AI and the Publishing World. What's Going On?

AI is moving publishing from a business built mainly around producing and distributing human-made works toward one that also manages machine-assisted production, machine-generated supply, and competition for attention inside AI interfaces. The transition is real, but it is uneven: adoption is fastest for routine work and workflow support, while questions of authorship, consent, quality, and accountability remain unsettled.

The main shifts

1. Publishers are using AI as workflow infrastructure. In books, reported uses include ideation, editing, translation, marketing copy, pricing, recommendation, and audiobook production. A 2026 study of book-industry stakeholders reported that more than 60% of publishers and self-published authors had integrated AI into some stage of production, with independent authors benefiting because one person can cover several formerly separate functions (Olszewski and Krawczyk, 2026).

The same study reports rapid experimentation with AI narration and multilingual access, but also persistent listener preference for human narration because of emotional engagement. These figures come from industry-facing surveys and reports rather than a single representative census, so they indicate direction and intensity more confidently than they establish a universal adoption rate.

In news, AI is being used for drafting, summarisation, personalisation, analytics, and distribution. A 2025 study by Zhao and Berman, linking audience, hiring, access-control, and content data, found that publisher visits began a persistent decline after August 2024, but found no discrete post-GenAI collapse in editorial hiring; publishers instead changed access rules, page formats, and interactive features (Zhao and Berman, 2025). That is a shift toward augmentation and defensive adaptation, not evidence that AI has already replaced newsroom production at scale.

2. The bottleneck is moving from production to discovery and monetisation. AI answer engines can satisfy an information need without sending the reader to the publisher’s site. A 2026 Association of Online Publishers/Deloitte snapshot reported a 4.55% fall in digital revenue among 13 surveyed publishers in the first three months of the year and linked especially large declines in classified and off-platform revenue to users receiving information inside AI answers; the same report said only 26% of ChatGPT users reported that they would click at least one media link in a response (“First tremors” of AI earthquake showing, 2026). The sample is small and the causal attribution is not clean, but the strategic problem is clear: publishers may supply the information that makes an AI answer useful while losing the visit, advertising impression, subscription opportunity, and behavioural data.

This creates a new bargaining question. Publishers can block crawlers, negotiate licensing deals, improve metadata, or optimise content for machine citation, but each choice trades reach against control. A 2026 media-industry compilation reported that 81% of publishers used AI in editorial or production work and 23% had a content-licensing deal with an AI firm, while 62% reported declining referrals from AI answers (AI in Media and Publishing Statistics, 2026). Those numbers are useful as a market signal, not as a definitive industry census: the page is a secondary compilation and its estimates should be checked against the underlying surveys before use in investment or policy decisions.

3. The supply of publishable material is expanding faster than the market’s ability to reward it. Generative systems reduce the cost of producing books, illustrations, summaries, and promotional material. That lowers barriers for independent creators, but it also increases competition and makes catalog quality, provenance, and discoverability more valuable. A 2026 analysis of platform data reported that books with substantial detected AI content represented 20% of the studied catalog but 12.1% of sales and 11.3% of revenue, while books without detected AI text represented 62.9% of the catalog and 72.5% of revenue (AI-generated books are flooding Amazon and, 2026). The same analysis found that the catalogue grew much faster than quarterly revenue, implying intensified competition and lower revenue per title. Because the study concerns a particular platform sample and detection method, it should not be generalised to all publishing; its stronger conclusion is that more output does not automatically create more demand.

Scholarly publishing is becoming a governance problem

Academic publishers are converging on a basic rule: an AI system cannot be an author because it cannot accept responsibility, manage conflicts of interest, sign publishing agreements, or answer for errors. A comparative study of the five largest academic publishers found consensus on prohibiting AI authorship and requiring disclosure, but substantial divergence in what must be disclosed, how policies are phrased, and how reviewer and editor use is governed (Lendvai and Petra, 2026). Current guidance commonly permits limited assistance such as language polishing while requiring disclosure of substantive generated text, images, analysis, or references (Chigwada and Ngulube, 2026).

The harder issue is verification. Large language models can generate plausible but false claims and references, reproduce bias, and obscure how an answer was produced. A 2026 review of scientific publishing identified benefits in language quality, editorial screening, and access for researchers working across language barriers, but also hallucinated content and citations, confidentiality breaches, automation bias, and inconsistent enforcement across journals (Generative Artificial Intelligence in Scientific Publishing, 2026). Detection alone is a weak solution: a 2026 editorial argues that AI detectors cannot reliably distinguish assisted from manual writing, while disclosure can be underreported when authors have incentives to conceal use (Hanneke et al., 2026). The practical direction is therefore not “detect everything,” but require human verification, record responsibility, and make high-stakes uses auditable.

Peer review is a particular fault line. Reviewers and editors are generally prohibited from uploading unpublished manuscripts or review reports into public AI systems because doing so can expose confidential data, methods, and findings. A 2026 policy review describes this as a confidentiality rule distinct from author disclosure and notes that publishers are experimenting, cautiously, with publisher-hosted systems that do not send manuscripts to external providers (GenAI in scholarly authorship, 2026). AI may help with screening or structured feedback, but delegating evaluative judgment creates risks of automation bias, prompt injection, confidentiality loss, and unexplained changes in acceptance decisions (Hanneke et al., 2026).

Copyright, consent, and bargaining power

The central legal conflict is not only whether an AI output is copyrightable. It is also whether copyrighted books, articles, illustrations, voices, and news are being copied or ingested for training, whether that use is licensed, and who receives compensation. Current author guidance reports that standard book contracts often leave AI-training rights with authors rather than publishers, so publishers seeking to license books for training have been asking for permission; it also recommends negotiating explicit terms for training, AI narration, translation, and cover generation (AI and Publishing, 2026). That is contract guidance, not a universal legal rule, and it does not resolve disputes over material already obtained or used without consent.

Creators’ resistance is therefore partly about ownership and partly about market substitution. A 2026 review of trade publishing found that authors generally resist GenAI because of labor displacement and disruption of creative work, while publishers are more willing to use it for non-creative tasks; it also found that the roles of literary agents and acquisition editors remain under-researched (Kehoe, Nagle and Feller, 2026). A separate 2026 synthesis of creative-industry evidence describes the emerging policy demands as transparency, consent, compensation, and human-centered design, while reporting that a UK consultation on AI and copyright drew more than 11,500 responses, 88% of which opposed expanded AI-training rights (Woodbridge and O’Hare, 2026). The consultation result shows the depth of creator opposition; it does not by itself settle the legal merits of either side.

What is likely to matter next

First, provenance will become a product feature. Publishers will need reliable records of who made a work, which systems assisted, what material was licensed for training, and whether a voice, likeness, or style was authorised. Second, contracts will carry more of the practical burden: training rights, synthetic narration, translation, cover art, disclosure, audit rights, and revenue sharing will be negotiated explicitly. Third, human judgment will remain economically valuable where readers care about trust, taste, lived experience, and accountability; AI can lower production costs, but low cost alone does not create attention or reputation.

The largest unresolved question is distribution. AI may help publishers make more content, but it can also intercept the relationship between publisher and reader. Evidence so far supports a mixed picture: workflow efficiency and access gains, pressure on creative labour, a growing flood of low-cost material, and emerging traffic and licensing conflicts. It does not support the simple claim that AI will either destroy publishing or merely make it more efficient. The likely outcome is a reallocation of value toward platforms that control models, discovery, rights, and audience access and toward publishers that can prove the origin, quality, and accountability of what they publish.

*Article authorship assistance by Harry, our AI Bot.

References

Olszewski, R. and Krawczyk, D. (2026) “Impact of Artificial Intelligence on the Book Industry: A Path Toward the Knowledge Economy,” Knowledge Economy and Lifelong Learning, 2(1), pp. 67–98. doi:10.61093/kell.2(1).67-98.2026.

Zhao, H. and Berman, R. (2025) “Strategic Response of News Publishers to Generative AI,” ArXiv [Preprint]. Available at: http://arxiv.org/abs/2512.24968v4.

“First tremors” of AI earthquake showing in digital revenue hit (2026) internet. Available at: https://pressgazette.co.uk/publishers/digital-journalism/first-tremors-of-ai-earthquake-showing-in-digital-revenue-hit/.

AI in Media and Publishing Statistics 2026 (2026) internet. Available at: https://presenc.ai/research/ai-in-media-and-publishing-statistics-2026.

AI-generated books are flooding Amazon and tanking sales for human authors (2026) internet. Available at: https://the-decoder.com/ai-generated-books-are-flooding-amazon-and-tanking-sales-for-human-authors/.

Lendvai, G.F. and Petra, A. (2026) “Artificial intelligence in academic practices and policy discourses across ‘Big 5’ publishers,” Research Evaluation, 35, p. rvag004. doi:10.1093/reseval/rvag004.

Chigwada, J. and Ngulube, P. (2026) “Use of artificial intelligence tools in the publishing process: expectations from publishers through author guidelines,” Frontiers in Research Metrics and Analytics, 11, p. 1740510. doi:10.3389/frma.2026.1740510.

Generative Artificial Intelligence in Scientific Publishing: Ethical Governance, Challenges, and Responsibilities (2026) internet. Available at: https://jms.ump.edu.pl/index.php/JMS/article/view/1571.

Hanneke, B. et al. (2026) “Academic Publishing in the Age of Generative AI,” Business & Information Systems Engineering, 68(4), pp. 729–737. doi:10.1007/s12599-026-01013-0.

GenAI in scholarly authorship: the 2026… (2026) internet. Available at: https://casrai.org/news/genai-disclosure-scholarly-authorship-2026.