The ground is shifting

I. It was never the readers

The story everyone tells about newspapers is that the internet gave away the news for free and readers stopped paying. It is a tidy story and it is mostly wrong. Readers did not abandon journalism. The top fifty American newspaper websites had more average monthly unique visitors in 2022 than in 2014 — about 8.8 million against 8.2 million.1 What collapsed was not the audience. It was the business model sitting underneath it, and that model had very little to do with journalism.

For most of the twentieth century a newspaper was two businesses wearing one coat. One of them reported the news. The other one sold small ads to people looking for a job, a flat, a second-hand car or a lost dog. The second business paid for the first. Classified advertising was high-margin, local, and almost impossible to compete with, because the value of a classifieds section is that everyone in the county already reads it — a network effect defended by a printing press.

The internet did not have to be better journalism to break that. It only had to be a better place to sell a sofa. Robert Seamans and Feng Zhu measured exactly this, comparing newspapers in markets Craigslist entered against newspapers in markets it had not yet reached. Classified advertisers saved an estimated $5 billion between 2000 and 2007 — money that came directly out of newspaper revenue. The papers most dependent on classifieds cut their classified rates, then raised subscription prices to compensate, then lost circulation because of the price rise, and then lost display advertising rates because circulation had fallen.2 A free website for used furniture set off a chain reaction that ended in fewer reporters.

The aggregate is brutal. US newspaper advertising revenue peaked at $48.67 billion in 2000 and was $9.76 billion in 2022 — a fall of about 80% in nominal terms, and far worse adjusted for inflation.3 Newsroom employment in the United States fell 26% between 2008 and 2020, from roughly 114,000 to about 85,000. Inside that number, newspaper newsrooms fell 57% — from about 71,000 to about 31,000 — while digital-native newsrooms grew 144%, from 7,400 to about 18,000.4 Digital growth did not replace what was lost. It was not close.

By 2025 the count of newspapers that had stopped printing since 2005 was around 3,500. Two hundred and twelve American counties had no local news source at all, and 1,525 had exactly one — usually a weekly. Nearly fifty million people now live with limited or no access to local news.5

The lesson is not that old media was lazy. It is that an industry can do its actual job well and still die, because the thing paying for the job was something else entirely — and that something else was easier to take than anyone expected.

Hold on to that, because it is the same shape as what is happening to books. Publishers are not bad at editing. The part of publishing that money controls is not the editing.

II. The re-intermediation of attention

Money does not evaporate. When it leaves an industry it arrives somewhere else, and it is worth being precise about where.

In 2024 Alphabet reported total revenues of $350.0 billion, of which $264.6 billion was Google advertising: $198.1 billion from Search and other properties, $36.1 billion from YouTube ads, and $30.4 billion from the Google Network — advertising Google places on everyone else’s sites.6 For fiscal 2025 the company reported total revenues of $402.8 billion.7

Set that beside the other number. Google’s advertising revenue in a single year was roughly twenty-seven times the entire US newspaper industry’s advertising revenue in that year, and more than five times what the whole American newspaper industry earned from advertising at its absolute historical peak.

Here is the part that matters for publishing. Google did not become the world’s largest advertising company by making advertisements. It makes almost none. It won by occupying the position between the person who wanted to reach an audience and the audience itself, and by being measurably better at that position than the incumbent — cheaper, faster, self-service, and accountable to the click. That is re-intermediation: not the removal of a middleman, but the replacement of an expensive, slow, relationship-based one with a cheap, instant, automated one.

Note the shape carefully, because the optimistic version of this story is wrong in an important way. The internet did not disintermediate advertising. It re-intermediated it, and the new intermediary is larger and more concentrated than the ones it replaced. A single company’s Network segment alone — $30.4 billion of ads sold on other people’s websites — is three times the advertising revenue of every newspaper in the United States combined.

So the honest question for publishing is not “will the gatekeeper be removed?” It is: when the gate moves, who ends up holding it, and on what terms? That question has an answer, and the answer is not automatic.

III. What self-publishing already proved

Self-publishing is not a recent invention and it is not a consolation prize. It is how a surprising amount of the canon reached the world in the first place.

Walt Whitman set type for and paid to print the first edition of Leaves of Grass himself in 1855. Virginia and Leonard Woolf bought a hand press and founded the Hogarth Press in 1917, and it published Mrs Dalloway and To the Lighthouse — Woolf is one of the few major novelists of the century who never had an editor with the power to say no to her. Beatrix Potter, turned down or told to change her book by publisher after publisher, printed The Tale of Peter Rabbit privately in 1901 rather than alter it; it has since sold in the tens of millions across more than thirty languages. Mark Twain, already famous, founded his own publishing house in 1884 and used it to publish Adventures of Huckleberry Finn.8

What is new is not authors publishing themselves. What is new is that they can now reach everyone.

Amanda Hocking began putting paranormal novels on the Kindle store in April 2010 because she wanted to fund a trip to a Muppets exhibition. By March 2011 she had sold more than a million copies of her first nine books.9 E. L. James published Fifty Shades of Grey in 2011 through a tiny digital press after it began life as fan fiction; the trilogy has sold in the region of 165 million copies.10 Colleen Hoover self-published Slammed in 2012 and has since sold more than twenty million books.

Andy Weir is the cleanest experiment of all, because the traditional industry got first refusal and declined. Unable to interest agents, he posted The Martian one chapter at a time, free, on his own website. Readers asked for a Kindle version, so he put one up at 99 cents — the lowest price Amazon would let him set. It sold 35,000 copies in three months, more than the free version had been downloaded. Crown bought print rights in March 2013 for over $100,000.11 The gatekeepers had not judged the book badly. They had never read it.

Hugh Howey then demonstrated the leverage this creates. After Wool succeeded as a self-published serial, he refused to sell the digital rights, turned down seven-figure offers, and in 2013 signed with Simon & Schuster for print only — a deal a major house had essentially never done before.12 He kept the ebook. He kept the reader relationship. The publisher got the thing publishers are still genuinely good at, and nothing else.

And in 2022 Brandon Sanderson asked his readers for a million dollars to self-publish four novels. They gave him $41,756,818, from 185,341 backers — the most-funded project in Kickstarter’s history.13 A single author, going direct, out-raised the annual revenue of most independent publishers.

The volume behind these names is real. Bowker, the American ISBN agency, recorded 152,978 self-published ISBNs in 2010 and 727,125 in 2015, growth of over 375% in five years.14 Self-published ISBNs passed one million for the first time in 2017, reached 1,551,391 titles with ISBNs and BISAC codes in 2018, and stood at about 2.3 million new self-published titles in 2021.15

It would be dishonest to stop there. The same period produced the Authors Guild’s 2023 income survey of 5,699 published authors, and its findings are grim. The median book-related income of a full-time, established author in 2022 was $12,000; with all writing-related work added in, $23,329. Across all respondents, median book income was $2,000 and median total author-related income was $5,000. Fifty-six per cent said that work other than books — journalism, teaching, editing, ghostwriting, events — more than doubled their income.16 The Guild has separately estimated a roughly 42% decline in author earnings over the preceding decade.17

So: the distribution problem is solved and the income problem is not. Both things are true. Anyone selling you self-publishing as a road to money is selling something. What it is actually a road to is existence — the book gets to be in the world, findable, readable, quotable, permanent. That is a different and more important thing, and it is the thing that used to be for sale.

IV. What inference does to the cost of a book

Now the third disruption, and this one is still in progress, so it deserves more care than the first two.

Publishing a book well has always required perhaps a dozen skilled trades: developmental editing, line editing, copy-editing, proofreading, typesetting and interior design, cover design, indexing, metadata and categorisation, ISBN administration, format conversion, distribution onboarding, accessibility, translation, audiobook production, and a jacket description that makes someone pick it up. At professional rates these total thousands of dollars per title before a single copy sells. That cost — not talent, not merit — is the actual gate. It is why a publisher’s advance functions less as payment than as capital, and why the person without capital does not get published.

Inference collapses the marginal cost of most of those trades toward zero. Not all of them, and not equally, and the honest accounting matters:

• Genuinely transformed. Typesetting, format conversion, metadata generation, categorisation, indexing, accessibility markup, translation drafts, and audiobook narration. These are mechanical transformations of a text, and machines now do them at a quality that is acceptable-to-excellent, in minutes.
• Substantially assisted. Copy-editing and proofreading, jacket copy, cover concepting, and the first pass of a structural edit. A machine finds the inconsistency; a person still decides what to do about it.
• Not solved. Judgement about whether a book is good. Knowing which reader it is for. Telling an author the hard truth about chapter nine. Building an audience. None of this is close, and claiming otherwise is how bad books get published.

The audiobook case shows the shape of the change most clearly, because it is already measurable. Audiobook narration was the single most expensive item on an independent author’s budget — a studio, a professional narrator and an engineer for a full-length book. Amazon opened its “Virtual Voice” synthetic narration to Kindle Direct Publishing authors in late 2023; by May 2024, over 40,000 audiobooks made that way were on Audible.18 In 2025 Spotify partnered with ElevenLabs to let self-published authors generate audiobook editions.19 A cost that ruled out audiobooks for most independent authors stopped ruling them out, in about eighteen months.

This is not an argument that machines should write books. It is the opposite. It is an argument that the machine should do the dozen jobs that stand between a written book and a published one, so that the writing — the part only a person can do — is the only part that has to be paid for in talent.

V. What that does not fix — and what we do about it

Section II is the warning. Cheap production does not by itself democratise anything, because when a gate becomes cheap to pass, the scarce thing moves: from production to attention. Two and a half million new titles a year is not a level playing field. It is a haystack. And whoever runs the search over the haystack is the new gatekeeper — which is precisely how advertising ended up more concentrated after the internet than before it.

So the honest position is this. Automating publication is necessary and not sufficient. It removes the thing money buys. It does not remove the thing distribution buys. Anyone who tells you it does is repeating the mistake of 2004.

What GroundSpring Publishing does about it is deliberately narrow and, we hope, deliberately honest:

The spring in our logo is not decoration. A groundspring is water that has been travelling underground for a long time, under its own pressure, and reaches the surface because the ground finally gives way. It does not need a pump. It needs the rock to move.

---

Endnotes

1. Pew Research Center, “Newspapers Fact Sheet,” 10 November 2023. Average monthly unique visitors to the top 50 US newspaper websites: 8,233,544 (2014) and 8,839,848 (2022). Average time per visit fell from 2.59 to 1.48 minutes over the same period — readers stayed, but browsed rather than settled. pewresearch.org ↩
2. Robert Seamans and Feng Zhu, “Responses to Entry in Multi-Sided Markets: The Impact of Craigslist on Local Newspapers,” Management Science 60(2), February 2014, pp. 476–493. doi.org/10.1287/mnsc.2013.1785 · full text (PDF) ↩
3. Pew Research Center, “Newspapers Fact Sheet,” 10 November 2023. Estimated US newspaper advertising revenue: $48,670,000,000 (2000) and $9,760,830,024 (2022). Digital advertising rose from 17% of newspaper ad revenue in 2011 to 48% in 2022 — a larger share of a much smaller total. pewresearch.org ↩
4. Pew Research Center, “U.S. newsroom employment has fallen 26% since 2008,” 13 July 2021, from an analysis of Bureau of Labor Statistics data. pewresearch.org ↩
5. Northwestern University Medill School, Local News Initiative, “The State of Local News 2025,” 20 October 2025. localnewsinitiative.northwestern.edu ↩
6. Alphabet Inc., Form 10-K for the fiscal year ended 31 December 2024. Google advertising $264,590M, comprising Google Search & other $198,084M, YouTube ads $36,147M and Google Network $30,359M. sec.gov ↩
7. Alphabet Inc., “Alphabet Announces Fourth Quarter and Fiscal Year 2024 Results” (total revenues $350,018M for 2024; $307,394M for 2023) and “…Fiscal Year 2025 Results” (total revenues $402,836M for 2025). FY2024 · FY2025 ↩
8. Alliance of Independent Authors, “Famous Self-Publishing Authors: Past and Present”; Hogarth Press founding and output. selfpublishingadvice.org · Hogarth Press ↩
9. Amanda Hocking began self-publishing in April 2010; by March 2011 she had sold more than a million copies of her first nine books. Amanda Hocking ↩
10. E. L. James, Publishers Weekly Publishing Person of the Year, 2012; the trilogy’s reported sales are around 165 million copies. Sales totals for blockbusters are publisher-reported and should be read as approximate. publishersweekly.com ↩
11. Andy Weir, The Martian: serialised free on his website, then sold on Kindle at $0.99, reaching 35,000 copies in three months; print rights to Crown in March 2013 for over $100,000. Andy Weir ↩
12. Hugh Howey’s print-only agreement with Simon & Schuster for Wool, reported January 2013; he retained the digital rights. Publishers Lunch ↩
13. Brandon Sanderson’s “Surprise! Four Secret Novels” Kickstarter closed 31 March 2022 at $41,756,818 from 185,341 backers, the most-funded campaign in the platform’s history. CNBC ↩
14. Bowker self-publishing reports: 152,978 self-published ISBNs in 2010 rising to 727,125 in 2015, growth of more than 375%. Publishers Weekly ↩
15. “Self-Publishing Is Thriving, According to Bowker Report,” Publishers Weekly, 17 February 2023: 1,551,391 titles with ISBNs and BISAC codes in 2018; 2,298,004 in 2021; 2,300,336 new self-published titles counting ISBNs alone, 15% below 2020. ISBN counts overstate distinct books — one title takes several ISBNs across formats — so read them as a measure of activity, not of books. publishersweekly.com ↩
16. “Author Incomes Post Small Gains,” Publishers Weekly, 29 September 2023, reporting the Authors Guild 2023 Author Income Survey of 5,699 published authors. publishersweekly.com ↩
17. The Authors Guild, “Key Takeaways from the 2023 Author Income Survey,” which puts median full-time author book income at $10,000 and notes a fall of roughly 42% since 2009. The Guild’s own figure and the Publishers Weekly figure in note 16 differ because they describe different cohorts (“full-time” against “full-time established”); both are given here rather than picking the more dramatic one. authorsguild.org ↩
18. “KDP Virtual Voice Beta Has Produced Over 40,000 Audible Titles,” Publishers Lunch, May 2024. Publishers Lunch · Amazon’s own description of the feature: KDP help ↩
19. Spotify’s 2025 partnership with ElevenLabs to let self-published authors produce audiobook editions. TechCrunch ↩

Bibliography

• Alphabet Inc. Form 10-K for the fiscal year ended December 31, 2024. US Securities and Exchange Commission.
• Alphabet Inc. Alphabet Announces Fourth Quarter and Fiscal Year 2024 Results and …Fiscal Year 2025 Results. US Securities and Exchange Commission.
• Alliance of Independent Authors. Famous Self-Publishing Authors: Past and Present. selfpublishingadvice.org.
• The Authors Guild. Key Takeaways from the 2023 Author Income Survey. 2023. authorsguild.org.
• Amazon Kindle Direct Publishing. Digital Book Pricing Page and Audiobooks with virtual voice. kdp.amazon.com.
• Milliot, Jim. “Author Incomes Post Small Gains.” Publishers Weekly, 29 September 2023. publishersweekly.com.
• Northwestern University Medill School, Local News Initiative. The State of Local News 2025. 20 October 2025. localnewsinitiative.northwestern.edu.
• Pew Research Center. Newspapers Fact Sheet. 10 November 2023. pewresearch.org.
• Pew Research Center. U.S. newsroom employment has fallen 26% since 2008. 13 July 2021. pewresearch.org.
• Publishers Weekly. “Self-Publishing Is Thriving, According to Bowker Report.” 17 February 2023. publishersweekly.com.
• Seamans, Robert, and Feng Zhu. “Responses to Entry in Multi-Sided Markets: The Impact of Craigslist on Local Newspapers.” Management Science 60, no. 2 (February 2014): 476–493. doi.org/10.1287/mnsc.2013.1785.

Last reviewed 26 September 2026. If a figure here is out of date or wrong, tell us and we will correct it and say that we did.