When we think of writing a book, a few images might come to mind: scribbling furiously by candlelight or staring at a blank page with no outline, like Stephen King, before finally bringing the fruits of one’s labor to a publisher, who holds the keys to the gates of literary acceptance. What doesn’t fit neatly into that vision is the role of the algorithm — the digital machinery that determines who sees what content, shaped by preferences and habits that are often unconscious.
While the omnipresent effects of the mysterious and vague algorithm are obvious in certain forms of media — for example, determining which videos are shown to social media users, which songs play next when playlists end, and which headlines rise to the top of Apple News — its implications for book publishing are less obvious. Books may seem, on the surface, to be the last bastion of traditional media. While some books may trend on “BookTok,” the TikTok subcommunity dedicated to book-lovers, or be prominently featured on Goodreads or Audible, the publishing industry has seemingly avoided being revolutionized in the same way as music, television, and film. Increasingly, this is an illusion.
As far back as 2012, the Wall Street Journal reported that e-books were enabling publishers to access in-depth data on how readers engage with books: which books get finished, how quickly they are being read, and even what kinds of love interests readers prefer. This gave publishing companies new insights into how to market existing works and what kinds of feedback to give authors. But what happens when decisions about what to publish are not based on human judgment, but market data?
Today, data analytics are frequently used to predict trends in order to shape published content to anticipated demand — not only at literary giants like HarperCollins, but also at scholarly publications like Elsevier. There is absolutely an argument to be made that this approach is not only efficient but also necessary to save what might otherwise be a threatened industry. After all, what’s wrong with publishing the kinds of books that people actually want to read? As literary agent Mark Gottlieb writes on his blog, “Analyzing sales data, online reviews, social media mentions, and reader demographics…enables [publishers] to make strategic decisions when acquiring and developing new titles, increasing the chances of success by aligning content with reader demand.”
While privacy concerns have been raised about the information collection required to facilitate data analytics, the potential stifling of literary expression is perhaps an even greater concern. By definition, data-driven publishing uses information about what has been popular to predict what will be popular, meaning that new, revolutionary works will have trouble breaking through. And while individual publishers could still theoretically choose to make risky decisions in a business model that prioritizes market data over human instinct, it is not hard to imagine a world in which the first round of rejections is made by artificial intelligence. Even as it is, the near-universal requirement of sending a query letter explaining how a text fits what a publisher is already looking for may dissuade authors from submitting works that contradict the data.
Beyond this, there is a broader question of whether authors should tailor their writing to maximize reader engagement at all. Important works of art rarely reflect what is already popular, but a data-driven approach offers little incentive to be ahead of one’s time. This is especially salient given that the book marketplace is heavily saturated, with over two million books self-published each year. The lure of traditional publishing, with its promise of advances, marketing, and potential adaptations, may incentivize authors to steer clear of risky stories to begin with. This goes further than avoiding controversy. Publishers may encourage authors to write for the average reader — based on aggregated data that shows, for instance, that romance-novel readers prefer love interests with brown hair, green eyes, and moderate chest hair. Such a model encourages publishers to choose a book everyone will finish, rather than a book that will be one person’s favorite. The amount of enjoyment any given reader gets out of a book matters little from a business perspective so long as they get to the end — unless there’s potential for a franchise. While individual books may see increased engagement, works that may have become classics — especially literary and nonfiction pieces, which take longer to read and are less likely to be finished — might never see the light of day.
Disincentivizing originality gives large language models an advantage over human authors. The HarperCollins–Microsoft artificial intelligence-training deal may produce books that perfectly fit the blueprint of what HarperCollins’ existing reader data suggests will be successful next, precisely because artificial intelligence can generate books that include only what’s already working. Human authors, on the other hand, may not be able to resist adding their own creative voice, unsupported by data as it may be.
The revolution in book publishing is currently in progress, but the extent of its reach and the intensity of its impact remain to be seen. As readers, we have the power to shape our literary landscape, whether by supporting independent, print-based publishers or by searching for books off the beaten path. And as writers, we have an opportunity to prioritize integrity over marketability, since humans may not be necessary to achieve the latter.
