Libby adds AI content filters as ebook platforms brace for a flood of synthetic books

Libby, the popular ebook lending app used by thousands of public libraries, is preparing to give readers a new choice: whether they want to see AI-generated books, narration, translations, and artwork in their library results. The move is the clearest sign yet that the public-library ebook ecosystem is trying to adapt to a publishing landscape increasingly shaped by machine-made content.

The new controls, which will appear in OverDrive’s Libby app settings, are meant to help readers and librarians sort through a growing volume of AI-assisted and fully synthetic titles. But the system comes with a major limitation: OverDrive is not planning to run its own AI detector to label books. Instead, the company says it will depend on publishers and distribution partners to identify AI content through standardized metadata.

That approach reflects a practical reality and a larger industry debate. AI-generated books are multiplying fast, but the tools for reliably identifying them remain imperfect. As a result, Libby’s new filters may help users avoid certain kinds of content, but only if the relevant publishers accurately disclose how a book was made.

Libby’s AI controls arrive as publishing changes fast

OverDrive’s new chief executive, Marc DeBevoise, has made it clear that the company sees artificial intelligence as unavoidable. He described AI as a major new phase for the business, while also arguing that the technology can bring benefits when used carefully.

For OverDrive, the challenge is to balance openness with transparency. The company wants to preserve the ability for libraries and readers to decide what they are comfortable accessing, while also avoiding a blanket rejection of AI-based tools that may improve discoverability, localization, and accessibility.

DeBevoise said the company wants users to know what is available and how it was produced, framing AI as something that must be disclosed rather than hidden.

The new filters are expected to cover several categories at once:

• AI-authored books
• AI-narrated audiobooks
• Machine translations
• AI-generated artwork or illustrated elements

That scope matters because AI is not just changing how books are written. It is also altering how they are translated, narrated, packaged, and distributed. For library readers browsing a catalog, those distinctions can be important, especially when they are trying to avoid synthetic voices or machine-translated literary texts.

How the Libby filter system is supposed to work

OverDrive is taking a metadata-first approach. Rather than scanning titles with an automated detection tool, it will rely on labels supplied by publishers and by the services that feed titles into its catalog.

That decision may sound modest, but it is strategically significant. AI detection systems are notoriously unreliable, especially when applied to long-form creative works. They can produce false positives, miss heavily edited material, or fail to account for mixed human-and-machine workflows.

By contrast, metadata labeling is cleaner in theory because it places responsibility on the source. The drawback is obvious: if a publisher does not disclose the role of AI, the label may never appear.

Why OverDrive is not using an AI detector

The company’s choice suggests caution. A false label could create distrust among authors and readers, while an inaccurate exclusion tool could suppress legitimate work. For a public-library platform that serves schools, universities, and local communities, the stakes are especially high.

Libby’s system is therefore less of an AI fingerprinting tool and more of a disclosure framework. Readers who want to avoid AI content will be able to toggle it off in settings, but the effectiveness of that setting will depend heavily on whether publishers and distributors comply.

That may be enough for now. It is also a sign that the publishing industry is moving toward a world in which AI labeling becomes part of the cataloging process, much like genre tags, language tags, and format indicators.

A catalog built before the AI boom now faces new pressures

OverDrive’s library network is enormous and deeply established. The company, which dates back four decades, originally helped digitize books for distribution on floppy disks and CD-ROMs before moving into ebooks in the early 2000s. Libby, its consumer app, launched in 2017 and has since become the most recognizable library-reading platform in many markets.

Today, OverDrive says it works with 92,000 public libraries, schools, and universities in more than 115 countries. Its catalog includes over 6 million books, and those titles have been borrowed more than a billion times.

Most of that catalog predates the modern generative-AI era. That matters because the vast majority of older titles were created long before large language models became widespread. In other words, most of what readers find in Libby was not produced in an AI-assisted environment.

DeBevoise argued that works published before the recent wave of generative AI are, by definition, not AI-generated, and that the bulk of the catalog remains outside the current debate.

But the catalog is not static. As more AI-assisted books enter the supply chain, even a historically human-made library platform may increasingly have to decide what to disclose, what to filter, and what to promote.

The bigger market is already struggling with AI slop

OverDrive is not facing this issue in isolation. Online book marketplaces and ebook retailers have been wrestling with a surge in low-quality, mass-produced content that is often described as AI slop.

Amazon, which offers a direct self-publishing pipeline through Kindle Direct Publishing, started limiting the number of books authors can upload per day in 2023 as part of an effort to slow floods of suspicious or repetitive titles. More recently, Kobo CEO Michael Tamblyn said the company rejects nearly half of self-published submissions because of AI-related concerns.

Tamblyn recently described the situation as an overwhelming stream of content, warning that publishers and storefronts are dealing with a torrent rather than a trickle.

That same pressure is now reaching library platforms. Even though OverDrive does not let authors upload books directly in the same way Amazon or Kobo do, it still depends on distribution channels that aggregate self-published titles and deliver them to multiple storefronts at once.

Why self-publishing intermediaries matter

A major example is Draft2Digital, which supplies books to several ebook marketplaces, including Apple Books and Google Play Books, and also feeds titles into the broader digital distribution ecosystem. According to the company’s policies, AI-generated books are permitted if they have undergone extensive human editing.

That kind of rule makes the spread of AI content hard to prevent entirely. A book may pass through several stages of human review and still include machine-generated passages, translations, narration, or illustrations. By the time it reaches a public-library app, its origins may be difficult to untangle without explicit disclosure.

For OverDrive, that is the central problem: it can manage access at the interface level, but it cannot fully police the production chain behind every title.

AI content controls aim for a middle path

OverDrive’s strategy is not to ban AI-assisted books outright. Instead, it is trying to offer filters so readers and librarians can choose their own comfort level.

That middle path is important in a library context. Public libraries are expected to provide broad access to information, not to impose a single editorial standard on what readers may borrow. A filter that can hide AI content on request is therefore less restrictive than a total ban and more flexible than a one-size-fits-all policy.

At the same time, the company appears to believe AI can be useful in specific areas, especially where it reduces costs or broadens access.

DeBevoise argued that AI can improve access to information and content when deployed responsibly, particularly in ways that help people find, understand, or localize material.

That framing suggests OverDrive is not treating AI as a threat to be eliminated. Rather, it is treating AI as a variable to be managed.

Discovery and recommendation tools are part of the equation

Libby first introduced some AI-powered features last year to improve book discovery, though those efforts drew criticism. That backlash underlines a recurring theme across the tech sector: users are often receptive to AI only when it is clearly useful, unobtrusive, and transparent.

In the library context, recommendations matter because readers rely on discoverability. If AI helps surface similar titles, improves search relevance, or makes metadata easier to navigate, it could become a practical part of the platform. But if AI appears to nudge readers toward opaque or low-quality content, trust can erode quickly.

OverDrive seems to be trying to preserve the upside while minimizing the discomfort.

Audiobooks are the strongest case for AI, and the biggest concern

One area where OverDrive sees real promise is audiobook localization. DeBevoise says AI may help make content available in more languages, allowing domestic books to travel internationally and international books to reach more local audiences.

That argument rests on a real economic hurdle. Producing high-quality human-narrated audiobooks in dozens of languages is expensive and time-consuming. Synthetic narration could make it possible to expand access far more quickly.

But OverDrive is not presenting machine narration as a full replacement for human performance.

DeBevoise said he still prefers audiobooks read by voice actors, while acknowledging that recording the same title in many languages can become prohibitively costly.

That is the practical tension at the heart of the issue. Human narration can carry nuance, emotion, and performance quality that synthetic voices still struggle to match. Yet the cost of human production can sharply limit multilingual availability, especially for smaller publishers and niche titles.

Audiobooks now dominate usage

Libby’s usage data helps explain why OverDrive is paying so much attention to audio. Although audiobooks account for only about 15 percent of the app’s catalog, they now generate roughly half of all usage.

That imbalance shows just how central audio has become to digital reading habits. For many users, listening is the preferred way to consume books, whether because of convenience, accessibility, or personal preference. It also means that any AI changes affecting audiobooks could have a disproportionate impact on reader behavior.

If AI narration becomes more common, readers may face a catalog split between human-read and machine-read titles. Libby’s filters are intended to help manage that split, but they will also reveal how much demand exists for each format.

Machine translation is useful, but not uncomplicated

Machine translation is another area where the promise of AI collides with quality concerns. For nonfiction, educational material, or straightforward genre publishing, automated translation may be a huge access booster. For literary fiction, however, nuance, voice, and cultural context can be lost.

Libby’s new tools will allow users to hide machine-translated books as well, which suggests the company understands that not all AI use cases should be treated the same way.

That said, the setting only works if the books are labeled properly. A translated novel that is not tagged as machine-translated could still appear in results, even if a user has asked to avoid that category. The feature is therefore only as strong as the metadata pipeline behind it.

Why literary translation remains controversial

Translators and editors have long argued that language is more than word substitution. It carries tone, voice, cultural reference, and subtle authorial intent. AI systems can produce readable output, but readers and critics may still see translated literature as an area where human judgment matters enormously.

That concern makes library filters useful, but also raises a broader question: should machine translation be disclosed in the same way as AI authorship? OverDrive’s answer appears to be yes.

The company is treating authorship, narration, illustration, and translation as part of the same transparency problem.

What this means for libraries, publishers, and readers

The immediate impact of Libby’s new controls will likely depend on how publishers respond. If AI labeling becomes routine, the feature could become a straightforward way for users to personalize their library experience. If publishers are inconsistent, however, the filter may create only a partial sense of control.

For libraries, the change could be useful in helping them respond to patron concerns without restricting access to content wholesale. A librarian can support broad availability while still recognizing that some patrons want to avoid synthetic media.

For publishers, the shift increases the pressure to disclose how a title was made. That may ultimately become a competitive issue as well as an ethical one. Readers who value human-made work may start looking for transparency the same way they look for cover art, format, or narrator information.

And for readers, the benefit is simple: more control over what shows up in the catalog.

• Readers can opt out of AI-generated content where it has been labeled.
• Librarians can serve different user preferences without removing titles entirely.
• Publishers are pushed toward clearer disclosure practices.
• AI-assisted localization and narration may expand access for multilingual audiences.

Why metadata may become the new battleground

The most important part of this story may not be the user-facing toggle in Libby, but the metadata behind it. In the digital book world, metadata determines whether a title can be discovered, categorized, translated, recommended, and now filtered for AI origins.

If standardized metadata becomes the industry norm, AI disclosure could eventually become as routine as page count or language. But if the ecosystem remains fragmented, users will continue to rely on incomplete labels and inconsistent reporting.

That would create a split reality: an app designed to filter AI content, operating inside a supply chain where AI content is often difficult to verify.

In that sense, OverDrive’s new controls are both a user feature and a signal to the industry. They indicate that AI disclosure is no longer a niche concern. It is becoming part of the basic infrastructure of digital publishing.

OverDrive’s bet on transparency over detection

For now, OverDrive is betting that transparency will work better than surveillance. Instead of trying to guess whether a book was made with AI, it is asking the people who publish and distribute books to say so themselves.

That may prove to be a sensible compromise. It avoids overreliance on flawed detection tools, while giving readers a way to shape their own library experience. But it also leaves a gap that only the wider publishing industry can fill.

The real test will come as AI-generated and AI-assisted titles become more common. If labeling becomes normalized, Libby’s filters may be an early model for how library platforms handle synthetic media. If not, the company may find that its new controls are only a partial defense against a rapidly expanding catalog of machine-made content.

Either way, the message from OverDrive is clear: AI is not a hypothetical future problem for library ebooks. It is already here, and the platform is now building tools to manage it.

Timeline: how OverDrive and Libby got here

As the ebook world absorbs the effects of generative AI, Libby’s new tools show how even a public-library app built around access and trust must now confront the question of provenance. The answer, at least for now, is not perfect detection. It is disclosure, filtering, and a hope that the industry will label itself honestly.

superintelligencenews@gmail.com