Generative AI is rewriting the rules of published content and catalogue workflows. Moving past the noise of speculative disruption, this guide examines actionable standards for ONIX and MARC21. Discover how to properly designate AI software as related works, attribute human prompters, and apply robust provenance tracking without compromising metadata integrity.
In the shadowed neon stacks of the information age, where algorithms quietly forge new works and human creators still hold the line, clear provenance has never been more critical. How do we flag generative AI involvement in published content without descending into chaos?
This guide cuts through the noise for two key standards: ONIX for the book trade and MARC21 for library cataloguing.
No fluff. Just steel.
EDItEUR, the body behind ONIX, has issued practical guidance in its Application Note: Aspects of AI in ONIX.
Core rule: disclose AI use when it goes beyond de minimis assistance. Transparency serves retailers, platforms, and readers.
<UnnamedPersons>09</UnnamedPersons> (List 19): Officially designates AI (Artificial Intelligence). Used when the creation of book content (text, images, or structural layout) is accomplished via a generative AI model or technique.Managed via <EpubUsageConstraint> blocks using List 145 (Usage type) to explicitly grant or restrict machine harvesting:
Paired with usage constraints from List 146 to define whether the rule is an open permission or a strict prohibition:
01: Permitted unlimited (Open permission / no restrictions).02: Permitted subject to limit (Allowed, but capped by characters, copies, etc., specified via <EpubUsageLimit>).03: Prohibited (Strictly blocked / opt-out enforced, commonly used for TDM and AI training blocks).<Contributor>
<SequenceNumber>1</SequenceNumber>
<ContributorRole>A01</ContributorRole>
<PersonName>Schmalfuß, Olaf</PersonName>
<ProfessionalAffiliation>
<Affiliation>DataMercs Cybernetics</Affiliation>
</ProfessionalAffiliation>
</Contributor>
<Contributor>
<SequenceNumber>2</SequenceNumber>
<ContributorRole>Z01</ContributorRole>
<UnnamedPersons>09</UnnamedPersons>
<BiographicalNote>Generated via Google Gemini large language model inference.</BiographicalNote>
</Contributor>
<EpubUsageConstraint>
<EpubUsageType>13</EpubUsageType>
<EpubUsageStatus>03</EpubUsageStatus>
</EpubUsageConstraint><UnnamedPersons>09</UnnamedPersons> (the code for generative AI).This approach is already influencing supply chains, with some platforms labelling or restricting AI-heavy content.
Aspects of AI in ONIX (PDF) (EDItEUR, current as of 2026).
Library systems move more deliberately, with a strong emphasis on metadata integrity. Groundbreaking practical guidance from the PCC Standing Committee on Standards (FAQ: Cataloging of Resources Generated Using Artificial Intelligence Software) has clarified how to handle synthetic contributions without breaking data structures.
The American Library Association (ALA) in their Guidance on the Use of Artificial Intelligence in Libraries highlight the “Human-in-the-Loop” mandate:
AI can draft 505 contents or summaries, but a human cataloguer must authenticate the record, ensuring ethical AI usage and preventing hallucinated subject headings from corrupting union catalogues.
$e prompter, $e editor, $e author).Computer poetry, AI art, or AI-assisted content (LCGFT/AAT) to explicitly characterize the nature of the generated intellectual or artistic output.$a, generation dates, and algorithmic confidence metrics via $c, separating algorithmic provenance from descriptive human notes.While field 883 remains the designated vehicle for machine-generated metadata provenance, it is still relatively bleeding-edge for many traditional libraries, implementation and support may variy across Integrated Library Systems (ILS).
$a – Generating application or system name (e.g., Gemini v2.5)$d – Date of generation (YYYYMMDD format)$q – Generating agency or prompting entity (DataMercs)$c – Algorithmic confidence metric or generation parameters (0.98 confidence)$u – Uniform Resource Identifier pointing to the provenance/audit log (https://datamercs.net/provenance/...)883. Always back up machine provenance with a human-verified 588 source of description note where interoperability is critical.OCLC’s Bibliographic Formats and Standards (BFAS) and the PCC Task Group on AI and Machine Learning reinforce these practices, balancing transparency with rigorous quality control.
To see how these principles translate into raw syntax, consider the following production-ready MARC21 bibliographic snippets aligned with the latest PCC and OCLC frameworks.
000 01890nam a2200385i 4500
001 9910987654321
008 260725s2026 bera b 000 0 eng d
020 ## $a 9783110999998
040 ## $a DE-100 $b eng $e rda $c DE-100
100 1# $a Schmalfuß, Olaf, $e author, $e prompt engineer.
245 10 $a Synthetic Worlds : $b A Neural Experiment in AI-Generated Cyber-Fiction / $c Olaf Schmalfuß, Gemini & Kizuna AI.
264 #1 $a Berlin : $b DataMercs Cybernetics, $c 2026.
300 ## $a 210 pages : $b illustrations (some color) ; $c 21 cm
336 ## $a text $2 rdacontent
336 ## $a still image $2 rdacontent
337 ## $a unmediated $2 rdamedia
338 ## $a volume $2 rdacarrier
500 ## $a Prose and neural-net code snippets were generated collaboratively via large language model inference using Google Gemini based on iterative prompt engineering by Olaf Schmalfuß; synthetic visual conceptualization and persona vectors co-designed with Kizuna AI (Virtual Entity).
520 ## $a An immersive literary and techno-cybernetic experiment exploring how neural text generation, human prompting matrices, and autonomous virtual entities reshape narrative topology, speculative structures, and the ontological architecture of the bibliographic metadata grid.
588 ## $a Description based on machine-vision analysis of the resource package; abstract and controlled subject vectors synthesized via automated parsing, curated and verified by human editorial oversight.
650 #0 $a Artificial intelligence $x Creative applications.
650 #7 $a Cyberpunk fiction. $2 lcgft
650 #7 $a Generative artificial intelligence $x Literary themes. $2 lcsh
650 #7 $a Generative KI. $2 gnd
650 #7 $a Human-AI interaction. $2 lcsh
650 #7 $a Human-computer interaction. $2 mesh
655 #7 $a Computer poetry. $2 lcgft
655 #7 $a Synthetic media. $2 lcgft
655 #7 $a Speculative fiction. $2 gnd
730 0# $i Related work: $a Gemini (Neural network software)
730 0# $i Related work: $a Kizuna AI (Virtual persona / Algorithmic character)
883 ## $a Automated metadata enrichment and vector indexing $d 20260725 $q DataMercs $q Gemini v2.5 $c 0.98 confidence $u https://datamercs.net/provenance/log-20260725To maintain interoperability with next-generation discovery layers, AI software and virtual personas must be anchored to structured authority files rather than treated as simple text strings.
bf:Work related to a bf:Software entity via custom relationship properties) or Schema.org (CreativeWork utilizing properties like softwareApplication or contributor paired with additionalType for virtual personas).MARC21
ONIX offers ready-to-implement, market-oriented solutions – ideal for publishers and distributors.
MARC21 focuses on precise, integrity-first cataloguing, providing nuanced frameworks for treating AI as a related work alongside human operators.
In both realms, the rule is the same: Transparency builds trust in the age of the EU AI Act and beyond.
Standards evolve. Check the sources. The archives endure when the records stay honest.
Title Graphic: “Ghost in the MARC”
generated with Grok Imagine by Grok (xAI) in collaboration with the author
For attribution, please cite this work as
Schmalfuß (2026, June 23). OS DataMercs: Ghost in the MARC: Cataloguing the Synthetic Shift in Bibliographic Metadata. Retrieved from https://www.datamercs.net/posts/2026-06-23-ghost-in-the-marc-cataloguing-the-synthetic-shift-in-bibliographic-metadata/
BibTeX citation
@misc{ghost-in-the-marc-ai-cataloguing,
author = {Schmalfuß, Olaf},
title = {OS DataMercs: Ghost in the MARC: Cataloguing the Synthetic Shift in Bibliographic Metadata},
url = {https://www.datamercs.net/posts/2026-06-23-ghost-in-the-marc-cataloguing-the-synthetic-shift-in-bibliographic-metadata/},
year = {2026}
}