Signaling AI Involvement and Human Authorship in ONIX Metadata
Artificial intelligence is a controversial topic in publishing, with discourse dominated by both publishers’ fears over the unauthorized use of published content for training large language models and the threat to creator’s livelihoods. But even in an industry with an often hostile view of AI, there are legitimate and increasingly common uses, most particularly across a range of administrative and workflow tasks. The recent survey report AI use across the North American book industry 2026 by BISG and BookNet Canada emphasized this.
The Case for Transparency
There is a seemingly inexhaustible range of books of non-human authorship. Almost 90% of respondents to the 2026 industry survey mentioned the above cited concerns over both the accuracy of AI-generated content and the ‘flooding’ of retail platforms by AI-generated books. Whether it's travel guides with text scraped automatically and indiscriminately from Wikipedia, or low-effort genre novels created by—or at least with the assistance of—an artificial intelligence, these increasingly crowd the market, impacting negatively on books by real humans. In reaction to this, bookstores and publishers who wish to differentiate their products from AI output should aim to be transparent by:
-
Identifying those books that are created or assisted by AI, and
-
Stating positively that other books have no AI involvement
In ONIX metadata, it’s possible to do both, and this should help boost trust in published content.
How to Indicate AI Contributions in ONIX
First, ONIX metadata treats AI as a contributor. A book written in Spanish by a Mexican writer, but translated into English using AI, could be listed as:
<Contributor>
<SequenceNumber>1</SequenceNumber>
<ContributorRole>A01</ContributorRole> <!-- written by -->
<PersonName>Valeria Lucia Garcia Moreno</PersonName>
<PersonNameInverted>Garcia Moreno, Valeria Lucia</PersonNameInverted>
</Contributor>
<Contributor>
<SequenceNumber>2</SequenceNumber>
<ContributorRole>B06</ContributorRole> <!-- translated by -->
<FromLanguage>spa</FromLanguage>
<UnnamedPersons>09</UnnamedPersons> <!-- generative AI -->
</Contributor>
Note the use of <UnnamedPersons> to specify the use of generative AI, a tag which is also used for attributions like ‘Anonymous’ or ‘et al’. (The language of the book, English, would be listed elsewhere in the ONIX.)
Any other role—even author—can be specified as AI, and it can also be useful to state that a human creator has made use of AI assistance. Unfortunately, there is little incentive for a publisher to add this to their metadata proactively, as this may compromise copyrightability in the U.S. Pressure from bookstores that wish to differentiate human creativity or expertise from AI-generated material will certainly be required.
One of the most common cases is the use of an AI-generated voice as the reader of an audiobook. These are often books that would not otherwise be available in audio format at all, making the range of available products more diverse and benefiting accessibility:
<Contributor>
<SequenceNumber>3</SequenceNumber>
<ContributorRole>E07</ContributorRole> <!-- read by -->
<UnnamedPersons>07</UnnamedPersons> <!-- synthesized voice -->
</Contributor>
Depending on the code value in <UnnamedPersons>, this might be an AI that has been licensed and trained to sound exactly like a real voice actor (an ‘authorized voice replica’ in the terminology recommended by the Audio Publishers Association), or a more generic AI-generated voice. Either way, the real name of the voice actor or the 'brand name’ of the generic voice can be included within the <Contributor> composite using <AlternativeName>.
The use of AI to generate image descriptions for accessibility purposes can also be a positive for accessibility of e-books.
How to Indicate No AI Involvement in ONIX
Second, there is the case where there is no AI involvement, and the author or publisher wishes to make this clear. This might be done in one of two ways.
1. Third-Party Human Certification
There is a small but growing number of organizations that will provide a third-party certification that an author or other creator is a real human, for example the Author’s Guild and the UK Society of Authors provide an online ‘Human Authored’ certificate. Proudly Human and Books by People do something similar, and there are doubtless many others. Each of these services is relatively new, and although they currently have some limitations, they aim to be a trusted source of provenance information.
Each effectively requires an affidavit or statement with some legal value from the author, they provide a logo for use on the book itself, and there is a URL linking to the online certificate that can be verified by a potential purchaser. It is this URL that can be included in the ONIX metadata:
<Website>
<WebsiteRole>55</WebsiteRole> <!-- human-created certification -->
<WebsiteDescription>Human Authored certified</WebsiteDescription>
<WebsiteLink>https://humanauthored.co.uk/books/can-you-solve-the-murder-by-antony-johnston</WebsiteLink>
</Website>
Note that these certifications certify the combination of the creator and the work, and that the certificate for one work does not automatically apply to other works by the same creator.
2. Publisher Declarations
The alternative here is that the publisher makes some declaration about human creation–perhaps a blanket statement of no use of AI for content creation across the publishing company as a whole, or a book-by-book statement. This may be something of a blunt instrument, but again, a URL leading to a statement can be included in <Website> using a different role code. Such a first-party statement would probably be most valuable for publishers who have well-known and respected brands.
Do involve your company’s legal department in creation of such a company-wide declaration.
The Road Ahead
Of course, short of legal requirements for disclosure, there is no real solution for bad actors, or for those legitimate publishers who avoid AI-generated content but provide neither certification nor declarations in their metadata. Further, experience with FSC logos for paper from well-managed and more sustainable forestry suggests that it will likely take a long time for certification-and-logo schemes like Human Authored or Proudly Human to become a strong factor in consumer selection. But the first step for publishers and booksellers must be towards transparency and provision of information about their use of AI in content creation, both in the metadata and on the book, so purchasers can be informed in advance.
Further details are available in EDItEUR’s ONIX application note Aspects of AI in ONIX.
Join the Conversation
Interested in helping shape the industry's response to AI? BISG's AI Working Group brings together professionals from across the publishing ecosystem to discuss emerging developments, share experiences, and identify opportunities for industry collaboration. We welcome your participation. To join, e-mail info@bisg.org to express your interest and get on the mailing list for meeting invites, agendas, and other key communications.