On Tuesday, Frankfurt, Germany–based digital book distribution company Bookwire hosted "All About AI In Publishing," a three-hour online event offering presentations on technological developments in the industry. The event attracted nearly 1,000 registered attendees.
At the event, executives from the company, which was an early advocate for broader global audiobook distribution and has championed using NFTs to create unique digital artifacts, revealed it is now positioning itself as an intermediary to facilitate formal rights licensing agreements to AI companies.
"AI companies value the quality of content because the difference between an average AI model and a high-performance AI model also lies on the quality of the content that has been trained on, not only in the algorithm," Nicholas Bräuer, general counsel for Bookwire, said.
The push toward formal licensing agreements has been much debated, especially in the wake of news that seemingly nearly all major AI platforms trained on at least some, if not significant amounts of, unlicensed content. Bräuer cited several pending lawsuits, including cases involving the New York Times and OpenAI, UMG Recordings and Suno AI, and various German cases including GEMA and OpenAI.
"[It's] very important that we obtain more clarity through court decisions about what constitutes the legal framework of AI training in different jurisdictions," Bräuer said.
Recent developments suggest AI companies are willing to pay for content access. Bräuer cited Anthropic's still-pending agreement earlier this month to a settlement totaling $1.5 billion, agreeing to pay $3,000 per title to authors, though he clarified the payment was "not for compensating the training itself, because it's under fair use, but for the illegal use of the data sets." Bräuer added that the proposed settlement "is also a good sign that AI companies acknowledge the value of content."
Eric Bartoletti, Bookwire's head of business development, identified four key factors driving AI companies toward licensing deals: quality content requirements, access to non-public data as a competitive advantage, legal compliance needs, and sustainability of data flows.
"AI companies depend on authors continuing to provide them with a substantial volume of attractive, diverse content in the future for training purposes," Bartoletti said. "It's just much easier for them if they get a broad offering of different formats, different genres, different languages they can select from."
The revenue potential extends beyond traditional training licenses. Bartoletti explained that AI companies seek content "not only to train their models, but also to embed them in the answers given, for example, by chatbots."
Bookwire's three-step process
Bartoletti said that Bookwire has developed a structured approach for publishers interested in AI licensing. The first step involves catalog assessment and establishing publisher requirements and restrictions.
"We just have a conversation getting a better understanding of your current catalog. What is in there, which genres, which languages, which formats," Bartoletti said. "What are your red lines that you're not willing to cross in terms of what your content can be used for?"
The second phase involves signing a non-binding letter of intent that allows Bookwire to present publisher catalogs to AI companies. And the final step occurs when AI companies make offers. Publishers retain decision-making authority over whether to proceed with specific deals.
Notably, Bookwire positions itself as an aggregator rather than acquiring broad rights. "You do not transfer any general AI rights to Bookwire," Bartoletti noted. "It's always deal specific."
Bookwire has reported interest from AI companies for both fiction and nonfiction content, each of which has differnt applications. Bartoletti explained that fiction proves valuable for fine-tuning language models, while nonfiction content serves information-based applications.
Bartoletti added that the company is also seeing "a lot of interest" in audiobooks in addition to e-books.
Bräuer noted that when it comes to the terms themselves, licensing agreements include restrictions on content reproduction. "We make sure that all the contracts we are closing have restrictions, so there's no verbatim reproductions of the book content, and if it's like a citation, it's limited to a certain amount of words like 50 words or 100 words," he said.
Contract duration varies between time-limited licenses, which expire and require the AI company to no longer return results based on that licensed content, and perpetual agreements. Bräuer stressed that "it's not possible to untrain or unlearn an AI model," and emphasized that pricing reflects these different terms: "Perpetual licenses are more valuable."
Bartoletti positioned AI licensing as essential for publishers to control the use of their content, particularly given the challenge of enforcing AI companies' compliance with opt-out measures. "Signing bilateral licensing agreements with AI companies is a great way to clearly define what AI companies are actually allowed to do with your titles, for which use cases they can use them, and to get your consent and a fair remuneration," he said.
The executives reported having already closed successful AI licensing deals and continue building their network of AI company contacts across different market segments, from large foundational model companies to specialized AI developers. They expressed a sense of urgency about addressing this issue as soon as possible.
"There's a lot going on, a lot is happening. There is a lot of momentum and we should all use that to our advantage," Bartoletti said. "The time for action really is now."
