What BiblioScan AI does
BiblioScan AI is an AI cataloging tool developed by BiblioLabTech for the slowest part of technical services: original cataloging of print material. It uses image recognition and language models to capture bibliographic data from photos of a book and pre-fill a MARC21 record in your library system, reducing manual work and transcription errors. It is not a generic OCR tool. It does not just read the text on a page; it classifies each piece of text by the role it plays in the record, which is the difference between transcribing a page and cataloging it.
How it works, step by step
- Capture. Someone on your team photographs the cover, the title page with its imprint data and the table of contents, using a phone or any camera. No dedicated scanner or workstation is needed.
- Recognition. OCR turns the image into text, and vision and language models identify what each fragment is: title, author, publisher, ISBN, year, edition.
- Mapping to MARC21. The data is organized into the fields and subfields your catalog expects, following the cataloging profile agreed with your library.
- Pre-filling. The draft record appears directly in Koha, DSpace, Alma or your current system, with no copying and pasting between tools.
- Review. The cataloger compares the proposal with the book, corrects it, decides classification and subject headings, and approves. Only then is the record saved.
Our guide to AI cataloging goes deeper into each stage.
Which MARC21 fields it fills, and what stays with the cataloger
Fields that transcribe printed information are resolved with high reliability; fields that require bibliographic judgment always need a person.
| MARC21 field | Content | Typical behavior |
|---|---|---|
| 020 | ISBN | Very reliable |
| 100 / 700 | Authors | Name transcribed as printed; the authorized form comes from authority control |
| 245 | Title and statement of responsibility | High reliability; check the split between subfields |
| 250 | Edition statement | Reliable when printed in the book |
| 264 | Place, publisher and date | Reliable; the classic error is taking the printing or copyright date |
| 300 | Physical description | Partial; pagination and dimensions are checked against the item |
| 490 | Series statement | Detected when printed on the book |
| 500 | Notes | Proposed when the item supports them; always reviewed |
| 650 | Subject headings | Candidates suggested from the table of contents; your vocabulary decides |
| 082 / 080 | Classification | Suggestion only; your collection policy decides |
Three decisions always stay human: authority control, subject headings assigned from your controlled vocabulary, and classification. That is why BiblioScan AI produces proposals and never writes straight into the catalog. An error caught in review costs seconds; an error propagated to thousands of records costs a cleanup project.
How accurate it is, and how we measure it
BiblioLabTech publishes a reference figure for BiblioScan AI of 99.9% extraction accuracy, with about 95% of the work automatic and 5% left to human review. Treat it as a reference, not a project guarantee: the real review rate depends on photo quality, the condition and type of material, and how demanding your cataloging policy is. Low-confidence cases, such as a blurred photo or a book without clear imprint data, are flagged instead of being filled with an invented value.
Every project therefore starts with a measured pilot. We establish how many records per hour your team produces today and with what correction rate, then measure the same with BiblioScan AI on your own material. It is the only honest way to give a figure, because every library starts from a different place.
Time saved compared with manual cataloging
| Method | Time per record | Where the time goes |
|---|---|---|
| Original manual cataloging | 10 to 25 minutes | Full transcription, authority checks, classification |
| Copy cataloging through Z39.50 | 2 to 6 minutes | Finding a good match and adapting it |
| AI proposal plus review | 2 to 6 minutes | Reviewing, correcting, classifying, assigning subjects |
These are orders of magnitude from BiblioLabTech's projects. The point is not that AI is as fast as Z39.50; it is that AI brings original cataloging down to copy-cataloging times for material that has no record anywhere else: theses, institutional publications, local and regional editions, old donations. Z39.50 resolves what the world has already cataloged, and BiblioScan AI resolves what nobody has.
Where it fits, and where it does not
BiblioScan AI works best with modern monographs with clear print, which make up the bulk of most backlogs. With rare books, non-Latin scripts, serials or audiovisual material, the gain is smaller and the review heavier, and the honest recommendation is to keep cataloging those by hand. It processes material in Spanish, English and Portuguese; collections in other languages should be tested in the pilot before you commit the workflow to it.
For born-digital documents such as PDF theses and reports, the right tool is MetaExtract API, which works from the document instead of a photo. Our guide to AI metadata extraction explains the difference. Libraries with both a print collection and a digital repository often use the two tools side by side.
Included with Koha, available for any system
Every Koha implementation by BiblioLabTech includes BiblioScan AI at no extra cost. It is also licensed on its own for libraries that already run a system, including commercial platforms: it integrates with Koha, DSpace and Alma, as well as ABCD and other library systems, without migrating your catalog.
BiblioScan AI is licensed as an annual service that covers implementation, configuration of your cataloging profile, support and improvements during the period. The price depends on the number of branches or libraries using it, the expected processing volume and the integrations required. Implementation takes six to eight weeks from diagnosis to deployment, including the pilot, or four to five weeks when your cataloging profile is already defined and your system exposes a standard integration.
Conditions for processing, retention and deletion of the images are agreed in the contract. Because they show covers and title pages of published material, no personal data is involved.
Where BiblioScan AI runs today
- Universidad Continental (Peru) uses BiblioScan AI and MetaExtract API on top of the commercial library platform it already had, without migrating. Both services were contracted at different times and both have been renewed. Read the case study.
- Universidad Católica Luis Amigó (Colombia) added BiblioScan AI for its print collection after putting MetaExtract API into production on its DSpace repository and ABCD catalog; BiblioScan AI has also been renewed. Read the case study.
- Escuela Colombiana de Ingeniería Julio Garavito (Colombia) catalogs in Koha with BiblioScan AI, with mandatory human review before each record is saved. Read the case study.
BiblioLabTech works remotely with libraries in any country. Tell us about your collection and your current system: the next step is a conversation about your material, not a generic demo.