"AI-powered" appears on almost every library software page now, and most of it means one chatbot bolted onto a search box. That is worth saying plainly before anything else, because the honest version of this topic is more useful than the marketing version — and it is a shorter list than the pitch decks suggest.
This is what actually exists, in a real library platform, today: ours. Not a demo, not a slide — features a librarian uses on a normal Tuesday.
What is real right now
- Drafting a catalog record from an ISBN. The assistant looks a title up against Google Books and Open Library and drafts the record — title, author, publisher, subjects where available. A librarian reviews it before anything is saved.
- A catalog health score. Dozens of data-quality checks — missing fields, likely duplicates, records with no usable classification — rolled into a single number you can track over time, so a cleanup project has a measurable before and after.
- Bulk fixes, reviewed once. The same correction applied across many records in one reviewed, all-or-nothing batch, instead of opening each record by hand.
- Suggested author marks and DDC numbers. Worked out from fixed cataloging rules and the library's own conventions rather than guessed, and confirmed by a librarian before they are saved.
- Reports in plain English. Ask "what was most borrowed this term" and get an answer, instead of building a query or learning a report wizard.
- Voice input, in more than one language, for when typing is not the fastest way to ask.
The detail that matters more than any single feature: nothing writes to the catalog unattended. Every one of the above is a proposal a person reviews and confirms. That is not a limitation bolted on for safety theatre — it is the design, because a catalog is not a place to let a language model improvise.
What is still marketing, everywhere
Being equally honest about the other side of the list:
- AI that assigns classification or call numbers on its own. Suggesting a number is real; saving it with no cataloger looking is not something a library should want, and not something we ship.
- Fully autonomous acquisitions or weeding. Suggesting candidates from real usage data is a reasonable near-term feature; deciding and ordering without a person in the loop is not something any library platform, ours included, actually ships.
- Robots that physically reshelve or find misplaced books. A genuinely interesting idea, and one we are looking at building — but it does not exist as a shipped product anywhere we are aware of, and a page that implies otherwise is selling a render, not software.
Why the distinction matters
A library evaluating software is usually doing it once, for years. Buying against a features page that turns out to be aspirational is expensive to discover after the contract is signed — and a supplier willing to overstate cataloging AI is worth being cautious about on everything else in the same page.
The useful question to ask any vendor is not "do you have AI" — everyone will say yes. It is: show me the exact write path, and show me what happens if it is wrong. If the answer does not involve a human confirming a change before it commits, that is worth knowing before you buy.
What this looks like in practice
Two follow-up pieces go into specific parts of this in more depth: how AI cataloging actually cuts copy-cataloging time, with a real before-and-after, and what changes when you can ask for a report instead of building one.
Want to see it against your own catalog? The AI cataloging assistant page has the full breakdown of what ships today and what we can build for your library specifically.