AI in Publishing: 2026 Statistics & Primary Sources
A neutrally-framed roundup of every verifiable statistic on AI in publishing — adoption among authors, use inside publishing houses, reader sentiment, rights and litigation, output volume, and economic data. Each stat links to the primary source. Updated quarterly.
contain AI-generated content
Source: NBER w34777 (Reimers & Waldfogel, 2025) ↗of authors want consent and compensation when their work is used to train AI
Source: Authors Guild AI survey, December 2023 (n=2,400+) ↗of authors use generative AI in their work (ChatGPT 85%, Claude 54%, ProWritingAid 50%)
Source: BookBub Partners survey, May 2025 (n=1,229) ↗of book-industry workers individually use AI / of organizations use AI
Source: BookNet Canada × BISG North American survey, April 2026 (n=559) ↗growth in self-published titles in 2025 — 3.5M new ISBNs (4M+ titles total industry-wide)
Source: Bowker via Publishers Weekly, 2025 ↗of AI-using authors do not disclose their use of AI to readers
Source: BookBub Partners survey, May 2025 (n=1,229) ↗About this dataset
Most published statistics on AI in publishing come from one of three places: industry-association surveys (Authors Guild, BISG, ALLi), academic working papers (NBER), or commercial aggregators that recycle older numbers without dating them. This page consolidates the first two and avoids the third.
Every statistic below links to its primary source. Where a stat is older than 18 months, we mark it [Historical] and note context. Where a sample is small or self-selected, we say so. Where a finding cuts against the prevailing narrative, we keep it in.
If you cite something from this page, please cite the primary source first. We're an aggregator; we don't want to displace the surveys themselves.
2. AI adoption inside publishing houses
The most authoritative dataset on industry-side AI use is the BookNet Canada × BISG North American survey (April 2026, n=559) — the first joint trade-body survey covering individual workers and organizations together. Stanford HAI's 2026 AI Index provides cross-industry context.
of individual book-industry workers report using AI in their job
of book-industry organizations report using AI somewhere in the business
of book-industry organizations explicitly do not use AI
of book-industry organizations have a formal written AI policy
of individual AI use is for administrative / marketing / data tasks
of organizational AI use is for administrative / marketing / data tasks
of book-industry respondents use AI for translation (one of the lowest-adoption use cases)
of organizations across all industries reported using AI in at least one business function in 2024
Cross-industry benchmark — publishing's 48% organizational adoption is roughly half the cross-industry average.
base salary for a Penguin Random House ML Platform Engineer (job posting closed Aug 2025)
Methodology note: the BookNet Canada × BISG survey was fielded in summer 2025 and reported in April 2026. The full breakdown by department (editorial, design, marketing) is not in the publicly released excerpt — the full report is paywalled through BookNet Canada.
3. Reader sentiment toward AI in books
Reader-side data is the thinnest part of this dataset — there is no industry-association reader survey on AI in publishing as of May 2026. The Jill Hamilton independent survey (n=285, January 2024) is the only public reader data on AI book covers. Sentiment events (the *Shy Girl* / Hachette cancellation, the Clarkesworld AI-submission flood) are included as qualitative anchors.
of surveyed readers say they would not buy a book with an AI-generated cover; 8.4% maybe; 5.6% yes
Small sample, self-selected via genre Facebook groups, subreddits, and Bluesky bookish communities. Treat as directional, not representative.
of readers in the same survey said they cannot reliably tell an AI cover from a human-illustrated cover
Hachette (Wildfire imprint) pulled Mia Ballard's *Shy Girl* in March 2026 after the New York Times presented evidence of AI-generated text. Ballard denies, attributes the patterns to a hired editor.
Clarkesworld magazine (Feb 2023) suspended new short-fiction submissions after a flood of AI-generated submissions; editor Neil Clarke reported a ratio of roughly 700 legitimate to 500 AI submissions per day at peak [Historical]
Clarkesworld 2025 submission volume — the magazine reopened with manual screening rather than automated AI detection. Neil Clarke (Aug 2025): "the US is now the primary producer of slop submissions."
(1) takes work from human artists; (2) covers all look the same / oversaturated style; (3) general dislike of the AI aesthetic
This section is the weakest in this dataset — and that weakness IS the finding. A rigorous, large-sample reader survey on AI in publishing has not yet been published by any major trade body or research institution as of May 2026. Reader sentiment is currently inferred from purchase patterns, social-media reactions, and small independent surveys.
4. Rights, consent, and litigation
Author-side opinion on AI training is overwhelmingly consistent: consent and compensation, not blanket opposition. The Authors Guild and Society of Authors UK surveys both find supermajorities favoring consent-based licensing rather than absolute bans. The legal landscape resolved partially in 2025 with the Bartz v. Anthropic settlement and continues in active discovery for the OpenAI cases.
of surveyed authors believe consent should be required before their work is used to train AI
Authors Guild called this the highest agreement on any question in the Guild's 111-year survey history.
of authors want compensation when their work is used to train AI models (separate question)
of authors support collective licensing as a path forward (vs. individual-only opt-in)
of authors would not license their work for AI training if they couldn't restrict the prompts (e.g., not for fan-fiction or imitation)
of book-industry respondents cite copyright/training-data controls as their #1 AI concern
of book-industry respondents report at least one substantive AI concern (only 2% report none)
Bartz v. Anthropic class-action settlement (Aug 2025) — largest US copyright settlement on record. ~500,000 works; ~$3,000 per work. Training on legally-acquired copies ruled fair use; training on pirated copies not.
Authors Guild v. OpenAI — motion to dismiss denied (Oct 28, 2025); settlement conference held March 13, 2026; case unresolved as of May 2026
NYT v. OpenAI — active discovery; data-preservation dispute and appellate proceedings ongoing May 2026
Silverman et al. v. OpenAI — consolidated into SDNY MDL 3143; fact discovery cutoff Feb 27, 2026
Disney/Universal v. Midjourney — filed June 11, 2025; Midjourney fair-use defense August 2025; in discovery as of May 2026
Authors Guild released new model contract clauses including (1) no consumer-facing AI upload without written permission with mandatory opt-out of training, and (2) no substantive AI editing (basic spell/grammar permitted)
EU AI Act Article 53(1) took effect: providers of general-purpose AI models must publish a sufficiently detailed summary of their training data and respect the EU Copyright Directive's text-and-data-mining opt-out rules. Full enforcement begins August 2026 (pre-existing models have until August 2027). Fines up to €15M or 3% of global annual revenue.
Penguin Random House became the first Big-5 publisher to add an explicit no-AI-training clause to the copyright pages of its books
HarperCollins (late 2024) became the first Big-5 to license backlist for AI training (to Microsoft) — opt-in only, with ≤200 consecutive words and/or ≤5% of text reproduction guardrails
Two-thirds of the Big 5 have now staked out opposing public positions on AI training (PRH against, HarperCollins licensing). Hachette, Simon & Schuster, and Macmillan have not published equivalent public policies as of May 2026.
5. Output volume and quality
The two most-cited academic findings on AI's volume and quality effects in publishing come from NBER working papers w34777 (Reimers & Waldfogel, on creative-product volume and quality) and w34255 (on ChatGPT use patterns). The Bowker / Publishers Weekly title-count data provides the industry-wide denominator.
"The diffusion of LLMs from 2022 to 2025 tripled new book releases" (paper uses "book releases", a broader category than ISBN-only filings)
"AI-containing books have lower quality, and their rising share — topping half of 2025 releases — drives the overall decline" in average book quality (measured by usage)
"AI books raised consumer surplus by seven percent in 2025" — the same paper finds an overall consumer-welfare gain despite the quality drop, because the supply curve shifted
This is the counterintuitive finding from the same paper as the quality-decline result. Both findings are in the paper's abstract.
total new ISBN'd books in the US in 2025, up 32.5% year-over-year
of those new ISBNs were self-published in 2025, up 38.7% year-over-year
traditionally published titles in 2025, up 6.6% year-over-year
Spines, a venture-backed AI-publishing startup (Series A $16M), aimed to release 8,000 titles in 2025 — up from ~2,000 in 2024 and 400 in 2023. Charges authors $1,200–$5,000 per book for AI-driven editing, cover design, and distribution across 100+ channels. The Society of Authors urged caution; independent publisher Canongate publicly condemned the model.
had adopted ChatGPT by July 2025; non-work usage rose from 53% to 70%+ between 2022 and 2025
of ChatGPT conversations fall into three categories: Practical Guidance, Seeking Information, and Writing — "writing dominates work-related tasks"
Citation hygiene: a previous version of [/blog/ai-book-publishing-complete-guide](/blog/ai-book-publishing-complete-guide) on this site framed the NBER w34777 finding as a "threefold increase in ISBN issuance." That was incorrect — the paper says "book releases," a broader term that includes platform releases without traditional ISBN filings. Corrected May 2026.
6. Tool-specific adoption among authors
Tool-specific share data is from the BookBub Partners survey (May 2025, n=1,229), which asked AI-using authors which specific tools they use. The sample skews indie-fiction.
of AI-using authors use ChatGPT (OpenAI)
of AI-using authors use Claude (Anthropic)
of AI-using authors use ProWritingAid
Sudowrite, NovelCrafter, Jasper, and other fiction-specific AI tools were each cited by under 25% of AI-using authors, with no single specialist tool reaching majority adoption in the BookBub sample.
7. Economic and labor data
Industry-level revenue, labor, and indie-author income data.
US book-publishing industry market size in 2026 (CAGR 0.7%); ~3,287 businesses; 9.7% average profit margin
median annual indie-author income (ALLi 2025 survey of authors deriving ≥50% of time or income from writing, n=2,000+); +6% year-over-year
median annual traditionally-published-author income reported by ALLi's comparison panel (vs. $13,500 indie median)
indie author monthly income distribution (broader sample than ALLi's qualifying authors)
share of indie author revenue coming from Amazon (2023 → 2024 → 2025) — a meaningful diversification trend
of indie authors now sell direct (Shopify, Payhip, BookFunnel) — and ~50% of authors earning over $10k/mo do
median email-list size for indie authors earning $10k+/mo (vs. far smaller lists for lower-earning authors)
of UK illustrators / translators surveyed already report having lost work to generative AI; 37% / 43% report income decreased
Methodology note: ALLi's income survey qualifies authors who derive ≥50% of time or income from writing (a serious-indie filter), so its medians are higher than broader self-publishing samples like Written Word Media's.
How we curated this
We include a statistic only if it meets all three of these criteria: (1) it comes from a named primary source, not a citation of a citation; (2) the survey size, date, and methodology are disclosed by the source; (3) we can link directly to the source page.
Stats older than 18 months are marked [Historical] and dated explicitly. We keep older stats when they're the only or canonical source on a topic (e.g., the Authors Guild 96% consent finding is from Dec 2023 but remains the most-cited industry survey).
Where samples are small (n<500) or self-selected, we flag it inline rather than excluding the data. We'd rather report a directional finding with caveats than pretend the field has more rigor than it does.
We do NOT include statistics from commercial aggregators (Gitnux, WifiTalents, WorldMetrics, AutoFaceless) that recycle older numbers without dating or attribution. Where Statista has a paywalled stat, we link to the originating survey instead.
Conflict of interest disclosure: ManuscriptReport (the publisher of this page) is a commercial AI book-marketing platform — we have a financial interest in publishing-industry AI adoption. We've tried to mitigate that by leading with stats that cut both directions (74% non-disclosure is a credibility risk for AI tools; 96% want consent is a constraint on AI training). If you spot framing you think is biased, tell us via our contact page and we'll review.
Updates: this page is reviewed quarterly. The changelog at the bottom records every substantive change.
Frequently Asked Questions
What percentage of authors use AI in their work right now?
How many publishers use AI in their workflow?
Are authors paid when their books are used to train AI?
What do readers think about AI-generated books?
Where can I find the latest publishing-industry statistics?
Is this dataset updated regularly?
Can I cite this page?
Why did the previous version of one of your blog posts have a different NBER finding?
Changelog
Initial publish. 16 primary sources, 94 verified data points. Hero stats anchored to NBER w34777 (50%+ of 2025 releases contain AI), Authors Guild (96% want consent, Dec 2023), BookBub (45% adoption, May 2025), BookNet Canada × BISG (46%/48% individual/org), Bowker (3.5M self-pub titles, +38.7%), BookBub (74% don't disclose).
Full sources
All primary sources cited above, organized by type. Where a survey has multiple coverage points, we link to the primary source (the survey publisher), not secondary coverage.
Industry-association surveys
- · Authors Guild AI survey (Dec 2023, n=2,400+) ↗
- · BookBub Partners — How Authors Are Thinking About AI (May 2025, n=1,229) ↗
- · BookNet Canada × BISG — AI Use Across the North American Book Industry (April 2026, n=559) ↗
- · ALLi 2025 Indie Author Income Survey ↗
- · Written Word Media 2025 Indie Author Survey (Dec 2025, n=1,346) ↗
- · Society of Authors UK Creator Survey (Jan 2024, n=787) ↗
Academic working papers and reports
Industry data and trade press
Legal and policy events
- · Authors Guild — Anthropic settlement summary (Bartz v. Anthropic, Aug 2025, $1.5B) ↗
- · Authors Guild v. OpenAI (CourtListener docket) ↗
- · NYT v. OpenAI — OpenAI's public response to data demands ↗
- · Silverman v. OpenAI (CourtListener docket) ↗
- · Disney/Universal v. Midjourney (NPR coverage, June 2025) ↗
- · Authors Guild — Model Contract AI Clauses (April 2026) ↗
- · Penguin Random House copyright-page AI clause (TechCrunch, Oct 2024) ↗
- · HarperCollins AI Licensing Deal (Authors Guild summary) ↗
Use this data in your work
Every statistic above links to its primary source. Cite the primary source first. If you publish a story that draws on this aggregation, we'd love to see it — tell us via our contact page. If your reporting involves the publishing-industry AI shift specifically, ManuscriptReport's commercial reports and tools are referenced openly on this site at /services and /tools.