Industries / Media, arts and recreation / Media and entertainment / Publishing
Finding and signing authors
Finding and signing authors is how publishers and magazines choose which books and stories to take on, and on what terms, whether the work arrives through a literary agent, an open call for submissions or an author who has already published it alone.
6
Milestones achieved, from 1881 to 2023.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1881
The first literary agency
Alexander Pollock Watt began handling publishing contracts for his friend, the novelist George MacDonald, around 1875, and by late 1881 had set up A. P. Watt & Co. in London. The University of North Carolina, which holds the firm's records, calls it the world's first literary agency. Watt took 10% of what his clients earned, which the archive calls the basic rate still used today.
Credit
Alexander Pollock Watt
2010
Seventy percent for self-publishers
Amazon added a royalty option to its Kindle Digital Text Platform, the service now called Kindle Direct Publishing, that paid authors who publish their own e-books 70% of the list price, less a delivery charge of $0.15 per megabyte, from 30 June 2010, for books priced between $2.99 and $9.99. Amazon's announcement set this against the 7 to 15% of list price that authors usually got for print books, or 25% of the publisher's receipts for e-books.
Credit
Amazon (Russ Grandinetti, vice president of Kindle content)
2022
Readers fund four novels
In a Kickstarter campaign that ran from 1 to 31 March 2022, 185,341 backers pledged $41,754,153 to Brandon Sanderson's company, Dragonsteel Entertainment, for four novels he had written in secret, against a goal of $1 million, by Kickstarter's count. The money came straight from readers, before any of the books reached a shop.
Credit
Brandon Sanderson; Dragonsteel Entertainment
2022
A court blocks a merger
Judge Florence Y. Pan of the US District Court in Washington, DC stopped Penguin Random House from buying Simon & Schuster. She defined the market as the rights to "anticipated top-selling books", those with advances of at least $250,000: about 2% of the books publishers buy, but 70% of the money spent on advances, about $1 billion a year. She found the merger likely to lower the advances those authors are paid.
LessonA merger can be stopped for what it would do to the people a company pays, here authors, and not only for what it would do to the prices readers pay.
Credit
US Department of Justice, Antitrust Division; Judge Florence Y. Pan
2023
AI stories close a magazine
Clarkesworld, a science fiction magazine that pays its writers and is usually open to anyone, stopped taking submissions on 20 February 2023. Its editor, Neil Clarke, told NPR that by then it had received 700 genuine submissions that month and 500 written by machine. Days earlier he had written that detection tools give false negatives and false positives, and cost more than short-fiction magazines can afford.
LessonOpen submission, the route by which unknown writers reach editors, became costly to keep open once machines could produce plausible stories in bulk.
Credit
Neil Clarke, Clarkesworld
2023
A daily cap on uploads
After suspected AI-written books appeared on Amazon, including books falsely listed under the name of the writer Jane Friedman and mushroom-foraging guides reported to contain possibly dangerous advice, Amazon required authors to disclose AI-generated content and, in September 2023, limited each account to three new titles a day. Miriam Johnson, a publishing lecturer at Oxford Brookes University, told The Guardian that people flooding the store would find a way around the cap.
LessonA platform can cap volume, but a disclosure rule relies on the author's word; Johnson asked how Amazon would check it.
Credit
Amazon Kindle Direct Publishing
Watch for
A mark for human writing
The Authors Guild opened its Human Authored certification, a mark an author can print on a book to say its text was written by people, to every author with books published in the US in March 2026, after a members-only beta from January 2025. It costs non-members $10 a title, certified books are listed in a public database, and the Society of Authors runs a matching scheme in the UK. Authors sign a licence and non-members verify their identity, but the text itself is not tested.
Achieved whenan outside party checks a sample of certified books, for example against drafts or with a detector whose error rates are published, or a major bookseller shows the mark in its listings.
Claimed by
the Authors Guild, March 2026
Software reads the slush pile
Trilogy, a publishing software company, launched Manuscript AI in January 2026. It gives unsolicited manuscripts (the "slush pile") a score out of 100, including a "sales potential" estimate from comparison with 20 years of bestsellers in the genre and a predicted star rating, and Trilogy says it trained the system on about 2.7 million copyright-cleared and public-domain books. Its managing director, Alex Dare, said 94% to 97% of manuscripts are never published.
Achieved whensomeone outside the company compares the scores with what editors later decided, or with the sales of the books that were published.
Claimed by
Trilogy, January 2026
Open
01
Score slush readers on decisions
A test of AI manuscript-screening tools against real editorial decisions. With a magazine's or small press's permission, under its submission terms and on its own computers so that no manuscript leaves them, run the tools over one past year of submissions whose outcome is known, and measure how many of the works the editors accepted each tool would have thrown out.
The briefCredit · unclaimed02
Counting books that earn out
An anonymous public table of book advances and how they paid off, built from royalty statements that authors volunteer. For each book it records the advance band, the genre, the size of the publisher and whether, and when, royalties covered the advance (the book "earned out").
The briefCredit · unclaimed
Open 01 · the brief
Score slush readers on decisions
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of AI manuscript-screening tools against real editorial decisions. With a magazine's or small press's permission, under its submission terms and on its own computers so that no manuscript leaves them, run the tools over one past year of submissions whose outcome is known, and measure how many of the works the editors accepted each tool would have thrown out.
Why it is still open
The figures for tools such as Trilogy's Manuscript AI are the vendor's own. The best-known academic attempt, "Success with Style" (Vikas Ganjigunte Ashok, Song Feng and Yejin Choi, 2013), predicted the success of novels from their writing style with accuracy of up to 84%, but it measured success among books already on Project Gutenberg, by download counts, not among submissions that editors had turned down. This research found published tests of AI against acceptance decisions only for scholarly papers.
Built with what exists
An open-weight AI model run on the publisher's own hardware, such as Mistral Large 4 (see the Note "A trillion open parameters"), or a small model trained further on one graphics chip (see the Note "One GPU to fine-tune"); an export from the publisher's submissions system; Python for the scoring.
First step
Ask one magazine that publishes its acceptance rate for permission to run a free open model over last year's accepted and rejected stories on its own machine.
Done looks like
For at least 1,000 past submissions, a published table giving, for each tool, the share of accepted works it would have rejected at the setting that removes half the pile, with every run repeated three times to show whether the scores hold steady (see the Note "All twenty times").
Open 02 · the brief
Counting books that earn out
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An anonymous public table of book advances and how they paid off, built from royalty statements that authors volunteer. For each book it records the advance band, the genre, the size of the publisher and whether, and when, royalties covered the advance (the book "earned out").
Why it is still open
The public numbers are scattered estimates. At the 2022 merger trial, Michael Pietsch of Hachette said about half the books it publishes earn out, an agent said 20% of the authors at one agency did, the agent Andrew Wylie said 5% of the books he represents did, and the government's economist said more than 85% of contracts for anticipated top-selling books never earn out. The Authors Guild's 2023 survey of 5,699 authors reports income, a median of $10,000 from books in 2022 for full-time authors, but not what becomes of each advance. In June 2020 the young-adult author L. L. McKinney started the hashtag #PublishingPaidMe, under which authors posted their advances to show the gap between what white writers and writers of color, especially Black writers, were paid; one author who took part, Martha Wells, noted that advances leave out royalties, audio and foreign rights.
justice.govauthorsguild.orgverakurian.substack.comgoodreads.com
Built with what exists
Statements shared through an author organization with each author's consent; an AI model that reads documents to pull figures from statements whose layouts differ by publisher, with each figure checked against the page (see the Note "Seven more flagged"); Python or a spreadsheet for the table.
First step
Collect 30 statements from volunteer authors and test whether the extraction gets every figure right.
Done looks like
At least 1,000 books from at least 20 publishers, each with its advance band and earn-out status after five years; 100 statements entered a second time by hand, matching the machine's figures in at least 99% of fields; the table published without names.