How much AI is in Spain’s bestselling books?

Artificial intelligence is also making its way into the world of books. EL PAÍS examined the use of AI in a sample of 119 popular books published in 2025 and 2026, analyzing titles that appeared on the bestseller lists at Amazon.es and the Spanish bookstore chain Casa del Libro, as well as the winners of a major literary prize.

All the books that the newspaper found to be largely AI-generated were self-published, many of them in the romance genre. These books have sizable readerships, with between 270 and 2,000 reviews on Amazon.

The most prominent case is Devuélveme mi país (Give Me Back My Country), by Spanish politician Alvise Pérez. In February 2026, it reached No. 1 on Amazon. According to the Pangram detection tool, 97% of the book was generated by AI. In other words, the tool attributes nearly all of the text to an AI system such as ChatGPT or similar software, with only 3% classified as exclusively written by a human. The tool does not indicate who used the AI or how it was used. It cannot determine whether the author asked the system to generate the entire text or provided notes and ideas that were later expanded by the AI.

Shortly after midnight, Pérez responded by email, later posting the message on social media. The statement, purportedly drafted by his own AI assistant, maintained that he had written the book “word for word.”

The other books found to make predominant use of AI are novels, almost all of them romances. According to Pangram, 100% of the text of Mr. Sterling: My Boss. A Contract. And a Heart at Stake, by the author writing under the pseudonym Violeta Dark, was generated by AI. The tool attributes 86% of My Boss Is the Antichrist (Sexy Version) by Estrella Correa to AI, as well as 75% of The Map Killer by Pablo Poveda and 53% of The CEO’s Christmas Bride by the pseudonymous Leila Yoon. EL PAÍS contacted these authors through their websites, email addresses and social media accounts, but no one responded.

For each book, the newspaper analyzed three 1,000-word excerpts, roughly equivalent to 10 pages, using the Pangram detection tool. In addition, it examined the full text of 12 of the books with the highest AI scores, including all those mentioned in this article. The same exercise was then carried out on 56 books published in 2019 and 2020, before today’s AI models existed. None showed any signs of AI generation: all received a score of 0%.

What about the books published in 2025 and 2026? Of the 119 titles analyzed, 35 showed some evidence of AI-generated text (29%), and 13 scored above 50%.

https://www.datawrapper.de/_/VwO3H/

AI found in self-published books, not in works by big-name authors

The use of AI was concentrated in self-published works: 13 of 44, or 30%. Romance novels dominate the category because they also make up a significant share of Amazon’s bestsellers.

Established authors, by contrast, did not test positive, and traditional publishers have safeguards in place, according to industry sources. While no publisher has confirmed using Pangram specifically, they say they do take AI into account. “Our rule is that everything must start with a human and end with a human. And from there we strive for a transparent conversation with authors and readers,” says Miguel Aguilar, an editor at Penguin Random House.

The detector found no traces of AI in books by some of the best-known authors in the Spanish-speaking world. The latest works by Javier Cercas, Antonio Muñoz Molina, Isabel Allende, María Dueñas, and Dolores Redondo all received scores of 0%, as did the winners of the Planeta Prize and the Nadal Prize. The same was true of other bestselling political books, including those by Iván Espinosa de los Monteros and Judge Manuel Marchena.

Even so, nine traditionally published books showed some traces of AI use. They include four nonfiction works, all by well-known authors and all commercially successful, with AI scores ranging from 12% to 34%.

Pangram is not infallible. In short passages, changing a single word is enough to alter its assessment. To test the tool, we carried out two checks. First, we analyzed 56 books published in 2019 and 2020: Pangram did not flag any of the 168 excerpts examined. Second, we generated 29 AI-written texts imitating the style of real books: Pangram detected every one of them.

Many books show signs of AI use only in specific chapters. The technology’s footprint is not evenly distributed throughout a book. Some chapters appear to be almost entirely AI-generated, while others show no trace at all. But although extensive AI use is currently concentrated among self-published authors in particular genres, more limited and controlled use appears to be spreading.

Among traditionally published nonfiction books, none of the 16 titles analyzed from 2019-20 showed any sign of AI involvement. By contrast, seven of the 22 nonfiction books published in 2025-26 did. Some influencers who publish books seem to rely on AI for the most straightforward explanatory sections. In that sense, AI is being used much as Wikipedia or online essay repositories once were: as a source for the most obvious passages.

None of these actions is illegal or a crime. Some publishers have already added AI-related clauses to their contracts, but AI will only become a central issue for the industry if readers decide it matters and choose not to buy books written with AI. The debate over AI authorship has gained momentum in 2026 across several countries, driven largely by improvements in the detection tool Pangram and the remarkable success of a debut French-language novel by the relatively little-known Canadian author Thélyson Orélien. Just this Monday, it emerged that the winner of the Caine Prize for African Writing, worth about $13,000, was likely written with AI. It is difficult to know whether, in a few years, with AI being used for all manner of tasks, readers will have adapted to its presence.

For now, it’s a controversial issue that writers are addressing with great passion. “I have written more than a dozen novels without using AI and I have done quite well, so, obviously, I would never use AI to write,” says Marcos Nieto Pallarés, the author of one of the novels analyzed by this newspaper, which received an AI score of 0%. “I have noticed a drop in sales regarding self-publishing. But seeing a cover on a list, it is hard to know whether it was written with AI, so I don’t know to what extent AI-written novels have flooded Amazon.”

109 books in a year and a half

Amazon requires authors to disclose AI-generated content, but it does not prohibit it. Content is considered AI-generated even if it has subsequently been edited. According to the platform’s content guidelines, “If you used an AI-based tool to create the actual content (whether text, images, or translations), it is considered ‘AI-generated,’ even if you applied substantial edits afterwards.”

“Our content guidelines determine which books may be published for sale, and we take both proactive and reactive measures to prevent, detect and remove content that violates those guidelines, whether generated by artificial intelligence or not,” says an Amazon spokesperson, without giving further details about how they do so.

“I have no proof” that AI is more prevalent in self-published books, says Yanira García Fernández, the author of another novel analyzed by the newspaper that received an AI score of 0%. “Although I do believe there is no kind of control. Traditional publishing filters submissions and Amazon does not do that prior work; anyone can create an account, upload their book and that’s it, which is why it is easier to publish anything because it’s as simple as clicking a button and letting the reader choose,” she adds.

Violeta Dark has published 109 books in five languages on Amazon since March 1, 2025, including at least 21 in Spanish. Mr. Sterling: My Boss. A Contract. And a Heart at Stake was released in August 2025 and is the title that has received the most reviews. It is also the first installment in a series called The Mr. Irresistibles.

As of October 2026, it remains Amazon’s No. 3 bestseller in contemporary fiction. This is how Dark describes herself on her website: “Violeta Dark creates intense, passionate love stories where strong women find a love that defies all odds. Her novels blend romance, suspense, and raw emotion, always guaranteeing that happy ending worth fighting for.”

Readers, too, are increasingly citing AI as a reason for criticism in Amazon reviews. “Not recommended, it starts well but takes turns that make no sense, it seems written by AI. I had read something by this author before and it wasn’t bad,” says a reader of Estrella Correa’s book. Commenting on Alvise Pérez’s book, another Amazon user wrote: “It seems written by an AI. Although his message is direct and sometimes repetitive, there are moments when everything sounds too similar, almost as if it had been written automatically and without many nuances.”

Is this just the beginning?

A scientific paper published in August analyzed a sample of 14,419 self-published ebooks on Amazon. The researchers defined AI use above 25% as substantial and found that one in five books exceeded that threshold. Among the self-published bestsellers analyzed by EL PAÍS in Spain, however, the share is even higher: 39%. AI-generated books are not yet a dominant presence, but it is obvious we are at the start of a new era.

Pangram uses its own method to detect AI-generated text. But there are also telltale patterns that have become informal clues for spotting AI in a piece of writing: negative contrasts of the form “not X or Y, but Z”; lists of three items; parallel structures; frequent use of colons and dashes; aphoristic summaries; and a uniform sentence rhythm.

“Her words cut through me like a knife. Precise. Direct. Relentless,” Violeta Dark writes near the beginning of her novel.

“I am not talking about an idealized past nor a nostalgic postcard, but about a shared inner certainty: you worked, you strived, you did your part,” is the second sentence of Alvise Pérez’s book. “We were a nation in control of its destiny,” is the final sentence of its opening paragraph.

This analysis points to a shift: in 2019 there was no trace of AI in the most popular books, and now signs of it are beginning to appear. It is also important to understand what these systems detect: they identify whether a text appears to have been generated by AI (because it follows certain patterns), but that does not reveal exactly what role a person played in the process. AI can be used for very lazy, almost automatic writing, but it can also be used to turn a person’s own ideas or dictation into written text. There is a lot of gray area between those extremes.

What lies ahead is a debate, though not an entirely new one. The memoir Open by Andre Agassi, for example, was written by journalist J. R. Moehringer after 250 hours of conversations with the tennis star. Who, then, is the author? Only Agassi’s name appears on the cover, even though he may not have written a single word.

Methodology. The test was carried out using a subscription to version 4 of Pangram, one of the detectors most highly rated by the scientific community.

To select the books, we started from the best-seller lists of Amazon.es (top 50 in Books and in Kindle) and Casa del Libro (top 10 for fiction and nonfiction). We did not review them daily: we used one snapshot of each list per month, saved on the Internet Archive, between January 2025 and September 2026. This is therefore a sample of popular books, not a census. To that list, we added winners of 13 literary prizes. From that list, we analyzed all the books we had available (119, approximately half). No book was chosen or discarded because of its author or content. All are adult books, originally written in Spanish and published in 2025 or 2026.

From each book, we randomly selected three 1,000-word excerpts to run through the Pangram software. Then, for 12 books showing signs of AI use, we analyzed the entire book.

For the control group, we analyzed a similar sample of books from 2019 and 2020 using the same procedure, when current AI models did not exist. That first control has a potential objection: perhaps Pangram labels the 2019–2020 books as human not because it can distinguish them, but because it already knows them and they were in the texts used for its training. That cannot be ruled out. But there are two reasons to think it is not the case. The first comes from our analysis: trade-published and prize-winning books from 2025 and 2026 also do not test positive. The second comes from the company itself. Its CEO, Max Spero, explained on the podcast Mixed Signals that they measured their false positives with millions of texts predating 2022 “that we know are not AI-generated, but that are also not in our training set.” According to their numbers, Pangram marks only one in 24,000 human texts as AI.

We also ran the reverse test. We generated texts imitating the style of real authors with four AI models, then analyzed them with Pangram. The software marked them all as AI-generated.

Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition