- In the literary world, the question of using AI is being raised more and more.
- Until now, its use has sometimes been highlighted, sometimes downplayed, or even hidden, and publishing houses have started to put in place labels – before the adoption of the AI Act on August 2nd by the European Union.
- This information “around” the book changes the game and plays a role in its material and symbolic economy.
In many companies, artificial intelligence (AI) has become an unsung collaborator. In France, 18% of companies with 10 or more employees reported using at least one AI technology in 2025, compared to only 6% two years earlier .
Are publishing houses an exception to this phenomenon? The book industry is certainly not immune to this trend. The survey conducted by the French Publishers Association (Syndicat national de l’édition) on the use of AI and the Canadian report BookNet 2025 reveal that book professionals primarily use AI for administrative tasks such as drafting emails, summarizing documents, and producing abstracts.
Solutions like Insight, developed by Veristage and presented as an AI “that works like publishers ,” allow for more in-depth editorial work by enriching metadata or preparing sales pitches. This gradual integration of chatbots is also highlighted by the recent report from the French National Assembly on the effects of artificial intelligence on creative work, which emphasizes the increasing spread of these tools across all cultural sectors.
These different uses of AI behind the scenes in publishing are rarely communicated to the public. However, when this tool is involved in the creative process itself, the reactions are quite different. Authors using AI are strongly encouraged to disclose these uses to inform readers. AI thus seems acceptable when it remains a tool for editorial production, but becomes a subject of justification as soon as it touches upon literary creation.
For editorial transparency?
It is in this context that transparency mechanisms have developed in the publishing industry, with diverse and sometimes nuanced positions.
Since 2024, several initiatives have sought to make the production conditions of works visible. Some rely on a logic of AI-free certification, attesting that a book was created without substantial recourse to a chatbot. Several labeling systems have thus emerged in France, such as the “Human Creation” label, the “Guaranteed without Artificial Intelligence” logo , and the terms “Human-Made” and ” Handmade, “ used particularly in the world of comics.
In the United States, the ” Human Authored “ certification created by the Authors Guild emerged from the same dynamic. Others, on the contrary, adopt a logic of signaling the use of AI, explicitly mentioning its presence in the editorial or creative process, for example by means of a banner on the cover or a mention on the back cover.
Raphaël Doan’s alternate history novel, * If Rome Hadn’t Fallen* , makes the use of AI a visible element of its publishing strategy. A banner has been added by the publisher, Passés Composés, stating: “The first history book written and illustrated with artificial intelligence ,” and the use of AI is also mentioned on the back cover . This editorial choice is not merely a gesture of transparency. Presenting the book as a “first” also makes it a selling point, with AI becoming both a vehicle for media visibility and information intended for the reader.
Conversely, the novel Tokyo Sympathy Tower by Japanese author Rie Kudan, who has acknowledged using AI to write this novel , makes no mention of AI in its French translation published by Éditions Denoël, nor on their website .
In self-publishing, two opposing logics coexist. Librinova displays a label visible in their publication as well as on their website since this structure is the initiative of the “Human Creation” label, while Amazon on Kindle Direct Publishing (KDP) imposes a mandatory declaration since 2023, without this information necessarily being visible to the reader at the time of purchase.
This diversity of practices shows that the issue is not solely about the presence or absence of information, but also about how it is made visible. Information is not simply added to the book; it is staged, sometimes highlighted, or conversely, kept at a distance. Not all publishing professionals choose the same medium or the same degree of exposure. Therefore, there is no single model of transparency today, but rather several editorial strategies for visibility.
This diversity is all the more interesting given that research shows that the mere presence of a mention of AI alters readers’ judgments of authenticity, creativity, and sometimes even their purchase intent . In other words, the meaning of a work is not constructed solely within the text. The information surrounding it also contributes to its interpretation and evaluation.
This quest for transparency extends beyond the publishing sector alone. It is part of a broader movement towards the standardization of AI usage.
Since August 2, 2026, the transparency obligations stipulated in Article 50 of the European AI Regulation require publishers to label content generated or manipulated by a robot in order to limit the risk of deception and strengthen user trust. While these provisions do not specifically target the book industry and do not mandate labeling of books, publishing initiatives are also part of this trend toward transparency for users. It is, however, legitimate to wonder whether literature created in whole or in part by AI might attract a new readership, just as certain video games produced by robots have captivated a segment of the public.
Towards an amended paratext
This evolution echoes the words of Gérard Genette in Seuils , published as early as 1987. He recalled that ” the ways and means of paratext are constantly changing according to the times, cultures, genres, authors, works, editions of the same work, with sometimes considerable differences in pressure “.
For Genette, a book never begins with the first sentence. It begins with its title, its author’s name, its cover, its series, its preface, or its back cover—all elements that already guide interpretation before the reader even enters the text. What if AI were precisely in the process of updating the categories of information surrounding the text?
Labels, banners, back cover blurbs, and author statements would then no longer be merely transparency devices. They could constitute the first manifestations of an amended paratext. The question would therefore no longer be simply whether to indicate the use of AI, but where this information should be placed in the book, and how it transforms the reading experience.
This hypothesis opens up a largely unexplored area of research. While several studies now analyze the effects of labels on the perception of works, very few examine their place within the material and symbolic economy of the book. Where do publishers choose to make AI visible? What mechanisms do they favor? And above all, what functions do these new labels fulfill? Do they serve only to inform, or do they also contribute to rebuilding reader trust in a context where the conditions of production of works are becoming increasingly difficult to identify ?
From this perspective, it might be beneficial to specify the identity of the AI used. The same requirement should also apply to proofreaders or graphic designers, who have never been named to date, unlike translators. The issue therefore goes beyond mere transparency. By making the use of AI visible (or, conversely, invisible), publishers are not simply adding information to the book. They are contributing to redefining the mediations surrounding the work and, perhaps, to the emergence of a new category of paratext.
Author Bio: Stéphanie Parmentier is a Lecturer at Aix-Marseille University (amU), PhD in French Literature and teacher-librarian at Aix-Marseille University (AMU)
