ɫ��ֱ��

In March 2026, Hachette pulled Mia Ballard’s Shy Girl after the author was accused of using GenAI to write the novel. The incident escalated the debate around AI adoption within publishing, sparked by OpenAI’s launch of ChatGPT in 2022.

I have taught students about digital publishing developments for over a decade. Tools that automatically generate text and images are not new, but OpenAI and its competitors have made them more accessible. My first classroom experiments with ChatGPT’s precursors asked students to create Twitterbots or generate abstract geometric book covers in homage to reissues of .

• Reaching for the pizza? How to help students make informed choices with GenAI
• GenAI has destroyed grading – and it’s made me a better instructor
• An insider’s guide to how students use GenAI tools

When I first introduced these tools, the students mostly saw them as gimmicks. This is no longer the case. During admissions interviews for my MA Publishing course, I ask applicants: what is the most pressing challenge facing UK publishers? The answer is overwhelmingly GenAI.

This response is understandable. Media coverage and industry hype present AI as a radical disruption to how we work and learn. This discourse can be overwhelming – and contradictory. Individual responses to GenAI in publishing vary, but there are three clear trends I have noticed in conversations with publishing students:

1. Concerns around what constitutes acceptable use of GenAI 

The Society of Authors launched a “” label in March 2026. There is nuance to the label, but as Ballard’s case demonstrates, publishers and educators could be viewing GenAI use as a binary rather than a spectrum. Most students and publishing professionals are unlikely to be either GenAI maximalists or complete boycotters. Negotiate acceptable use – both individuals and organisations can have reasons for doing so.

2. The impact of GenAI on the job market, especially on entry-level roles

Publishing courses are vocational. Students sign up with the assumption that it will prepare them for entry-level jobs in the industry. AI companies’ long-promised productivity gains are likely to affect these jobs if tasks can be automated. Students have an understandable fear of obsolescence and uncertainty around what the workplace will look like after graduation. Ensure there is sufficient space in the curriculum to discuss these concerns and reassure students of the value of human labour within the industry.

3. Navigating personal and professional ethics 

Castigating perceived misuse of GenAI can lead to a culture of fear. Australian students’ emotional responses to GenAI depend on group dynamics, according to a . Individuals have their own position on issues of sustainability, transparency and equity with GenAI that should be respected. Exploring how these underlying emotions and personal values align with expectations in corporate cultures is invaluable to prepare students to enter the industry.

Integrating practice with criticism

The best way to help students navigate these complex issues is domain-specific exploration of the tools. Combine hands-on experience with opportunities for critical reflection to help students see both opportunities and limitations for GenAI. I have found two specific tasks to be the most productive.

Book cover design

Students started experimenting with GenAI to generate book covers as early as 2021. These early experiments revealed all sorts of problems: garbled text, a three-dimensional render of the cover, imagery reflecting a superficial understanding of the book’s title.

The tools have improved but the fundamental issues remain unresolved. Book covers are multimodal, featuring sophisticated interplay between images, text and layout conventions. Students discover this when they are allowed to use GenAI in a task to create a book cover. 

It is difficult to “one shot” a book cover using these tools. The students who want to continue using GenAI after encountering this initial friction use it more diligently to generate minor parts of their visuals, such as a publisher’s logo. 

The task also encourages students who are less confident in their design skills to consider how to present briefs in a concise and effective manner to generate acceptable images.

Historic examples of text generation

Asking students to critically annotate an AI-generated text is a well-rehearsed trope across disciplines. The thorny issue of detection and false positives makes it difficult to rely on current systems. For example, students may feel uncomfortable using em-dashes if they are told it is an indicator of AI-generated text.

I have found analysing pre-ChatGPT computer-generated text more productive than using contemporary GenAI. Lilian-Yvonne Bertram and Nick Montfort’s is a rich anthology of early text-generation experiments that are often weirder and more creative than the generic output of contemporary services. The generation mechanisms are often simpler and easier to explain. This provides foundational knowledge for students to reassess current developments.

Creativity through friction

These two tasks rely on friction between GenAI outputs and human expectations to prompt students to reconsider the role of human creativity in publishing workflows. Students discover that these powerful tools cannot replace humans with a few clicks of a button – and that’s reassuring. There is still value in human creativity and decision-making within publishing.

In May 2026, a further debate broke out around allegations that up to three 2026 Commonwealth Short Story Prize regional winners may have used AI to generate their stories. Granta, which published but did not select the stories, but elected not to retract the stories “until definite evidence comes to light”. These controversies show that the debate around acceptable uses of GenAI is unlikely to diminish soon. Publishing graduates need to be equipped to critically and confidently engage with these debates rather than shy away from them.

Simon Rowberry is associate professor in publishing at UCL.