Libby’s AI filter will let readers opt out, if labels work

Libby is preparing AI content controls that will let readers choose whether they see AI-generated material in the app. The catch is that the filters depend on publishers accurately labeling AI authorship, AI-narrated audiobooks, machine translation, and AI-generated art.

WTF Index IDIOCRACY
◄ Terminator 0 Idiocracy 2 ►

The story focuses on AI-generated publishing content and unreliable labeling, a mild concern about quality and reader trust rather than danger or autonomy.

Libby’s AI filter will let readers opt out, if labels work

Libby is preparing a new way for readers and libraries to manage AI content, but the system will only be as reliable as the labels attached to each book.

OverDrive, the company behind the ebook lending app used through public libraries, is getting ready to introduce AI content controls inside Libby. The planned settings will let readers decide whether they want to see AI-generated content, including AI authorship, AI-narrated audiobooks, machine translation, and AI-generated art.

How Libby’s AI filter is supposed to work

The core idea is simple: readers would be able to choose in Libby’s settings whether AI-related content appears for them. That gives users more control without removing the material from the broader catalog.

Marc DeBevoise, the new CEO of OverDrive, frames the change as part of a larger shift in digital publishing. “AI is the new frontier for us,” he says.

The planned controls cover several kinds of AI involvement, not just books written by generative systems. A title could be affected because it uses machine translation, synthetic narration, AI-generated cover or interior art, or AI authorship.

That broader scope matters because AI is entering publishing in more than one place. A reader may be comfortable with a machine-translated title but not with a book generated by AI. Another may object to synthetic audiobook narration but not to AI-assisted discovery features. Libby’s approach is meant to expose those distinctions through content controls.

The label problem

The biggest limitation is built into the design. OverDrive has decided against using an AI checker to identify AI-generated books. Instead, it will rely on publishers self-labeling works through standardized metadata.

That means the filter is not a universal detector. It is a preference tool that depends on accurate information entering the catalog before the reader ever sees the title.

DeBevoise describes the goal as transparency: “We need to tell people what’s available [and] how it was created.” But that promise depends on whether the people and companies supplying books apply the right labels consistently.

The source article points to a clear consequence: Libby’s machine translation filter can only work if books are labeled correctly. The same logic applies to other AI categories. If an AI-narrated audiobook or AI-assisted title is not marked in the metadata, a reader’s opt-out setting may not catch it.

Why OverDrive is taking a middle path

OverDrive is not presenting AI as something to reject outright. DeBevoise argues that AI may help in areas such as content recommendations and localization, while still giving readers and librarians a way to avoid AI-generated material if they choose.

Libby already introduced some AI features for book discovery last year and later faced backlash. The new filters appear to be an attempt to separate AI as an internal discovery tool from AI as part of the books and audiobooks available to patrons.

OverDrive also starts from a catalog with a long history. The company was founded 40 years ago to digitize books for distribution on floppy disks and CD-ROMs. It began offering ebook lending with local libraries in the early 2000s and launched Libby as a consumer-facing app in 2017.

Today, OverDrive works with 92,000 public libraries, schools, and universities in over 115 countries. Libby’s catalog includes over 6 million books, and those books have been borrowed over a billion times.

Most of that catalog predates the recent wave of modern LLMs. DeBevoise puts it plainly: “Everything before 2020 [or] 2022 is by definition not AI.”

Self-publishing raises the stakes

The challenge is what happens next. The source article notes that Amazon began restricting the number of books self-published authors can upload per day in 2023 to combat AI slop. It also says Kobo CEO Michael Tamblyn revealed last month that Kobo rejects nearly half of all self-published books over AI concerns.

OverDrive is different because it does not let authors directly upload books in the same way Kobo and Amazon do. But it does work with self-publishing intermediary Draft to Digital, which also supplies self-published books to digital storefronts including Apple Books and Google Play Books.

Draft to Digital allows AI-generated books if they have gone through “extensive editing from a human.” That makes it likely that some AI titles will enter Libby’s catalog, even without a direct upload system.

For libraries and readers, the practical question is not whether AI books can be kept out completely. It is whether the catalog can describe them clearly enough for people to make informed choices.

Audiobooks show both the promise and the tension

Audiobooks are a major part of the issue because they are a major part of Libby usage. The source article says audiobooks make up only 15 percent of Libby’s catalog but account for roughly half of all usage in the app. DeBevoise calls audio “the modality of choice.”

That explains why AI-narrated audiobooks and localization are central to the debate. AI could help make works available in more languages, but DeBevoise still says he prefers audiobooks read by actual voice actors over fully synthetic narration.

The article also notes that AI critics have raised concerns about AI translation, especially for literary works. Libby’s planned controls acknowledge that concern by including an option to filter out machine translation.

The result is a compromise: Libby will not simply block AI content, and it will not rely on automated detection to classify it. Instead, OverDrive is building reader controls around publisher-supplied metadata. That gives readers a choice, but it leaves trust, accuracy, and enforcement at the center of the system.