?

Wandering clauses, parentheticals, and all. But notice it has the essentials: who the learners are, what the experience should feel like (“hopefully not boring!”), how students would reach it, and an open invitation to plan. That one message led to a working prototype the same afternoon.

Because these are small, self-contained tools, you can iterate freely: “make it harder,” “add journal articles,” “use APA instead.” You don’t need to know how it’s built to keep improving it.

And bigger things, too

This very website — the Dewey Lookup tool you’re reading guides on — was built this way: a librarian with an idea and an unmet need, working step by step with an AI to design it, write it, and put it online. If a solo librarian can stand up a public web tool and a free library system, the ceiling on what’s “too technical” has moved a long way.

The prompt habits that make it work

You don’t need special words. A few simple habits get dramatically better results:

Always give context. Tell the AI who you are, what you’re trying to accomplish, and any constraints. “I’m a solo school librarian formatting a contents note for our catalog” gets a better answer than a bare request.
Let the AI sharpen your prompt. When you’ve worked through something in a chat and expect to do it again, ask: “Rewrite this as a clear, reusable prompt I can paste next time.” Save the result.
Show an example. One good example of the output you want — a sample contents note in your exact format — teaches the AI faster than a paragraph of description.
Ask for steps, not just answers. “Walk me through this one step at a time” turns a daunting task into a series of small, doable ones.
Name the format you want. “Give it to me as a bulleted list,” “as a single paragraph,” “as a table” — saying so up front saves a round of edits.

You’re still the librarianAI can be confidently wrong, so verify anything that goes into a record, in front of students, or out to your community. The assistant drafts and explains and unblocks; the professional judgment stays yours. That’s not a limitation — it’s the right division of labor.

None of this requires you to become technical. It requires you to be willing to ask, to give context, and to check the work — things librarians are already good at. The tools that used to belong only to bigger libraries are within reach now, one conversation at a time.