Disclosure: this article contains affiliate links. If you buy through them we may earn a commission at no additional cost to you. Our recommendations are based on the evidence recorded alongside this article, and prices were verified on the date shown.

What changed and why it matters

New Scientist published a piece on August 18, 2026 under the headline "Feeding our books into generative AI risks creating a cultural void", putting a name to a worry that's been building for years around AI training pipelines that ingest published books. The headline is the news hook: a mainstream science publication is now treating book-to-AI training as a story with cultural stakes, not just a copyright dispute between publishers and AI labs.

That framing matters because the decisions about which books get scanned, licensed, and folded into training data are being made now, by a small number of companies, largely out of public view. A reader searching for this headline today is looking for a plain explanation of what's actually being claimed — not a legal summary and not a hot take.

The detail

"Feeding books into generative AI" describes the practice of using the text of published books as training material for large language models — the systems behind AI chatbots and writing tools. Books are attractive training material because they're long, edited, and information-dense compared to scraped web text, which is part of why AI labs have sought access to large book collections in the first place.

The "cultural void" in the headline names a downstream risk rather than a measured outcome. The concern is structural: if the same relatively narrow set of books keeps getting reused as training data — because it's already digitized, already licensed, or already cleared of rights disputes — then the language, references, and worldview an AI system reproduces could narrow along with it. Books that are out of print, never digitized, written in less-represented languages, or stuck in unresolved rights disputes risk becoming invisible to anyone whose first stop is an AI answer rather than a library, a bookstore, or a human recommendation.

It's worth being precise about what this piece can and can't tell you. The headline flags a mechanism — a narrowing of source material feeding AI outputs — without this article asserting how far that narrowing has already progressed, because that level of detail isn't part of the record cited here. Treat "risk" as exactly that: a plausible future outcome worth watching, not a completed one.

Who this affects most directly:

• Authors and publishers, whose backlist and rights decisions now double as decisions about what shapes future AI output.
• Readers and researchers, who increasingly meet books secondhand, through an AI summary or answer, rather than the text itself.
• Teams building AI products, who inherit whatever gaps exist in the training data their models were built on.

What it means for you

If you build, ship, or evaluate AI-powered products — rather than just read about the debate — the practical takeaway mirrors what publishers are already wrestling with: know what's actually feeding the systems you depend on, and don't assume broad AI fluency means broad cultural coverage. A model can sound authoritative on a topic it has thin or skewed source material for.

That awareness has a second, more immediate implication: infrastructure decisions. Teams building anything AI-facing still have to decide where the product runs, how it scales past a prototype, and what that costs in practice. Before committing budget to an AI-tooling stack, you can [check current Vercel pricing](See current pricing at the merchant) to see what deploying and scaling an AI-facing product actually costs at the infrastructure layer.

Best for: teams already building or deploying AI-integrated products who need a hosting and deployment layer they can scale without re-platforming.

Skip it if: you're still validating whether you need an AI feature at all — that decision belongs upstream of any infrastructure commitment.

Pricing clarity matters beyond hosting bills, too. If you're comparing vendor quotes for AI or data-heavy work more broadly and can't yet tell whether a number is fair, the MI Measure Price Calculator: How to Choose walks through how to read a quote against a public price list before you commit.

The decision this piece leaves you with isn't whether to use AI tools — most readers already do. It's whether you're treating "the AI said so" as equivalent to "the culture said so." Those are different claims, and New Scientist's headline is a reminder to keep asking which books, and whose, shaped the answer you got.