Survey reveals public skepticism on AI-written literature

A survey conducted by Ipsos bva – CESI engineering school between October 9 and October 12, 2026, among 1,000 French people aged 18 to 75 years, revealed that 69 percent of respondents believe that a text written by artificial intelligence is not genuine literature. The survey, released on October 28, 2026, comes amid significant debate over whether a book written with AI should still be considered literature.

Public insists on clear labeling of AI use

According to the survey, 92 percent of respondents believe that any use of artificial intelligence in the writing of a book should be clearly indicated on its cover. These results were revealed just days before the Goncourt Prize, following the Thélyson Orélien affair, which has raised questions among publishers and literary juries.

Controversy over a prize-winning novel

The controversy began with the novel C’était ça ou mourir, published by Grasset on August 19, which was awarded the Fnac Novel Prize on September 21. The book quickly found itself at the center of plagiarism accusations and suspicions concerning the use of artificial intelligence. On September 25, the Goncourt Academy removed the novel from its selection, while the Femina subsequently excluded it, and the Renaudot decided to keep it.

Publisher ends collaboration over AI concerns

The controversy took on another dimension on October 21 when Boréal editions, a Quebec-based publisher, announced that they would end their collaboration with Thélyson Orélien after an external analysis revealed traces associated with artificial intelligence in the metadata of certain documents. The publisher considers the involvement of this technology to be established, but the elements made public do not allow the precise determination of its extent.

Public opinion on excluding AI-suspected books

The survey also found that 46 percent of respondents believe that a book merely suspected of having been written with AI should be removed from literary prize selections. This position raises a difficulty: from what elements can a jury exclude a novel, and how can it be avoided that an insufficiently substantiated suspicion becomes a public condemnation?

Majority accepts AI-assisted literature with human input

The survey highlights that a majority of respondents, 55 percent, acknowledge that a book can be considered literature if "the human remains the main author." This distinction is worth emphasizing, as it separates the use of a tool during the creative process from the delegation of the writing itself. Requesting a translation, verifying information, or correcting a formulation does not necessarily imply the same work as asking a software to produce several chapters.

Public divided on AI-generated literature

The question of establishing the boundary between these practices is complicated, especially since the transparency so widely requested does not suffice to convince readers of the literary value of a production. Only 40 percent of respondents adhere to the idea that a book created with AI can be considered literature if this use is declared. Nearly half, 49 percent, express their disagreement. As for considering these texts as "a new form of literature in its own right," only 32 percent of participants are favorable.

Public disappointment if AI wrote a beloved novel

The study also presents a revealing situation: discovering that a particularly appreciated novel was entirely written by an artificial intelligence. In this case, 62 percent of those surveyed would feel disappointed or deceived, while 27 percent state that it would not change their judgment. Therefore, the text would not always be appreciated independently of its origin. For part of the public, the relationship with the author, their experience, and their way of storytelling remains inseparable from the pleasure of reading.

Public prefers clear labels over detection tools

Among the proposed measures, 66 percent of respondents prefer the mandatory inscription "written with AI" on the affected works, ahead of an editorial charter or the development of independent detection devices. However, it would be necessary to define what this indication would cover. An automated correction, a document search, or the generation of several hundred pages do not constitute equivalent interventions.

Publishers to use AI detection software

The Madrigall group, including Gallimard and several houses in the group, intends to strengthen its vigilance. According to information reported by ActuaLitté on October 24, these publishers will submit manuscripts to a detection software starting in January 2027. The device must notably concern Flammarion, Denoël, P.O.L, and Minuit. The methods of processing the results remain to be specified, as these software do not provide certainties.

Limitations of current AI detection tools

The example of Je, by Lilia Hassaine, published by Gallimard, illustrates these limitations. Submitted by ActuaLitté to the Pangram detector, the novel obtained a score of 21 percent of content supposedly generated by AI. The author then submitted a result of 10 percent, according to her, from analyses conducted by Gallimard and by herself. None of these percentages demonstrate that part of the book was artificially produced. Therefore, caution is necessary, especially when such evaluations risk compromising a publication or damaging an author's reputation.

Professionals call for transparency across book production

The professionals also demand more transparency. The writing of novels is only part of the problem. Artificial intelligence now intervenes at various stages of book production, from illustration to translation, sometimes without the reader being informed. Literary translators have been reflecting on these practices for several years. As early as March 2023, the Association of French Literary Translators (ATLF) and ATLAS called for greater transparency regarding automated systems used by publishing houses.

Concerns over rights and working conditions

The issue goes beyond informing the public: it concerns working conditions, pay, and recognition of professionals when their activity risks being reduced to revising a text produced by a machine. Another area of mobilization is the use of protected works to train models. In May 2026, 81 organizations in the cultural, creative, and press sectors asked lawmakers to examine a bill adopted by the Senate aimed at facilitating the defense of rights holders against AI providers. These organizations advocate for prior authorization, transparency of uses, and negotiated compensation.

Survey findings align with international trends

The distrust expressed in the survey is not absolute. The French results align with those of a recent American survey relayed by ActuaLitté. In the United States, 85 percent of the 5,025 book buyers surveyed wanted to be informed when artificial intelligence had participated in the writing of a book. The methodologies and populations studied are different, so the proportions cannot be directly compared. They nevertheless reflect a concern that goes beyond the French market.

Previous cases show growing scrutiny of AI in publishing

There are no shortage of previous editorial precedents. In March 2026, Hachette Book Group had abandoned the publication of Shy Girl, by Mia Ballard, in the United States and interrupted its sale in the United Kingdom, after suspicions regarding the use of AI in this horror novel. The author contested having used it herself to write it, mentioning the involvement of a person tasked with working on a previous version.

Public demand for clarity before major literary awards

In France, the Thélyson Orélien affair has now given this question new visibility. Literary juries have been forced to take a stance, publishers are beginning to strengthen their controls, and the Ipsos bva survey reveals a particularly strong expectation from the public. The mention on the cover would not solve everything. It would require distinguishing technical interventions from the actual creative work and establishing verifications sufficiently reliable to not penalize authors. But just days before the Goncourt Prize award, the realization is clear: if the machine can participate in writing, the readers surveyed want to know how much it was involved. It remains for the professionals to determine how to provide this information without turning every manuscript into an object of suspicion.