Can you make money writing a novel with AI? What the evidence shows

A broader market study—with important limits

A working paper by Tuhin Chakrabarty, Xinyue Liu, Jane C. Ginsburg and Paramveer Dhillon examines 14,419 predominantly self-published genre-fiction ebooks released from January 2023 through March 2026, with sales data through June 2026. The researchers classified books using full-text AI detection. Their “substantial” category means more than 25% of the text was flagged; it is not an author-confirmed account of how a book was written. Read the working paper, version 3, August 3, 2026.

The paper reports that selling titles grew faster than revenue, with revenue per selling book falling across most genres. These are market measures, not verified author profits. Its comparisons are observational, and the paper is listed as under review. Detector classifications introduce uncertainty; “no detected AI” does not establish that an author never used AI. Publication status and abstract.

For a prospective writer, this is useful context about competition. It cannot tell you what a careful revision workflow is worth, whether your particular story will find readers, or what you should expect to earn.

Three examples, three kinds of evidence

Jennifer Lepp: a working novelist using AI in her writing process

Josh Dzieza’s July 2022 reporting for The Verge followed Jennifer Lepp, who writes as Leanne Leeds, using Sudowrite in her fiction workflow. It described annual income that could reach the low six figures. Lepp was already an established novelist; the figure was not presented as profit from a separately measured set of AI-assisted books. This is a credible account of a professional writer adopting AI, with no basis for calculating how much extra income the tool produced. Read the reporting, July 20, 2022.

Joanna Penn: a book campaign with a public funding record

The Kickstarter campaign for J.F. Penn’s Death Valley closed in April 2025 with 231 backers pledging £10,794. That total is visible on the campaign itself, making it a stronger checkable financial record than an isolated earnings screenshot. It measures pledges to a book project, however, rather than the author’s take-home income. View the campaign, March 30–April 15, 2025.

In her year-end review, Joanna Penn described AI use for the cover’s background image, custom endpapers, a trailer and an audiobook narrated with her own voice clone. The account supports a specific example of AI-assisted production and publishing; it does not establish that AI wrote the novel’s prose. Read Penn’s account, December 29, 2025.

A campaign’s obligations matter as much as its headline total. To understand its financial outcome, we would need the amounts actually collected and spent on producing and fulfilling rewards, fees, promotion and other expenses. The public pledge total alone cannot answer that question.

Coral Hart: a substantial claim with substantial limits

AFP’s October 2026 profile of Coral Hart reported more than 370 books since February 2025. It described AI-generated first drafts and editing followed by human proofreading and rewriting. Hart estimated sales of more than one million books, and the report put her earnings above $1 million that year. These figures are reported claims, not independently audited results. The article does not provide accounts establishing net profit or a detailed reconciliation of revenue and costs. Read Thomas Urbain’s AFP report, October 25, 2026.

The same report describes prior publishing experience, training other writers and a forthcoming software business. That commercial context is worth retaining when assessing an unusually large success story. It does not disprove the claim, but this article should not turn it into a typical outcome or a forecast for someone writing a first book.

What “making money” actually tells you

Before comparing two authors’ results, establish what each number counts. A customer’s payment for a book, a royalty deposited in an author’s account and money left after expenses describe different stages of the same transaction. Campaign pledges add another distinction: a commitment to pay is not, by itself, a final account of collected funds and fulfilled obligations.

Also ask what came before the result. An established catalog, an engaged mailing list or years of publishing experience may contribute substantially to sales. None of these examples isolates the effect of AI against an otherwise identical publishing process. They are individual cases, not a representative survey from which to estimate a beginner’s likely earnings.

A modest book budget makes the distinction clearer

Consider an invented example, unrelated to the authors above. A writer receives $1,000 after retailer deductions and spends $250 on editing, $150 on design, $200 on promotion and $50 on software. That leaves $350 before tax and the value of the writer’s time, assuming no other expenses. Calling the project a “$1,000 success” would conceal most of the information needed to decide whether to repeat it.

For your own project, keep a simple record of receipts, expenses and hours. Separate a tool’s usefulness from its commercial return: it might help you finish a difficult scene without increasing sales, or reduce a production expense without improving the prose. Either can matter, provided you know which benefit you are buying. Set a spending limit you can afford before treating projected sales as money available to spend.

What the evidence cannot establish

The individual cases largely concern established commercial authors. They do not provide a dependable earnings forecast for a first-time novelist, and they do not isolate the commercial effect of AI from experience, readership, release strategy or other advantages. A large result in one publishing business is a reason to investigate its circumstances, rather than a budget for your own book.

These sources also leave a practical question unanswered: how do sales change when an author explains AI assistance openly to readers? We should not infer either acceptance or rejection from evidence that was not designed to measure that response. Nor does this selection establish that literary fiction, memoir or other kinds of writing cannot succeed. The scope of the evidence is narrower than the range of stories people want to tell.

What to take into your own work

The most useful next step is a small, finished piece of work that readers can judge. Choose a scene, revise it through to completion and ask readers whether they understand the stakes, believe the characters and want to continue. Keep the decisions that improved it. Our guide, How to write a novel with AI without producing AI slop, offers a disciplined workflow for that process, including continuity and revision.

Keep a contribution record as you work: what you wrote, what a tool generated, what you retained and what you changed. It helps you explain your process honestly and answer later publishing questions. It does not establish a copyright guarantee. The guides Can I call myself an author if I used AI to write my novel? and Can you copyright an AI-assisted novel in 2026? examine those separate questions.

Before paying for a launch, check the intended publishing route’s current requirements against the actual manuscript, artwork and audio. Our Publishing an AI-assisted book: KDP, Lulu, and other options is a starting point; follow its official links and recheck the rules before submitting. Preserve your files and versions using the practices in How to protect your novel while collaborating with AI.

There is room for ambition here, especially for someone with a story they have struggled to bring to the page. Build the financial expectations around the work you can finish, the readers you can reach and the costs you can sustain. Another author’s headline earnings cannot make those decisions for you.

Sources and limits

Sources were checked on November 6, 2026. The article uses independent reporting, an author’s first-person account, a public campaign record and a working paper under review, linked beside the claims they support. The study section was checked against the paper itself, rather than only secondary summaries. PennedCanon has not audited these authors’ accounts or verified their net profits. Publication dates are included because these examples describe particular periods, rather than current annual earning rates.

This is a selected set of cases, not an income survey or an earnings promise. Prepared with AI research and drafting assistance for editorial preparation.