What Is One Book Worth?

Publishers Weekly built the U.S. Book Show in the space BookExpo left behind, and six editions in, it has grown into the industry's insiders' event: two days in New York, seven hundred professionals, and the best programming in the business for the young people finding their place. The organizers aim the show squarely at the future, and this year the future meant AI.

The CEO panel spent most of its hour on AI in publishing, and the room stayed calm. Dominique Raccah of Sourcebooks warned against a "witch hunt." David Shelley of Hachette worried that detection tools would breed suspicion between publishers and the writers they depend on. Madeline McIntosh of Authors Equity gave the show its quotable line: every publisher is "five minutes away from having an AI controversy." I watched that panel. Everyone on it was careful, reasonable, and choosing their words very carefully.

The money went unmentioned. For a show aimed at what comes next, the hours on stage were not all that futuristic: the newest revenue line in the business, licensing books to AI companies, barely surfaced across two full days. That silence is worth examining, because the numbers are large and getting larger.

Start with the tally. Announced deals for text licensing now total north of $300 million, and Thad McIlroy, who tracks this market as closely as anyone, called that number low when he ran it in November. Wiley booked $40 million in AI licensing revenue in fiscal 2025. Taylor & Francis structured more than $10 million upfront, with recurring payments through 2027. HarperCollins offered authors $5,000 per title in an opt-in program that remains the only Big Five trade deal on the books, and Bloomsbury is now exploring opt-in deals of its own. The buy side has published its own forecast: the Association of American Publishers, in its own amicus brief, sized the AI training license market at $2.5 billion today and projected $30 billion within a decade.

So the money is real, it is moving now, and the industry's own trade association expects it to grow twelvefold.

Here is the strange part. The only benchmark price we have comes from the Bartz v. Anthropic settlement, which priced the works in that case at roughly $3,000 each, $1.5 billion in total, with payments scheduled through 2027. A federal court proceeding produced the only per-work price for a book the public has ever seen. The deals announced on the open market, meanwhile, arrive as totals: publisher-level, aggregate, and unitemized.

That is the gap. Every number in the tally above is a publisher's number. Wiley can report $40 million; nobody outside the building can say what one author with one backlist title received, or consented to, or gave up in the exchange. Was her book in the deal at all? Under what guardrails? For how long? The announcements don't say, and in most cases the author can't find out. HarperCollins stands out in this landscape for a simple reason: we know its per-title terms because authors were asked to accept them.

The gap in knowledge about pricing persists because the infrastructure barely exists. Most trade contracts predate AI training as a concept, which means the rights in question generally belong to authors, not publishers, and nobody built the infrastructure for authors to exercise them: to register a work, set terms, grant or refuse permission, and collect. Thirty-odd startups are now racing to build exactly that. Amlet is one of them, which is why I work here, and every company in this race is betting on the same outcome: a market where per-title terms are ordinary information rather than trade secrets.

Mark Williams at The New Publishing Standard wrote the sharpest piece to come out of the Book Show. His argument runs that publishing's leadership defends creative labor loudly while automating operational work, and that the industry's stance "is not an anti-AI position. It is a position about who controls the economic benefits of AI." He is right, and authors should take the point personally. If the fight is over who controls the economic benefits, then the author's share depends entirely on who sits at the table when terms are set.

The market for books as training data will mature when authors can understand their compensation: what this title earned, what its author agreed to, what the model may and may not do with it. Watch for the terms of Bloomsbury's opt-in, for whether book publishers step into Microsoft's new content marketplace, and for whether anyone follows HarperCollins in publishing a per-title number.

Until then, picture the market as it stands like a hotel with a lit marquee and no rates posted at the desk. You can tell money changes hands. You just can't know what a night costs.