The Digital Librarian

Generative AI is reshaping the publishing industry—and libraries are beginning to notice the impact on our collections. AI-generated books are flooding the marketplace, and AI audiobook narration is increasingly prevalent, sometimes with the assistance of established digital publishing platforms. As the world’s largest online book retailer, Amazon has been negatively impacted by an upsurge in “scam” materials that are meant to imitate popular, highly anticipated releases (think of the book version of mockbuster films), and had to remove nonfiction materials offering incorrect, potentially life-threatening advice! Given how central our collections are to the library’s mission, these industry developments demand our attention.

Why it Matters:

In February 2025, 404 Media published an expose titled “AI-Generated Slop Is Already In Your Public Library,” confirming the suspicions of many library professionals; that AI generated content is already negatively impacting our collections. The article outlined the pervasiveness of low-quality, AI-generated titles in library digital platforms, including Hoopla and Overdrive.

In the case of Hoopla, the problem is compounded by two key factors: libraries purchase access to a broad, pre-curated collection they do not select themselves, and they pay per item borrowed. As a result, materials may not meet a library’s acquisition standards, yet increasingly tight budgets are being spent on potentially low-quality, low-interest titles. Additionally, the sheer volume of substandard content can overwhelm our collections, making it harder for users to discover high-quality materials and diminishing the overall browsing experience.

The insinuation of AI-generated materials has not spared library audiobook collections either. In late 2024, the website Smart Bitches, Trashy Books explored the rise of AI-narrated titles on Overdrive in the article “AI Audiobook Narrators in OverDrive and the Issue of Library AI Circulation Policy.” In it, a librarian shares their experience stumbling upon a book narrated by “Scarlett Synthesized Voice,” only to identify more than 100 AI narrated titles upon further investigation! The piece notes the overall lack of policy governing AI-generated content within library collections.

So Let’s Talk Policy!

In considering a policy response to generative AI, we should ask ourselves several questions:

What is the intent of my policy? What am I trying to achieve?

Such goals may include:

• Quality control: avoiding the addition of low-quality/inaccurate additions to the library collection. Does the library distinguish between fiction and non-fiction? An AI hallucination within a piece of fiction may be annoying, but harmless. An AI hallucination within a medical text could have catastrophic consequences!

• Transparency: Does the library disclose AI “authorship?”. I say disclose, because AI-generated titles often employ a pseudonym to mask the AI origins.  Will you label both physical/digital items to this end? Do you locate these titles in separate collections?

• Prohibition: The library has determined that AI-generated content is irreconcilable with the library’s mission and values, and will not add such content to its collection.

Do I need a new policy, or can I view AI through an existing policy lens?

Libraries all have (or should have) an existing collection development policy. These policies have selection criteria that speak to quality, factual accuracy, and diversity viewpoint. They may require, or at least encourage, that materials are selected based upon reviews from an accepted review source. Applying this criteria will give you some rules of engagement, and may very well exclude a great deal of AI-generated content. I find that much of the language that is used to address self-published titles is often applicable in this new context.

Is the policy enforceable?

When we set out policy goals, we should ask ourselves, are they achievable? AI content can be notoriously difficult to identify until after it is already acquired. This would make prohibition or even consistent labeling difficult. Here an absolutist stance might be less accurate than one where the library policy states it will make every attempt to acquire human-authored content, or to disclose a title’s AI origins. It is also worth asking yourself how the library will respond to patron requests for AI-generated content within the broader context of your collection development policy.

Am I enforcing my policy as written?

Consistency is important! For example, if your policy rationale for excluding AI-narrated audiobooks is rooted in quality assurance, then you should be prepared to add AI-generated content when and where the quality is good, assuming it otherwise meets your selection criteria! If you’re finding that adherence to library policy is problematic, then it’s the policy that must change, not your enforcement of it!

Looking at Sample Policies

Prohibition on AI-Generated Materials

The Jasper-Dubois County Public Library in Indiana has a specific AI Generated Materials Policy within their overall in collection development policy. It states

“The library is committed to maintaining a collection that reflects human creativity, knowledge, and expertise. As such, the library does not purchase, acquire, or accept donations of books that are primarily generated, authored, or written by artificial intelligence. This policy ensures that our collection upholds quality, authenticity, and intellectual integrity for our community.”

Likewise, the General Criteria for Materials Selection of the Cranston Public Library in Rhode Island states simply:

“Works that are entirely written by or narrated by artificial intelligence (AI) will not be considered for the collection.”

Like many public libraries, a portion of Cranston’s digital collection is cooperatively purchased via a consortium, in this case the Ocean State Libraries (OSL) eZone. There is a separate collection development policy governing this eContent. It states

“OSL seeks to avoid all AI generated content in the eZone. All AI content that violates vendor policy of what is allowed in their marketplace will be reported to the vendor.”

Middle Ground?:  Distinguishing Between Selection Criteria and Weeding Criteria

The North Olympic Library System in Washington seeks to exclude AI-generated (included AI-narrated versions of human-authored) materials from their selection criteria, but does not automatically weed this content upon discovery within their collection. Instead, they subject it to their standard weeding criteria, and label it as AI generated within their catalog. Their policy also draws a distinction between AI-generated and AI-assisted works.

NOLS respects the intellectual property of human authors and creators. NOLS will make reasonable efforts not to purchase AI-generated content, or AI-generated audio editions of human-created works. AI-generated content inadvertently added to the collection will be labeled as such in the catalog record but will not be weeded unless it meets one or more criteria for weeding (poor circulation, damaged, superseded, etc.). AI-assisted content is permitted in the NOLS collection, subject to the same requirements and criteria as wholly human-authored works. AI-assisted content is work that is written by a human but for which the author used AI tools to edit, refine, or error-check the work. 

Allowed, with Designation:

The Kenosha Public Library in Wisconsin applies their selection criteria to human-authored and AI-generated materials equally, but states

“Artificial intelligence-generated content is designated accordingly and is not misrepresented as human-generated content.”

Vendor/Distributor Transparency is Essential!

Regardless of the specifics of your collection development policy, it is most effective when backed by transparency. This starts with publishers and extends to the vendors and distributors libraries rely on. These intermediaries hold significant purchasing power and can pressure publishers to disclose the use of AI. Likewise, libraries should leverage their collective financial influence to shape vendor practices. Notably, Hoopla has responded positively to library feedback following community concerns about AI-generated content.

To preserve the quality of our collections, libraries should insist that AI-generated titles are clearly identified during individual title-level acquisitions, and that platforms offering pre-curated collections provide tools to filter or exclude such content. This level of transparency should also extend to end users, enabling patrons to make informed choices while browsing the library’s collection.

Items of Interest

• I’ll be giving a virtual talk “In It Together: Libraries, AI, and the Power of Collaboration” at the 2025 Northwest Interlibrary Loan and Resource Sharing Conference on September 9. Registration is now open.

• On Thursday, October 16, I’ll be delivering a virtual keynoter for the University of the West Indies Mona Library’s LIS Symposium 2025. My talk will be “Our Age of Uncertainty: Libraries, AI, and the Information Landscape”.

• I’m honored to have been named a Library Journal 2025 Mover & Shaker.

• I’ll be delivering keynotes for the Pennsylvania Library Association and Michigan Library Association annual conferences. Always happy to connect in-person if you’ll be in attendance!

As always, if you’re looking for a speaker for your library event or staff development day, let’s talk! I cover the intersection of emerging technologies and libraries, keeping it conversational, informative, and entertaining. I am currently booking for 2026.