--- title: "AI in Publishing: What story do the numbers tell?" description: "Two industry studies this month have reinforced how I’m thinking about AI in publishing.  The first is coverage from the IPG Autumn Conference in the UK , including new research from Dr. Miriam Johnso..." url: https://www.boxcarmarketing.com/ai-in-publishing-survey-2026/ date: 2026-09-29 modified: 2026-09-29 author: "Monique Sherrett" image: https://www.boxcarmarketing.com/wp-content/uploads/claudio-schwarz-jrixOdR1POA-unsplash-mind-the-gap.jpg categories: ["Marketing Strategy & Tips"] type: post lang: en --- # AI in Publishing: What story do the numbers tell? Two industry studies this month have reinforced how I’m thinking about AI in publishing.  The first is coverage from the **[IPG Autumn Conference in the UK](https://independentpublishersguild.com/IPG/Latest/Blogs/IPG/Posts/Publishing%20insights%20from%20the%20IPG%20Autumn%20Conference.aspx)**, including new research from Dr. Miriam Johnson of Oxford Brookes University and a keynote from Eleanor Drage, author of [*What If We Got AI Right?*](https://www.amazon.ca/What-If-Got-Right-catastrophising-ebook/dp/B0FXNCB4ZM) The second is the new **2026 Book Industry Study Group & BookNet Canada survey on [AI use across the North American book industry](https://booknetcanada.ca/wp-content/uploads/2026/09/AI-use-across-the-North-American-book-industry-2026.pdf)**, released today, September 29. Taken together, they suggest that something has shifted.  We are moving beyond *Can we use AI?* toward harder questions: **Should we use it? Where? Under what conditions? And what happens to trust when nobody is quite sure what anyone else is doing?** These are my initial takeaways. There is more in both sets of research that I want to dig into, including the coverage from *Publishers Weekly* and *Quill & Quire*. ## 1. AI may be changing trust more than workflows The most interesting finding from the IPG coverage was about trust. Dr. Miriam Johnson (Oxford Brookes University) presented the findings of the IPG 2026 AI Member Survey (n=71). Her report outlined how AI is affecting independent publishers in the UK and the impact on trust. She described a growing “mutual suspicion” of publishers wondering whether authors are using AI without disclosure, while authors wonder whether publishers are doing the same with manuscripts, blurbs and other work. The IPG research went so far as to suggest that AI may have damaged trust more than it has changed workflows. [The Bookseller](https://www.thebookseller.com/news/news-top-stories/ai-debate-has-shifted-from-can-we-to-should-we-ipg-autumn-conference-2026-hears?utm_source=chatgpt.com) We’re seeing what that looks like in real time. Haitian-Canadian novelist Thélyson Orélien was removed from the Prix Goncourt longlist after an anonymous account used an AI detector to claim his novel was largely AI-generated. Orélien denies using AI, and other detectors reportedly produced conflicting results. Whatever eventually comes of the broader allegations around the book, the AI accusation alone demonstrates the problem: **an AI-detector result can create enormous consequences without establishing what actually happened.** [The Guardian](https://www.theguardian.com/books/2026/sep/25/thelyson-orelien-goncourt-prize-france?shem=aimgspe%2C&utm_source=chatgpt.com) That matters because the North American data shows just how anxious the industry is. [Read the full report on AI use across the North American book industry 2026.](https://booknetcanada.ca/research/ai-use-across-the-north-american-book-industry-2026/) In 2026, 89% of respondents were concerned about inaccurate AI information and 89% about low-quality or fraudulent AI books flooding retail platforms. Concern about author and creator care, including protection of IP, reached 85%. When everything is unclear except the stigma, you get a pressure cooker. This is why I no longer think of an AI policy primarily as risk management. **It is a statement about the relationship. **On both sides of the Atlantic, the data shows trust-based relationships being formalized into paperwork (corporate AI governance, AI policies, and updates to author contracts). Transparency is more important in an AI era. What can staff expect from you? What can authors expect? What can freelancers do with manuscripts? What happens if an AI detector flags a book? Will you stand behind an author when the evidence is weak? Those questions need answers before there is a crisis. ## 2. Publishers are formalizing AI, but not embracing it The other interesting pattern is that AI use is becoming more structured. In North America, organizational AI use rose from **48% in 2025 to 63% in 2026**, while individual use actually fell from **46% to 38%**. Formal AI policies increased from **31% to 49%**. [BISG/BNC](https://booknetcanada.ca/blog/2026/09/29/results-from-the-ai-use-across-the-north-american-book-industry-survey-2026/) The UK shows something similar. IPG research found **76% of surveyed independent publishers now use AI**, up from 42% three years ago, while nearly three-quarters now have a formal policy. [George Walkley / Outside Context](https://www.georgewalkley.com/IPG-Autumn-Conference-2026/?utm_source=chatgpt.com) This is not an industry going all-in on AI, it’s more like a cautious shift from experimentation to **controlled, organizational use**. And a clearer dividing line is emerging. Kate Parkin of Firefinch Publishing described their approach at the IPG Autumn conference as using AI **“operationally not generatively.”** [George Walkley / Outside Context](https://www.georgewalkley.com/IPG-Autumn-Conference-2026/?utm_source=chatgpt.com) North American behaviour mostly aligns with that. The most common organizational uses are **administrative and operational work (56%)**, **marketing (54%)**, data analysis and reporting (41%), and metadata and title optimization (37%). Rights and licensing management remains at just 4%.  That metadata use case makes particular sense to me. Publishing has valuable backlists and small teams with limited time to refresh hundreds or thousands of title records. There is a reasonable case for using AI to help identify gaps, inconsistencies or opportunities. **Overall, my read is that the numbers tell us that narrow practical use cases are seeing real productivity gains, but there is no decline in the level of skepticism and criticism of AI. ** ## 3. “Human in the loop” is harder than it sounds We use the phrase *human in the loop* far too casually. AI can generate reams of material very quickly. Someone still has to know whether it’s right. **For example**: a distributor described AI misapplying an immigration BISAC code while missing it where it applied (BISG/BNC, page 39).  Sometimes all we’ve done is move the bottleneck. The BISG/BNC 2026 survey includes self-reported positive experiences — 72% of professional AI users said it improved their efficiency or productivity — but also respondents describe the extra time needed to correct hallucinations, bad classifications and poor outputs.  A human reviewer is useful only if they know enough to catch the mistake, which is possibly why we see more adoption among seasoned professionals (use peaks at 15-35 years in the industry) vs. those early or even mid career (BISG/BNC, see the chart on page 14 and the findings on page 16). I’d like to see the breakdown by seniority of perceptions of AI training. **35%** of North American respondents said they have ethical objections to AI training and that it is not relevant to their work, while another **21%** have ethical objections, even though AI training is relevant to or required in their work. **Then we have the industry stance**: 79% favour staying informed vs. 33% for careful adoption. 57% agree the industry shouldn’t get involved at all.  There is acceptance that AI is having an impact on the industry and that it’s important to stay up to date, but AI literacy is not the same thing as supporting AI adoption, nor is literacy enough. AI training develops depth of understanding and hands-on skill. And the North American data shows resistance to AI education and training (BISG/BNC, page 30). **My two cents:** If you’re going to tell authors, staff and contractors what they can and can’t do with these tools, somebody in the organization needs enough practical experience to understand what the tools actually do, where they fail, and what meaningful review looks like. ## What I recommend for a small or medium press ### Do now **a) Create a Yes / No / Ask First list.** Forget the 20-page policy nobody reads. Focus on one page that covers the basics, in plain language. For example: **YES:** audience or market research, first-pass metadata checks, selected internal administrative work. **NO:** unpublished manuscripts in public AI tools, generated cover art, undisclosed editorial rewriting. **ASK FIRST:** author-facing copy, contracts, rights work, anything involving sensitive creator material. Then document the assumptions underneath those decisions. **b) Audit three workflows people are already using.** For each one, identify: **Input → tool/account → output → who checks it → skills required to check it → time actually saved.** If you don’t know whether it improves the work or saves time, **it’s an experiment, not a workflow.** **c) Fix one genuinely tedious process.** For a five-person press, I would start with metadata cleanup, campaign reporting, backlist research or repetitive internal admin. Time the process now. Test a different approach. Time it again. And don’t assume the answer is simply *human did X, now AI does X*. Ask whether the workflow itself should change. A great use case is having AI audit an existing workflow. **d) Tell authors what you’re doing.** Put a short statement in submission guidelines and onboarding materials explaining what you use, what you don’t use, and how manuscript material is protected. If your authors would be surprised by how you use AI, that’s a problem. If you’re surprised by how your authors or freelancers are using it, that’s also a problem. ### Watch **a) AI licensing.** The UK is already developing collective licensing mechanisms. Canadian publishers don’t need to copy them, but they should know which rights they control before somebody comes asking to license them. [Independent Publishers Guild](https://independentpublishersguild.com/IPG/Latest/Blogs/IPG/Posts/Ten-things-we-learned-at-the-2026-Autumn-Conference.aspx?utm_source=chatgpt.com) **b) AI-mediated discovery.** IPG also highlighted the growing importance of publisher-owned metadata as AI becomes part of book discovery. Before chasing the latest GEO tactic, get the underlying title, audience and descriptive data right. [George Walkley / Outside Context](https://www.georgewalkley.com/IPG-Autumn-Conference-2026/?utm_source=chatgpt.com)  This has also been reinforced by recent webinars by Tricia McCraney ([see the TechForum replay](https://www.youtube.com/watch?v=Zs4mDQHb_h8)) and Cam Lennon ([see PW+AI](https://pwlearninglab.com/fall26-session-1/)) ### Wait **a) Big AI transformation projects.** Start with an actual publishing problem.There are more questions than answers here, and many apparently simple use cases become considerably less simple once you look at the content going in, the accuracy coming out and the human work required to check it. Start with small, narrowly defined tasks where AI is a good potential solution. ### The takeaways from these two industry snapshots: 1. **Publishers use of AI varies by organization size and type**. There are lessons that could be shared. I want to dig more into small vs. large orgs adoption. **Small-press data**: 46% of 5-10 person orgs discourage AI, their individual use is the lowest (33%), and 43% are still developing a policy, the highest of any size band. I’d like to better understand this segment. 2. **UK publishers have similar qualms about AI as their North American counterparts**, but my anecdotal experience is that UK and US publishers are actively experimenting and moving forward faster than Canadian publishers. The IPG study had 71 responses from independent publishers and BISG/BNC included 230 publisher respondents, plus 541 from non-publisher roles (Libraries, service providers, literary agencies, distributor/printer/wholesalers). So it will be interesting to see the BISG/BNC data on the breakdown by respondent type, and what is shared specific to Canadians. 3. **AI governance and policies are required for trust-based relationships.** An AI policy should be a working document that is easy to reference, that everyone inside and outside the organization understands. **[Do an AI Governance Sprint](https://docs.google.com/document/d/1qqnVeDAeGaFF_Ds0KbQDunwlVioovjfDANsyDMmga4c/edit?usp=sharing)** → Here is a free Boxcar Marketing resource to help marketing teams verbalize what is acceptable and unacceptable AI use.  4. **AI training is not valued as much as it should be.** I’d love to see fewer vague experiments and more concrete how-to examples of where AI earns a place in the workflow. 55% of AI-using orgs don’t know where they’ll use it next, and one respondent said AI left them “differently busy” (BISG/BNC, page 39). We can do better.