The 2026 sapphic fiction reader survey (Part 2)

TL;DR

• Sapphic fiction authors are struggling financially. More than half are making less money from their writing than in the past, and one of the causes is the flood of AI-generated books that makes it harder for human-written stories to get noticed.
• Generative AI and assistive AI are not the same thing. When people speak out against AI, they are usually talking about generative AI, which replaces creatives rather than just supporting them.
• Readers are most opposed to generative AI when it comes to writing the book itself and narrating the audiobook. Attitudes become more divided regarding covers and marketing content. Translations sit somewhere in the middle.
• The majority of sapphic fiction readers reject AI-generated books. 87% wouldn’t read an AI-generated book, while only 1% say they don’t mind it.
• Audiobook listeners rejected AI narration at the same rate as AI-generated books (87%), while only 1% wouldn’t mind listening to AI-generated audiobooks.
• 62% of readers rejected AI translations outright, while 3% said they wouldn’t mind them.
• Nearly half of readers (44%) would not read a book with an AI-generated cover, while the other half would consider reading one under certain circumstances. 8% said they don’t mind AI covers.
• Attitudes toward AI-generated promotional images and video were similar to attitudes toward AI covers: 42% of readers said they would not pick up a book from an author who regularly uses AI-generated content to promote their books, while 12% wouldn’t mind.
• Younger readers of sapphic fiction were more likely to reject AI-generated books, audiobooks, translations, covers, and promotional material, while older readers tended to be more accepting of AI-generated content.
• Readers from predominantly English-speaking and non-English-speaking countries were equally likely to give AI-generated books a “hard pass,” but readers from non-English-speaking countries were more likely to reject AI-generated translations.
• Attitudes toward generative AI varied between countries. Readers from France, Brazil, and Canada rejected AI-generated content at higher rates, while readers in Germany, the UK, and the US tended to be somewhat more accepting.
• Most readers are somewhat confident in their ability to recognize AI-generated books and covers, saying they could identify AI use when it was obvious.
• More readers (28%) were very confident in their ability to recognize an AI-generated cover than in recognizing an AI-generated book (14%), but the proportion of readers who said they were not very or not at all confident was also higher for AI-generated covers (32%) than AI-generated books (25%).
• Younger readers of sapphic fiction felt more confident they could recognize AI-generated books and covers, while confidence was lower among older readers.
• Readers under 35 were more confident in their ability to identify AI-generated covers than their ability to recognize AI-generated books. Readers aged 35-44 were about equally confident in both, while readers aged 55 and older were more confident that they could identify AI-generated books than covers.
• Confidence in recognizing AI-generated books and covers varied by country. Readers in France, Brazil, and Germany tended to be most confident, while readers in Australia felt least confident.
• The most frequently mentioned perceived “AI tells” were a high number of continuity errors and repetitions as well as an implausibly high number of books published within a short timeframe.
• The only irrefutable proof of generative AI use is prompts left in the book or an Amazon/Audible label such as “Virtual Voice” or “Kindle Translate” for audiobooks and translations, while an unusually high publishing frequency can be a strong indicator.
• Many of the things readers perceive as AI tells are indeed common in AI-generated books, but they also occur in badly written, poorly edited human-written books. Very few of the mentioned issues can be taken as definitive evidence that generative AI was used, particularly when they occur in isolation.
• My advice to readers is to look for good writing instead of AI evidence. While it’s difficult to distinguish an AI-generated book from a badly written one, a well-written book is easy to recognize and unlikely to be AI-generated.
• Readers can support human authors without becoming “AI detectives.” Leave reviews for well-written books, but don’t shy away from reviewing badly written books either. Don’t accuse authors of using AI without proof; instead, focus your review on the issues that affected your reading experience.

INTRODUCTION

Thank you for taking the time to check out part two of the sapphic fiction reader survey report!

In part 1, we looked at the book-buying and reading habits of sapphic fiction readers. How many sapphic books are they typically reading per month? Are they reading more or less than they did in the past? Where do they buy their books and get their book recommendations? What influences their decision to pick up a book?

In this second part of the survey report, we will take a closer look at how sapphic fiction readers feel about AI-generated books, covers, translations, and audiobooks and at how confident they are that they could distinguish an AI-generated book from one written by a human author.

But before I discuss the survey results, I’d like to preface this section with a few words of context. This is important, so please don’t skip this part of the blog post.

AI is a tricky topic, and I don’t want this survey report to be used to normalize generative AI in creative spaces, nor do I want it to fuel witch hunts against individual authors, accusing them of using AI without solid evidence. Both are harmful to the sapphic book community in different ways.

There are two points I’d like to address before we dive into the survey results: First, I want to explain how generative AI is harming sapphic fiction authors and what effect it has on the sapphic book market. Second, I’d like to make an important distinction between two different types of AI and clarify which one people are talking about when they reject AI in creative spaces.

How is generative AI affecting sapphic fiction authors?

If you have read the results of my author survey, you might already know how sapphic fiction authors are doing right now. If you haven’t had the time to check it out yet, let me summarize it for you:

80% of sapphic fiction authors are making either no or very little money from their writing. For most, it’s a passion project. That’s not exactly new. Writing in such a small niche has never been a way to get rich, but authors need to recoup the costs of editing and cover art or—if they write full-time—find a way to pay the bills, and both are becoming increasingly difficult.

According to the author survey results, more than half of sapphic fiction authors saw a decline in their writing-related income in the past year, and nearly one-third of authors even reported a significant decline. There are several different reasons for writers’ income going down, and the influx of AI-generated books is definitely one of them.

Most authors named the flood of AI books as their biggest challenge right now. With the help of generative AI, a single user can mass-produce a huge amount of books within a very short time. In fact, I just saw an ad for an AI tool that creates an entire book within an hour. According to a 2026 study, the number of new releases that are published every month has tripled since the launch of ChatGPT, and the sapphic book market is affected by it too.

The flood of AI-generated books makes it much harder for authors to get any visibility for their books. They have to compete with a myriad of AI books in search results, Amazon recommendations, bestseller lists, ads, and on social media—and most of the time, they are losing. The sheer volume of AI-generated books drowns out books written by human authors.

As a result, sapphic authors are forced to spend more time marketing their books instead of writing.

If their writing-related income goes down, they can no longer invest in expensive formats like audiobooks or translations, which further reduces the share of audiobooks with human narrators and translations by human translators in the market.

New authors are especially affected because AI undermines the trust between authors and readers. As we saw in part 1 of the reader survey report, many readers have understandably become more hesitant to pick up a book by a new or new-to-them author for fear that it might be AI-generated.

And authors aren’t the only creatives being affected. Generative AI used to create covers, audiobooks, or translations is having similarly devastating effects on human cover artists, audiobook narrators, editors, and translators. Many of them are losing their jobs or part of their income because some publishers and authors turn to generative AI instead of paying a human professional.

To give you an idea of how generative AI affects translators, publishers, and authors publishing in languages other than English: Here’s the breakdown of books in the German lesbian romance top 100 bestseller list on Amazon.de as of September 26, 2026:

• Translations I’m confident are AI-generated: 76% (including 15% Kindle Translate books)
• Translations I suspect are AI-generated but can’t reliably identify: 6%
• Books that are miscategorized and not actually sapphic: 6%
• Translations by publishers who hired human translators: 5%
• Translations by indie authors who hired human translators: 4%
• Books by German authors: 3%

So only 3% of the books in the German top 100 were authored by German writers and only 9% were translated by human translators, while at least 76% are AI-generated translations. That gives a pretty clear indication that generative AI pushes German authors and translators to the margins in their own market.

What do we mean when we talk about AI in creative spaces?

I’d like to emphasize an important distinction between two different kinds of AI.

Generative AI creates something new—a scene, a chapter, or an entire book; a cover; an audiobook; a translation; or a social media graphic or video—from a prompt. But it doesn’t create this content out of nothing. It generates it based on patterns the AI model has learned during training on large datasets that included millions of books and artwork downloaded from piracy sites and used without the permission of or compensation for writers and artists.

By contrast, assistive AI doesn’t create something new. It tries to improve a text, an image, or an audio recording that humans have created, e.g., grammar/spell checkers, tools that analyze manuscripts and flag clichés or overused words, software that helps audiobook narrators with noise reduction, and tools that remove an object in the background from an image. These tools have been around for many years, and they are generally accepted among readers, writers, and other creatives.

Of course, the distinction between assistive and generative AI isn’t always clear-cut. For example, if an author writes the first draft of a book themselves but then uses AI to heavily rewrite their manuscript in a way that completely changes the style and voice, can that still be considered assistive? I would argue that at some point, the AI revisions become so substantial that the book is no longer really the author’s.

Some people insist that AI translations are assistive, but even Amazon KDP considers them generative. AI doesn’t just assist translators; it replaces them entirely. With AI translation programs such as Kindle Translate, the entire process is automated, with no human translator or editor involved. A good translation isn’t just a mechanical process of replacing the words of one language with their equivalents in another. A lot of creativity goes into translating a novel. Word-for-word translations often don’t work because humor and banter depend on cultural context, wordplay, and double meanings of words that might not exist in the target language. The same emotion is expressed in very different ways in different cultures and languages. Translators often have to rewrite entire passages to achieve the same emotional effect as the original.

When you hear people in creative spaces speak out against AI, they are usually talking about generative AI—AI that replaces human creatives such as authors, translators, editors, narrators, and cover artists, enabling people to mass-produce AI content in a short amount of time, flooding the market, and making it harder for readers to find human-created books.

That’s why the survey focused on readers’ attitudes toward generative AI.

SAPPHIC FICTION READERS’ ATTITUDES TOWARD GENERATIVE AI

With that context and these caveats in mind, let’s take a look at the survey results and how readers feel about AI-generated books, audiobooks, translations, and covers.

Would readers buy or read a sapphic book if they knew or suspected it was written by AI?

When asked whether they would buy/read a book that was written by generative AI, readers answered:

• Hard pass (Would not read an AI-generated book): 87%
• Depends on the story (Prefer books by human authors, but would buy an AI book if the book description sounds really interesting): 5%
• Depends on the price (Wouldn’t pay for it, but would read it if it was free or included in Kindle Unlimited): 4%
• Undecided: 4%
• Sure (Don’t mind reading AI-generated books): 1%

The majority of sapphic fiction readers reject AI-generated books. 87% would not read a book if they knew or suspected it was AI-generated under any circumstances.

9% prefer human-created books but might read an AI-generated book under specific circumstances, for example, if the blurb sounds compelling or the book is free or included in their subscription.

Only 1% said they don’t mind reading AI-generated books.

Interestingly, the acceptance of AI-generated books seems to be higher among older readers than among younger readers. Here’s how the percentage of readers who said they wouldn’t buy or read an AI-generated book breaks down by age group:

• <25 years: 97% chose the “hard pass” answer
• 25-34 years: 93%
• 35-44 years: 88%
• 45-54 years: 84%
• 55-64 years: 82%
• 65 or older: 80%

So for 20% of readers aged 65+, an AI-generated book is not an automatic dealbreaker, while only 3% of readers under 25 feel the same way.

Readers from predominantly English-speaking countries were no more accepting of AI-generated books than readers from countries in which English is not the predominant language. In both groups, 87% of readers said they wouldn’t read or buy an AI-generated book.

However, there were differences between individual countries: Here are the percentages of readers who said they wouldn’t buy or read an AI-generated book in the countries with the highest numbers of participants:

• Brazil: 93% chose the “hard pass” answer
• Canada: 91%
• Netherlands: 90%
• United States: 86%
• United Kingdom: 86%
• France: 85%
• Australia: 85%
• Germany: 84%

Would readers buy or read a sapphic book if they knew or suspected that its cover was AI-generated?

When asked whether they would buy/read a book with an AI-generated cover, readers answered:

• Hard pass (Wouldn’t read a book with an AI cover): 44%
• Depends on the author (Don’t love AI covers, but would still buy the book if it’s an author they trust not to use generative AI for their writing): 22%
• Depends on the story (Prefer books with a cover created by a human cover artist, but would buy a book with an AI cover if the book sounds really interesting): 13%
• Depends on the price (Wouldn’t pay for it, but would read it if it was free or included in Kindle Unlimited): 6%
• Undecided: 7%
• Sure (Don’t mind reading books with an AI cover): 8%

Nearly half (44%) of sapphic fiction readers would not read a book if they recognize the cover as AI-generated.

41% prefer human-created covers, but they would read a book with an AI-generated cover under certain circumstances—if they trust the author not to have used generative AI for their writing (22%), if it’s a compelling blurb (13%), or if the book was free or included in their subscription (6%).

The relatively large group of readers who chose the “depends on the author” category is particularly interesting. Their response might indicate that, in general, these readers do take an AI-generated cover as a sign that the story might be AI-generated too, so they are only willing to overlook the AI cover if it’s an author they trust to not use AI to write the book.

The group that said they don’t mind AI-generated covers remains a small minority (8%), indicating that while AI covers might meet less opposition than AI-generated books, they are still not broadly accepted by the majority of sapphic fiction readers.

The acceptance of AI-generated covers seems to be higher among older readers. Here’s how the percentage of readers who said they wouldn’t buy or read a book with an AI-generated cover breaks down by age group:

• <25 years: 73% chose the “hard pass” answer
• 25-34 years: 64%
• 35-44 years: 49%
• 45-54 years: 37%
• 55-64 years: 27%
• 65 or older: 27%

As with AI-generated books, younger readers were more likely to view an AI-generated cover as a dealbreaker, while acceptance increases steadily among older readers.

The acceptance of AI-generated covers also differs between countries. Here are the percentages of readers who said they wouldn’t buy or read a book with an AI-generated cover in the countries with the highest numbers of participants:

• France: 59% chose the “hard pass” answer
• Brazil: 55%
• Canada: 52%
• United States: 44%
• Australia: 41%
• United Kingdom: 39%
• Germany: 38%
• Netherlands: 37%

As with AI-generated books, Brazil and Canada were among the countries with the highest percentage of readers who reject AI-generated covers.

Would readers listen to an AI-generated audiobook (Virtual Voices)?

For this question, I looked at only those sapphic fiction readers who listen to audiobooks. When asked whether they would listen to an AI-generated audiobook, readers answered:

• Hard pass (Wouldn’t listen to an AI-generated audiobook): 87%
• Depends on the story (Prefer human narrators, but would listen to AI narration if the book description sounds really interesting): 2%
• Depends on the price (Wouldn’t pay or use a credit on it, but would listen to AI narration if it was free or included in their subscription): 3%
• Depends on the author (Prefer human narrators, but I would listen to AI narration if it’s an author they trust not to use generative AI for their writing): 3%
• Undecided: 4%
• Sure (Don’t mind listening to AI-generated audiobooks): 1%

Readers’ attitudes toward AI-generated audiobooks are remarkably similar to their attitudes toward AI-generated books. Clearly, audiobook listeners appreciate what human narrators bring to the experience.

The vast majority of sapphic fiction audiobook listeners strongly reject AI-generated narration—at exactly the same rate as for AI-generated books (87%).

8% of audiobook listeners prefer human narrators but would be willing to listen to an AI-generated audiobook under specific circumstances, e.g., if it was free or included in their subscription (3%), if they trusted the author not to have used generative AI for their writing (3%), or if the story blurb sounded compelling (2%).

Only 1% of audiobook listeners doesn’t mind AI-generated audiobooks at all—exactly the same percentage as readers who said they don’t mind AI-generated books.

As with AI-generated books, the acceptance of AI-generated audiobooks seems to be higher among older readers. Here’s how the percentage of readers who said they wouldn’t buy or read an AI-generated audiobook breaks down by age group:

• <25 years: 91% chose the “hard pass” answer
• 25-34 years: 92%
• 35-44 years: 88%
• 45-54 years: 88%
• 55-64 years: 81%
• 65 or older: 80%

The acceptance of AI-generated audiobooks also differs between countries. Here are the percentages of readers who said they wouldn’t buy or read an AI-generated audiobook in the countries with the highest numbers of participants:

• France: 98% chose the “hard pass” answer
• Brazil: 92%
• Canada: 92%
• United Kingdom: 88%
• Germany: 88%
• Australia: 87%
• United States: 85%
• Netherlands: 83%

Would readers buy or read a sapphic book if they knew or suspected it had been translated into their language by generative AI?

When asked whether they would buy/read a book that had been translated by generative AI, e.g., Kindle Translate, readers answered:

• Hard pass (Wouldn’t read an AI-generated translation): 62%
• Depends on the author (Would prefer a human translator but would still buy an AI translation if it’s an author they trust not to use generative AI for their writing): 11%
• Depends on the story (Prefer books by human authors and translators, but would buy an AI translation if the book description sounds really interesting): 5%
• Depends on the price (Wouldn’t pay for it, but would read it if it was free or included in Kindle Unlimited): 5%
• Undecided: 14%
• Sure (Don’t mind AI-generated translations): 3%

The majority of sapphic fiction readers wouldn’t read a book if they knew it had been translated into their language by generative AI. However, the percentage of readers who consider AI a dealbreaker drops from 87% for books and audiobooks to 62% for translations.

21% of readers prefer human translators but would read an AI translation under certain circumstances, e.g., if they trust the author not to have used generative AI for writing the original story (11%), if the story premise was compelling (5%), or the book was free or included in their subscription (5%).

The proportion of readers who don’t mind AI-generated translations is small (3%), but slightly bigger than the percentage of readers who doesn’t mind AI-generated books or audiobooks (1%).

As with AI-generated books, the acceptance of AI-generated translations seems to be higher among older readers. Here’s how the percentage of readers who said they wouldn’t buy or read an AI-generated translation breaks down by age group:

• <25 years: 73% chose the “hard pass” answer
• 25-34 years: 73%
• 35-44 years: 62%
• 45-54 years: 61%
• 55-64 years: 56%
• 65 or older: 53%

The acceptance of AI-generated translations also differs between countries. Here are the percentages of readers who said they wouldn’t buy or read an AI-generated translation in the countries with the highest numbers of participants:

• France: 87% chose the “hard pass” answer
• Brazil: 72%
• Canada: 70%
• Netherlands: 67%
• Australia: 67%
• Germany: 65%
• United Kingdom: 63%
• United States: 58%

Interestingly, while readers from predominantly English-speaking and non-English-speaking countries were equally likely to give AI-generated books a “hard pass,” there was a difference in attitudes toward AI-generated translations. 70% of readers from countries where English is not the predominant language said they would not read an AI-generated translation, compared to 60% of readers from predominantly English-speaking countries.

The survey doesn’t reveal the reason for that difference. One possible explanation is readers’ familiarity with translated literature. Readers from countries where English is not the predominant language are more likely to encounter translated books. For example, only about 3% of books published in the US are translations, while in Germany, France, Italy, and Spain, translations make up around 12-15% of new releases. In smaller markets, translations even account for 20-40% of new releases. Because readers from non-English-speaking countries are, on average, more familiar with translations—perhaps including AI translations, they might be more aware of how much can get lost in translation and more skeptical about trusting AI with the nuances of their language.

If readers see an author use a lot of AI-generated images or videos to promote their books on social media, would that affect whether they read or buy that author’s books?

When asked whether they would buy/read a book from an author who regularly uses generative AI to promote their books, readers answered:

• Hard pass (Wouldn’t buy or read a book from an author who uses AI-generated images or video to promote their books): 42%
• Depends on the author (Don’t love authors using AI to promote their books, but would still buy the book if it’s an author they trust not to use generative AI for their writing): 22%
• Depends on the story (Don’t love authors using AI to promote their books, but would still buy the book if it sounded really interesting): 10%
• Depends on the price (Wouldn’t buy their books but would still read them if they are free or included in Kindle Unlimited): 4%
• Undecided: 10%
• Don’t mind authors using AI-generated images or video to promote their books: 12%
• Really enjoy it when authors use AI-generated images or video to promote their books: 0.5%

Reader reactions to AI-generated social media/promotional images or video resemble the findings on AI covers.

AI-generated images or video used for marketing are a dealbreaker for nearly half of sapphic fiction readers. 42% say they would not buy or read a book by an author who uses AI-generated content for their marketing, even if the book itself was not created by AI.

36% don’t love authors using AI to promote their book but would be willing to overlook AI marketing under specific circumstances, e.g., if it was an author they trust not to have used AI for their writing (22%), if the story blurb sounded compelling (10%), or the book was free or included in their subscription (4%).

12% of readers said they don’t mind authors using AI-generated images or video to promote their books, while only 0.5% said they enjoy it.

The acceptance of AI-generated promotional images and video seems to be higher among older readers. Here’s how the percentage of readers who said they wouldn’t buy or read a book by an author who uses AI-generated content on social media by age group:

• <25 years: 75% chose the “hard pass” answer
• 25-34 years: 67%
• 35-44 years: 48%
• 45-54 years: 32%
• 55-64 years: 23%
• 65 or older: 21%

The acceptance of AI-generated promotional content also differs between countries. Here are the percentages of readers who said they wouldn’t buy or read a book by an author who uses AI-generated content on social media in the countries with the highest numbers of participants:

• France: 59% chose the “hard pass” answer
• Brazil: 59%
• Canada: 49%
• Australia: 43%
• United States: 42%
• Netherlands: 39%
• United Kingdom: 38%
• Germany: 36%

At one glance: Sapphic fiction readers’ attitudes toward generative AI

Here’s a direct comparison of how sapphic fiction readers responded to AI-generated books, covers, audiobooks, translations, and promotional content:

Readers are most opposed to generative AI when it comes to writing the book itself and narrating the audiobook: 87% of readers would give AI-generated books and audiobooks a hard pass, while only 1% are accepting.

Attitudes become more divided when it comes to the parts of the publishing process that might be seen as “promotional packaging.” AI-generated covers and marketing material are a dealbreaker for nearly half of readers, while the other half might still pick up the book.

Interestingly, translations sit somewhere in the middle. While only 3% don’t mind AI translations, they are a hard pass for 62% of readers—fewer than those rejecting AI-generated books and audiobooks but more than those who reject AI covers and marketing content.

Translations also have the highest rate of respondents who are undecided, suggesting that some readers may not yet have formed a clear opinion about where they stand on AI translations.

SAPPHIC FICTION READERS’ CONFIDENCE IN RECOGNIZING AI-GENERATED CONTENT

Important note: The survey measures how confident readers feel in their ability to recognize AI-generated content; it doesn’t establish how good readers actually are at telling AI books or covers from human-created ones.

How confident are sapphic fiction readers that they could tell whether a book was written by generative AI instead of a human author?

When asked how confident they were that they could identify AI-generated books, readers said:

• Very confident (Can usually tell): 14%
• Somewhat confident (Can tell when it’s obvious): 61%
• Not very confident (Probably often miss it): 16%
• Not confident at all (Have no idea how to recognize AI-generated books): 9%

Three-quarters of readers are at least somewhat confident in their ability to recognize when a book was generated by AI instead of written by a human author. However, most (61%) say they can only recognize AI-generated books when it’s fairly obvious.

25% of readers are not very or not at all confident they could tell when a book was generated by AI, with 9% having no idea at all how to recognize AI-generated books.

Confidence levels varied significantly by age group. Here’s the percentage of readers in each age group who said they were very or at least somewhat confident that they could identify an AI-generated book:

• <25 years: 85%
• 25-34 years: 81%
• 35-44 years: 80%
• 45-54 years: 79%
• 55-64 years: 69%
• 65 or older: 63%

Younger readers were more likely to feel confident in their ability to recognize AI-generated books.

There were also differences between readers from different countries in how confident they felt about identifying AI-generated books. Here’s the percentage of readers in each country who said they were very or at least somewhat confident that they could identify an AI-generated book:

• France: 91%
• Germany: 82%
• Brazil: 79%
• Netherlands: 78%
• United States: 74%
• Canada: 76%
• United Kingdom: 71%
• Australia: 69%

How confident are sapphic fiction readers that they could tell whether a book cover was created by generative AI instead of a human cover artist?

• Very confident (Can usually tell): 28%
• Somewhat confident (Can tell when it’s obvious, e.g., weird hands): 40%
• Not very confident (Probably often miss it): 22%
• Not confident at all (Have no idea how to recognize AI-generated covers): 10%

More readers are very confident that they could usually tell when a cover is AI-generated compared to their confidence in being able to recognize an AI-generated book (28% vs. 14%).

However, fewer readers describe themselves as somewhat confident when it comes to covers. 40% say they can tell in obvious cases, compared to 61% for AI-generated books.

Overall, 68% of readers are very or somewhat confident that they could identify an AI-generated cover, while 32% are either not very or not at all confident.

So it’s not quite as simple as saying readers are more confident that they would recognize an AI-generated cover compared to a book. The degree of confidence varies between the two. Readers are more likely to say they are very confident about identifying AI covers (28% vs. 14%), yet they are more likely to say they are at least somewhat confident about recognizing an AI-generated book (75% vs. 68%).

The number of readers who say they have no idea how to tell whether a cover or a book has been generated by AI is nearly the same: 9% for books and 10% for covers.

Confidence levels varied significantly by age group for covers too. Here’s the percentage of readers in each age group who said they were very or at least somewhat confident that they could identify an AI-generated book:

• <25 years: 93%
• 25-34 years: 88%
• 35-44 years: 79%
• 45-54 years: 65%
• 55-64 years: 54%
• 65 or older: 41%

Interestingly, readers under 35 were more confident in their ability to identify AI-generated covers than in their ability to recognize AI-generated books. Readers aged 35-44 were about equally confident in both, while readers aged 55 and older were more confident that they could identify an AI-generated book than an AI-generated cover.

Younger readers were more likely to feel confident in their ability to recognize AI-generated covers.

There were also differences between readers from different countries in how confident they felt about identifying AI-generated covers. Here’s the percentage of readers in each country who said they were very or at least somewhat confident that they could identify an AI-generated cover:

• Brazil: 90%
• France: 89%
• Germany: 74%
• Canada: 71%
• United States: 67%
• United Kingdom: 67%
• Netherlands: 59%
• Australia: 57%

WHAT READERS PERCEIVE AS SIGNS OF AI-GENERATED BOOKS

Important caveat: This section is not an instruction manual on how to spot AI

The results in this survey report reflect what readers perceive or assume to be indicators that generative AI was used, not necessarily actual reliable signs. Please do not mistake these results for a “checklist” for determining whether a book was AI-generated or not.

Make sure to carefully read my comments on each perceived “tell.” I’ll explain for each one how strong or weak the connection to generative AI actually is, relying not just on my own experience but linking to scientific research.

I want to make it clear that while many of the things readers mentioned are indeed issues that are often associated with AI-generated books, there can also be other explanations for most of them. A book by an inexperienced author who didn’t invest in professional editing or an author who was rushing to get a book out might have some of the same problems.

So very few of the things listed below should be taken as definitive proof that generative AI was used, particularly when they occur in isolation. If a lot of them occur together, it might become more likely that generative AI was used, but it still doesn’t constitute irrefutable evidence, so please be very careful with accusations.

I actually debated whether I should include this section in the survey at all, but I know readers are already making these assumptions, regardless of whether I talk about them or not—after all, the list you’ll find below consists of things readers spontaneously mentioned in an open-ended question, not options they picked in a multiple-choice question that I spelled out for them.

So I decided to include this section because it gives me a chance to examine these judgments and perhaps correct some misconceptions. Hopefully, it will help prevent authors from being wrongfully accused of using generative AI for things like using Oxford commas or em dashes while also guiding readers toward signs that more reliably point toward a book being AI-generated.

How do readers recognize a book might be AI-generated?

For this question, readers were asked to name specific things that would make them suspect a book might be written by AI. Because it was an open-ended question, readers could give any answer they wanted rather than choosing from predefined options.

With the caveat I explained above in mind, here’s the list of most commonly mentioned things readers perceive as being typical for an AI-generated book. They are listed in order of how frequently readers mentioned them, not by how reliable they actually are as indicators of whether a book might be AI-generated.

• Continuity errors, inconsistencies, and contradictions (15%). AI tends to have a “bad memory” and can’t keep track of the details mentioned earlier in the book. As a result, character names, appearance, background info, or location can suddenly change with no explanation; timelines don’t add up, and there might be plot holes and storylines that don’t go anywhere. For example, the main character might drive a BMW in chapter 2 but a Toyota in chapter 17 or a brother who was said to have died suddenly makes an appearance later in the story. Research shows that continuity errors are indeed a well-documented issue in AI-generated books, with inconsistencies becoming more apparent as the story gets longer. I’m not talking about one little mistake in a full-length novel; I’m talking about a pattern of continuity issues. However, they could also be caused by an author who didn’t take the time to reread and revise her book and who didn’t work with beta readers and editors who could have helped to catch these mistakes.
• Repetitions of words, phrases, plot points, or situations (15%). Research shows that constant repetitions can indeed be a problem that occurs in AI-generated books, particularly when they happen without a stylistic purpose. However, human authors often have favorite phrases and overused words, so they might repeat themselves too. Most of these repetitions are usually caught and edited out during the revision stage, but you might be dealing with a poorly edited book.
• Very high volume of books published within a relatively short time (13%). It’s a well-established fact that AI can generate large amounts of text much faster than any human could ever write. But please remember that there are prolific authors who could have a new book out every month, especially if they write full-time or write shorter works. There’s no specific number of books per year that can be seen as clear proof of generative AI use, but to give you an indication of what might or might not be possible for a human author: Romance novelist Barbara Cartland is considered one of the most prolific authors ever. She holds the Guinness Book of Records record for the highest number of novels written in one year: 27. Most of her books were on the short side for a novel, around 40,000 words, and might be considered novellas by today’s standards. So I think it’s safe to say that outputs such as that of Aurora North (170+ novels in 8 months) or Blair Lennox (36 full-length novels in 1 month) are impossible for a single human author who isn’t using AI. However, it’s important to not just look at the number of books an author has published but also at the time frame and the length of their books. For example, if you look at my Amazon profile, it will tell you that I currently have 139 books out. That’s because I’ve been publishing for 19 years; only 54 of these books are novels, and half of them are my own translations of my books into German, my native language. Also be careful not to wrongly accuse an author who is re-releasing all of their books after receiving their rights back and starting to self-publish.
• AI-generated cover (6%). An AI-generated cover is not proof that the author also uses generative AI for their books, but it’s an indication that they are willing to use AI for at least part of their publishing process. To quote two of the readers participating in the survey: “If an author is happy to use AI for their cover, then I assume they are more likely to use it in their writing” and “Not every AI cover has AI text, but most AI-generated books have AI covers.” That’s definitely true for Aurora North and Blair Lennox. However, there are authors who create an AI cover for their book because they feel they can’t afford to hire a human cover artist. (Authors, if you are in that position, there are graphic designers who offer very affordable covers starting at $10).
• Recycled storylines (5%). Several readers reported that a book being similar to others they have already read makes them suspect it’s AI-generated. Research shows that AI often ends up producing very similar stories, all based on the same template. AI models learned story structure from commonly used beats and patterns in stories gathered from the Internet or downloaded from piracy sites. That can result in plots that feel cut-and-paste, like a story you’ve read a hundred times before. There are also scammers who use generative AI to basically create knockoffs of popular sapphic books. I’ve blogged about one specific AI scammer here. However, human authors also use familiar patterns and tropes, especially in genres such as romance. Just because there are some similarities between different fake-dating stories or between different enemies-to-lovers romances does not mean the author used generative AI. Fractal Enigma, the AI content farm Aurora North and Blair Lennox belong to, relies heavily on tropes, but so do many legitimate writers. The difference is that a skilled human author will put their own spin on these tropes and create unique characters, while AI—or an inexperienced or careless author—may create books with nothing to distinguish them from other novels containing the same tropes.
• Lack of emotional depth (4%). Many readers feel that the character emotions in AI-generated books seem superficial, generic, and flat. Research shows that AI-generated stories often score lower for emotional depth and psychological complexity. However, some human authors might also struggle to portray emotion in subtle and immersive ways.
• Overuse of certain words and phrases (4%). Another issue that many readers perceived as a sign of an AI-generated text was specific words and phrases that are overused. Most often, readers mentioned the phrase “It’s not X; it’s Y.” For example: “It wasn’t the pain that frightened her. It was the silence.” Research has found evidence that certain words or phrases are indeed more common in AI-generated than in human-created texts. According to an estimate, “it’s not X; it’s Y” or “not just X but Y” sentences occur three times more often in AI-generated texts than in human writing. However, that doesn’t mean that human writers don’t use those phrases. It’s more useful to look for an unusually high frequency of such phrases rather than an occasional use.
• Author has no online presence (3%). Some readers concluded that if an author has no website and isn’t active on social media, this could be a sign that their books are AI-generated. While it’s true that many people who mass-produce books with the help of generative AI don’t bother setting up websites or interacting with people on social media, there are certainly many valid reasons for why a human author might not have an online presence, particularly for authors within the LGBTQ+ community. Some authors would rather keep their private lives private or want to protect their mental health by not being on social media.
• Stilted, unnatural-sounding dialogue (3%). AI-generated dialogue can sometimes sound too formal. Readers also observed that characters sometimes seem to be talking “at” instead of “to” each other. Research has shown that AI-generated dialogue can lack in subtext and is often too expository, with characters overexplaining their emotions or the situation in dialogue, and that AI-generated conversations exhibit a greater sense of formality. However, there are also human authors who struggle to write dialogue. In some cases, dialogue could be too formal because the author is not a native speaker.
• Reviews and ratings indicate it’s AI (3%). Unfortunately, the average star rating is often not very helpful in indicating books that might be AI-generated. Some AI-generated books do have lower average ratings, especially if a lot of readers picked up on the fact that it’s AI-generated, but others have ratings that compare to human-created stories. More useful than ratings are individual text-based reviews. If dozens of reviewers mention that they suspect a book to be AI-generated or many reviews point out issues such as repetitive phrasings or inconsistencies, it could be a warning sign that warrants caution. However, reviewers can also be mistaken, so don’t take a single person accusing an author of AI usage as definitive evidence.
• Lack of character depth (3%). Research shows that AI-generated characters tend to be less varied and more stereotypical, leaning more heavily on archetypes. AI characters are also more likely to have their motivations and personality traits spelled out and explained (“telling” instead of “showing”). AI tends to lack emotional complexity for morally ambiguous characters and when portraying deep psychological struggles. However, human-created fiction, especially romance, is also often based on familiar character types such as the aloof ice queen. Skilled writers can take these archetypes and give them depth, contradictions, vulnerabilities, and individual traits to make them feel complex and three-dimensional, whereas a less experienced writer might fail to do that, resulting in characters that feel more flat and one-dimensional.
• Wordiness and filler (3%): Research shows that AI-generated texts tend to be more wordy, with padding, long-winded sentences, and details that don’t move the plot forward and unnecessarily stretch out the story. However, some human authors also tend toward wordiness, especially in early drafts. A good editor and thorough revisions by the author will usually cut unnecessary details and tighten the prose, but if the author didn’t hire an editor or take the time to revise, padding might be left in the published story.
• Overly descriptive writing (2%). Related to the wordiness issue, AI-generated texts are more likely to contain a lot of overdone, drawn-out descriptions. That’s especially noticeable at chapter openings. If each chapter is created by a new prompt, AI treats it like a story opening and is trying to set the mood. Research also shows that AI and human authors typically write different kinds of descriptions. AI models are more likely to write atmospheric descriptions that don’t focus on what the characters do, which sets the mood but stops the forward momentum of the story. By contrast, human-written stories tend to include more descriptions driven by the characters’ interaction with the setting, e.g., moving through a room, touching objects. However, this tendency toward lengthy, atmospheric descriptions is not exclusive to AI; some human authors—especially in genres such as literary fiction or Gothic horror—also make extensive use of descriptions to create atmosphere.
• Prompts left in the text (2%). This one should definitely be at the very top of this list. Unlike most of the other potential signs, if the author accidentally left prompts (instructions to the AI model) in the text, it’s direct evidence that generative AI was involved in producing the story.
• Blank or generic author’s bio (2%). Many AI users who are mass-uploading AI-generated books to platforms such as Amazon indeed don’t take the time to set up an author’s profile, so if you click on the author’s name, you either find no bio at all or one that sounds very generic, with little to no information about the person behind the writing. However, there could also be other reasons for why an author left their profile blank or vague. Some authors prefer to stay anonymous to keep their personal lives private. For LGBTQ+ authors who aren’t fully out, it might be especially important to protect their privacy. It’s also possible that inexperienced authors might not realize that they have to set up their profile via Amazon Author Central.
• Spelling and grammar mistakes (2%). Some readers assume that AI-generated stories would contain more typos, spelling errors, and grammar mistakes. I actually think it’s the opposite. AI-generated text often seems flawless at first glance, especially when it comes to spelling. The writing appears polished when you look at individual sentences, while problems such as inconsistencies become more apparent the further you read.
• Overuse of figurative language (2%). Research shows that AI-generated stories do typically contain more metaphors and similes than human-created writing. More importantly, AI often produces metaphors that may sound poetic but don’t always make sense or have any connection to the rest of the story. In comparison, a skilled writer might introduce a metaphor early in the story and develop it throughout the book. However, an inexperienced writer who is trying too hard to make their writing sound poetic might also overuse figurative language.
• Formulaic, predictable plot (2%). Research shows that AI tends to default to the same narrow set of story blueprints, instead of telling unique stories. AI models were trained on large quantities of existing stories. They generate text by predicting what’s likely to come next, leaning on patterns and story beats found in their training data. This can make AI-generated books feel more predictable and formulaic than human-written stories. However, a formulaic plot is not proof that the book was generated by AI. There are plenty of predictable human-authored stories out there. Some “formulas” are well-established genre conventions that readers expect and enjoy in certain genres. For example, romance novels often include a “meet-cute” scene at the beginning of the story and a “grand gesture” by one character toward the end. A skilled writer will work within these conventions while still making the story feel fresh and original, while an inexperienced writer may rely too heavily on predefined beats, treating them like a checklist rather than just a starting point.
• Excessive use of the em dash (1%). Research shows that AI has used the em dash more frequently than human writers in the past. However, while some AI models still use them excessively, many newer models use them at the same or even lower rate than human writers. And please keep in mind that there are plenty of authors who love em dashes—myself included. A good editor will flag them when they become so frequent that they start to stand out, but excessive use could simply be a sign that the manuscript needed more editing.
• Mixed-up pronouns (1%): Generative AI sometimes does mix up character pronouns, e.g., using he/him for a character whose pronouns are she/her. I’ve frequently seen it happen in AI-generated translations. That happens because AI models have absorbed a lot of gender stereotypes during training. They were trained on a large number of stories in which certain names were usually linked to specific pronouns. If a character has a name historically associated more strongly with another gender, AI may fall back on those learned associations rather than consistently use the pronouns established in that story. Of course, a single incorrect pronoun by itself doesn’t necessarily mean the book is AI-generated. It might be a simple typo that slipped past the editor or the character may have had a different gender in an earlier draft.
• The story premise sounds good and the book starts promising but then goes downhill (1%). That actually is a tendency I’ve noticed in AI-generated fiction. The blurb sounds appealing because it’s based on popular tropes, and the story’s beginning can be engaging, but the longer the story goes on, the harder it becomes for the AI model to sustain a coherent plot and consistent characterization. As I mentioned earlier, AI works by predicting what comes next, but it struggles to keep track of details and events from earlier in the story, so inconsistencies start to pop up and the plot starts to become repetitive. However, a promising start followed by a big drop-off in quality could also be a sign of an inexperienced human author who had a good idea but didn’t have the skills to execute it.
• Flowery prose (1%): Research shows that generative AI has a tendency toward using flowery, overwritten language that sounds poetic but doesn’t add much meaning. That includes the excessive use of metaphors and similes described above. However, some authors also use poetic prose as part of their writing style, and inexperienced writers may unintentionally overdo it.
• No translator or cover artist credited on the copyright page (1%). Human authors often credit their translator, editor, and cover artist on the copyright page of the book, while low-effort AI-generated books that were created without professional involvement might not include those credits. However, in some cases, there might be other reasons for missing credits, e.g., a cover artist who prefers not to be named or an author who couldn’t afford to hire someone and did everything by themselves. Unfortunately, there are also cases where authors claim a human translator or cover artist worked on a book and even name them in the credits when they actually used generative AI for the translation or the cover.
• Multiple translations being published within a relatively short time (1%). Simultaneous human-created translations into multiple languages are not that unusual for books by big-name authors, published by mainstream publishers who can afford it. However, translations by skilled human translators are expensive, and indie authors would also have to pay an editor to go over the translation. Since sapphic fiction is a relatively small market, especially in many non-English-speaking countries, it can take a considerable time to recoup the costs of a translation. Few indie authors can afford to invest thousands of dollars (or whatever their currency is) per language to publish translations of their book into Spanish, Portuguese, German, French, and Italian simultaneously. Professional translations also take time. If an author publishes translations of their novel in multiple languages within only a couple of weeks of the original’s release, that can raise legitimate questions about whether there was enough time for human translators to translate the book. Of course, an author publishing AI-generated translations doesn’t necessarily mean their original book was also generated by AI, but it does show that the author was willing to use generative AI for at least part of their publishing process.
• Lack of variety in sentence structure (1%). Research shows that AI-generated text typically contains less variety in sentence length and structure compared to texts by human writers, who vary their rhythm more. However, some human writers tend to have very repetitive sentence patterns, and if they publish their book without thorough editing, their writing can have a similar lack of variety.
• Missing or AI-generated author photo (1%). Some readers felt that a lack of an author picture or an AI-generated photo on the author’s profile might make it more likely that the pen name is part of an AI content farm rather than a legitimate author. AI publishing operation Fractal Enigma indeed uses AI-generated photos for their supposed authors. However, the lack of a photo on an author’s Amazon page could have reasons other than AI involvement: Many legitimate authors write under pen names and prefer to keep their identity private. Especially for sapphic fiction authors who are members of the LGBTQ+ community, it might not be safe to reveal their face online, so they prefer to instead use a logo, a cartoon illustration resembling them, or no picture at all.
• Rule of three (0.5%). The rule of three (also called tricolon) is a rhetorical device that groups words, phrases, or ideas into groups of three. It’s a fundamental writing principle that many authors use instinctively. There are many famous examples in literature, slogans, and public speaking, including Abraham Lincoln’s “Government of the people, by the people, for the people.” Because AI is trained on human-written text, it tends to frequently use the rule of three too. Research shows that some AI models do overuse it, so several back-to-back uses in sentence after sentence might become suspicious, but occasional use of the rule of three is very common in human writing and is not an indication that AI was used.
• Same character names (0.5%). Research shows that most AI models indeed tend to suggest the same character names over and over again. You’ll often see first names like Elara, Elena, Mara, Vivian/Vivienne, and Seraphina as well as last names like Vance, Thorne, Chen, Voss, Sterling, Ashwood, and Hart. Of course, not every use of those names indicates that the author is using generative AI, but if you encounter a sapphic romance between Elena Voss and Maya Chen and then find two different novels (not part of a series) by the same author featuring a protagonist named Elena Vance, that’s definitely a red flag, especially if it occurs alongside other patterns commonly associated with AI-generated fiction. At the very least, it indicates that the author might be using AI to pick their characters’ names. By the way, the example I used above isn’t hypothetical; those are name combinations I found in Aurora North’s novels.

Summary of the list of things readers often perceive as red flags for an AI-generated book

Apart from prompts left in the text, the strongest sign of substantial AI use is an implausibly high number of books released within a short window of time. Similarly, the release of an unusually large number of translations into multiple languages within a short timeframe can be a strong indicator that generative AI was used for the translations.

A definitive sign of generative AI that few readers mentioned is an Amazon/Audible label. While Amazon unfortunately doesn’t label ebooks, physical books, or covers that have been created with generative AI, they do make AI-generated audiobooks and at least some of the AI-generated translations evident.

If a book was translated with Kindle Translate, Amazon’s generative-AI translation tool, you’ll find the words “Translated with Kindle Translate” beneath the cover on the book’s Amazon page. You can also recognize Kindle Translate books by their covers. They have a white frame, and at the top, it says: “Deutsche Ausgabe” (German edition), “Edição em português” (Portuguese edtion), “Édition française” (French edition), etc.

However, if the author or publisher used their own AI tool to create the translation, you won’t find any warning label.

For an AI-generated audiobook on Amazon or Audible, you will see “Virtual Voice” listed as the narrator.

Most other perceived indicators are often associated with AI-generated books, but they are also typical for human-authored stories that were published before they were ready. While they could point toward a book being AI-generated, they could also be signs of a badly written, poorly edited book. Very few of them are reliable proof that generative AI was used, especially when only a few of them occur in a book.

A couple of continuity issues, a tendency to be wordy, a one-dimensional main character named Elena, and an author without an online presence do neither prove nor rule out that the book was AI-generated.

At the same time, we shouldn’t go too far in the other direction and explain away a combination of pretty much all of these characteristics occurring together.

If an author with an AI-generated author photo releases 170+ novels in less than a year, each one with an AI-generated cover, most of them with AI-typical names and a dozen reviews either suspecting the books are AI-generated or pointing out endless repetitions and continuity errors, it’s reasonable to question if the books were really written by a human.

Please don’t go on witch hunts accusing authors of having used generative AI merely because they prefer not to share a picture of their face or use a few too many em dashes. But please let’s also not pretend that AI-generated books aren’t flooding most book markets, including sapphic fiction.

MY ADVICE TO READERS

I know readers are getting increasingly frustrated with the onslaught of AI-generated books. Many of them are desperate for some guidance. I’ve been asked by readers repeatedly if I can provide them with a checklist of “AI tells” that can help them make sure they aren’t buying or reading AI-generated books. For the reasons I explained above, that’s not possible.

But there’s some good news: While distinguishing an AI-generated book from a badly written or poorly edited one can be tricky, it’s easy to tell a well-written book from an AI-generated one.

So here’s my advice to readers: Instead of looking for AI evidence, look for good writing. A book that doesn’t just have a promising story idea but great execution in all the aspects mentioned above can’t be created by AI. Pick books with complex characters and emotions; dialogue that sounds natural; tropes used in a fresh, unique way; and a plot that makes sense and moves forward at a steady pace, without constant repetitions and unnecessary filler.

If you do that, you won’t waste your time or your money on a book that is AI-generated or simply badly written; you’ll instead spend them on a book that is genuinely worth reading.

Support authors who took the time to learn the writing craft and thoroughly revise their stories to make them the best they can possibly be. Recommend well-written books, ask your library to add them to its catalogue, and write a review.

Don’t shy away from writing a review for badly written books either. Be careful not to make any “this is AI” accusations unless you have irrefutable proof. Instead, point out the issues that affected your reading experience—endless repetitions, plot holes, things that didn’t make sense, or characters that felt flat, for example. No matter whether they are indeed signs of AI use or not, a review like that will help your fellow readers decide if the book is worth their time.

LEAVE A COMMENT

What do you think about the survey results? Do they reflect your own thoughts and experiences? Please scroll to the bottom of this page and leave a comment!