A deep dive into the seismic shift reshaping how readers find books – and what publishers must do about it

“Keyword Searching is archaic in comparison to the capability of LLMs to unlock the context of natural language. It’s not an incremental improvement. It’s a step change” – Cameron Drew.

This isn’t hyperbole. It’s a warning wrapped in an opportunity.

The End of Guesswork

For two decades, publishers have played a tiresome game: anticipating which clumsy keyword strings – “historical fiction Tudor court” or “business leadership productivity hacks” – might match a reader’s intent. LLMs render this archaeological approach obsolete. They “understand” nuance, mood, and the vague ache of “something like Normal People but set in Berlin.” Readers no longer translate desire into search-engine syntax. They simply ask.

But the transformation runs deeper than convenience. Traditional search required users to compress complex literary desires into brutally reductive terms. A reader seeking a novel that captures the particular melancholy of midlife transition, wrapped in darkly comic observations about academia, set somewhere vaguely Scandinavian, would have to reverse-engineer that craving into “literary fiction professor Norway” – and hope. The mismatch between intent and expression meant countless perfect reader-book pairings never happened.

It’s not just for books. I’m typing this into the TNPS LinkedIn newsletter interface, but then will copy across to the TNPS WordPress-hosted blog. Nowadays I paste the final text into an AI and ask it to come up with a key-phrase, slug and a 156 character metadata description that may or may not find me a few extra readers out there in the cyber-wilderness. But just how much reach can a 156-character metadata description have? It probably worked well ten years ago, but we’re almost in 2026, and the times, they are a-changing.

The Voice Revolution: When Search Becomes Storytelling

Here’s what the keyword optimisation playbooks missed entirely: we’re entering an era where readers won’t type their queries at all.

Voice interfaces are rapidly becoming the default mode of AI interaction. And when readers speak their searches rather than type them, everything changes. Consider the difference between these two queries for the same desire:

Typed: “mystery small town secrets”

Spoken: “I’m looking for a mystery novel, something atmospheric where everyone in a small town seems to be hiding something, kind of like Big Little Lies meets The Wicker Man, but I want it to be genuinely unsettling, not cozy. And I’d prefer a British or Irish setting if possible.”

That spoken query is 53 words. The typed version? Three.

People are comfortable speaking at length in ways they’d never type. We naturally provide context, qualify our requests, make comparisons. A reader standing in their kitchen asking their smart speaker for recommendations will unselfconsciously describe mood, pacing preferences, even which books they tried but abandoned and why. This isn’t just more data – it’s better data, capturing the associative, meandering way we actually think about what we want to read.

For publishers, this shift has profound implications. Voice queries will surface books based on tonal qualities, narrative rhythm, even emotional resonance – attributes that keyword tagging could never adequately capture. The novel that evokes “that particular Sue Miller sadness” or has “the propulsive dread of early Tana French” becomes discoverable in ways the BISAC system never imagined.

Discovery Becomes Conversation

The implications ripple outward. Author websites, devoid of strategic keyword stuffing, might suddenly surface more often. The long tail lengthens: obscure backlist titles could be unearthed by chatbots that grasp their thematic echoes with contemporary bestsellers. A debut novel from 2014 that sank without trace might find its audience in 2025 because an LLM recognises its kinship with this year’s breakout hit.

Yet this intimacy comes at a cost. When a reader asks, “What should I read next?” and the LLM responds with a synopsis rather than a retailer link, might publishers face potential disintermediation.

The Luddite Fringe think so, although I use the word “think” in its loosest possible way, of course.

The Psychology of Reading: Why Synopses Sell Books

This brings us to perhaps the most misunderstood aspect of the LLM search anxiety: the bizarre notion that readers who receive a synopsis won’t buy the book.

Let’s be clear: this fear betrays a fundamental misunderstanding of why people read.

We Don’t Read for Information Transfer

If readers simply wanted narrative data, SparkNotes (other study guide brands are of course available – all published by, you guessed it, publishers) would have destroyed the publishing industry decades ago. Film synopses would eliminate cinema. Restaurant reviews would replace dining out. But humans are not information-processing machines seeking the most efficient route to story data. We read for the experience of reading.

Consider narrative non-fiction, which has exploded over the past two decades. Readers often know the outcome before opening the book. We know how World War II ended. We know who walked on the moon. We know Theranos collapsed. Yet we devour these stories because the journey matters more than the destination. The way Erik Larson builds suspense and Bill Bryson builds humour around historical events we can Google in seconds is precisely why we read Erik Larson and Bill Bryson.

Fiction Readers Seek Immersion, Not Plot Points

For fiction, the case is even stronger. Literary readers reread favorite novels multiple times – with full knowledge of every plot twist. Genre readers often consume entire series where narrative beats become predictable. Mystery enthusiasts know the detective will solve the case. Romance readers expect a happily-ever-after. They read anyway.

Why? Because reading fiction offers:

  • Linguistic pleasure: The sentence-level craft that can’t be summarised
  • Emotional regulation: The controlled experience of feelings in a safe space
  • Cognitive transportation: The neurological state of being absorbed in a fictional world
  • Parasocial connection: The relationship with characters that develops over hundreds of pages
  • Identity exploration: Trying on different perspectives and lives
  • The reading experience itself: The ritual, the escape, the temporal space carved out from daily demands

A synopsis can tell you that The Goldfinch involves a boy, a painting, and a bombing. It cannot replicate the experience of Donna Tartt’s maximalist prose, the slow-burn revelation of character, the precise way she builds beauty from trauma.

The Synopsis as Sales Tool

Here’s the counterintuitive reality: detailed synopses likely increase sales. When an LLM provides a rich, contextual summary that resonates with a reader’s query, it’s not satisfying their desire – it’s igniting it. The synopsis functions as an extended recommendation, a curatorial act that says “this book contains what you’re seeking.”

Readers who get excited by synopses want more. They want to inhabit the world the synopsis sketches. They want the full texture of the experience. The synopsis is a movie trailer, not the movie. And effective trailers sell tickets.

The publishers who fear synopses are the same ones who once feared Amazon’s “Look Inside” feature, who resisted ebooks for fear of piracy, who worried bookstore browsing would kill mail-order catalogues. Every generation of publishing has its ghost stories about technology that will end reading. Every generation has been wrong.

Metadata’s New Metamorphosis

Traditional metadata – BISAC codes, blurbs, keywords – was designed for machines that matched, not understood. LLMs demand richer, narrative-driven context. Publisher catalogues must evolve into semantic goldmines: author Q&As, thematic essays, even character backstories become discoverable assets. The ISBN is no longer enough; the story about the story matters.

But what does this actually look like in practice?

From Taxonomies to Topographies

The old model treated books as points on a grid, intersecting categories like “Mystery” and “Historical.” The new model treats them as territories in a landscape, with varying terrain, microclimates, and uncharted regions. LLMs can navigate topology, not just taxonomy.

This means publishers should be creating:

Comparative context: “Readers who loved X will find echoes of Y in this book because…” – but articulated with specificity, not marketing pablum.

Tonal mapping: Not just “dark” or “uplifting,” but granular emotional texture. Does the darkness come from dread or from nihilistic humor? Is the uplift earned through suffering or granted through whimsy?

Stylistic fingerprints: Sentence rhythm, vocabulary range, use of metaphor, narrative distance. These are discoverable attributes for LLMs trained on text.

Reader experience indicators: Pacing (slow-burn vs. propulsive), difficulty level, whether it rewards rereading, ideal reading environment.

Cultural positioning: Not where it sits in the market, but where it sits in the conversation. What is it in dialogue with? What traditions does it honour or subvert?

The Author as Metadata Generator

Forward-thinking publishers are already beginning to treat author interviews, reading guides, and other editorial content not as marketing afterthoughts but as core metadata. When Ann Patchett discusses her writing process, she’s generating searchable context. When an editor explains why they acquired a manuscript, they’re creating discovery assets.

This content needs to be:

  • Structurally tagged for machine reading
  • Hosted in accessible, crawlable locations
  • Rich enough to provide genuine insight
  • Authentic rather than SEO-optimised to the point of meaninglessness

Threat or Collaborator?

The existential question looms at least in the minds of the Flat Earthers and the Luddite Fringe, that thrive on existential crises and falling skies. Will LLMs become direct competitors, synthesising summaries that satisfy queries without selling books? Or will they (Luddite Fringe, look away now!) partner with publishers, licensing editorially curated content to ground their models in authoritative voice?

The answer likely lies in the middle: a complex, evolving relationship where LLMs surface books that readers then buy, while publishers provide the rich contextual layer that helps LLMs make better recommendations. Think of it as curated discovery at scale.

The Amazon Precedent

Amazon, after all, didn’t destroy bookselling by replacing it – they transformed it by making discovery easier and purchase frictionless. Yes, independent bookstores suffered. Yes, the industry transformed painfully. But more books are sold now than in the pre-Amazon era. More readers have access to more diverse voices. More publishers have access more consumers across more formats. The long tail exists because Amazon made it browsable.

LLM search could follow a similar trajectory: disrupting existing power structures and discovery mechanisms while ultimately expanding the overall market by connecting more readers to more books.

The critical variable is whether publishers position themselves as essential partners in this transformation, or as obstacles to be routed around.

The New Gatekeepers: Or Are They?

One underexplored dimension of LLM search is the question of curation and bias. Traditional search engines had their biases – recency, popularity, commercial arrangements – but they were relatively transparent. You could game them, which meant you could also understand them.

LLM recommendations are opaque. We don’t know why one book surfaces over another. Training data, model architecture, fine-tuning choices, corporate partnerships – all of these invisibly shape what readers discover. This should concern publishers.

But it should also energise them. In a landscape where algorithmic opacity reigns, authentic editorial voice becomes differentiating. Publishers who develop reputations for genuine curation – who build direct relationships with readers through newsletters, podcasts, and community spaces – create discovery channels that don’t depend on LLM favour.

The future may be less about optimising for AI and more about building human trust that exists parallel to AI. Let the LLMs handle broad discovery; publishers handle depth and discernment.

The Path Forward: Eight Strategic Imperatives

This step change demands that publishers stop optimising for robots and start feeding them meaning. Specifically:

1. Audit and enrich existing metadata: Every title in the catalogue deserves more than BISAC codes. Commission brief essays on theme, tone, and context.

2. Create discovery-friendly editorial content: Interviews, essays, reading guides – all properly tagged and hosted where LLMs can access them.

3. Experiment with LLM partnerships: Approach OpenAI, Anthropic, Perplexity. Offer curated content in exchange for fair attribution and commercial arrangements. For smaller publishers, there are numerous new outfits providing a means to connect small publishers with the needs of the AI giants.

4. Invest in voice-optimised presence: As queries become spoken, ensure your books can be pronounced correctly and that long-form spoken descriptions exist.

5. Build direct reader relationships: Newsletters, communities, subscription models. Don’t depend entirely on intermediated discovery.

6. Train your teams in LLM literacy: Editors, marketers, and metadata specialists need to understand how these systems work – and how to feed them effectively.

7. Advocate for transparency: Push for industry standards around LLM recommendation explainability. Readers deserve to know why they’re seeing certain books.

8. Embrace the synopsis: Provide official, rich, detailed synopses rather than leaving LLMs to generate their own. Control the narrative about your books.

The Philosophical Stakes

Beneath all the tactical considerations lies a deeper question: What does it mean to be a publisher when discovery is conversational and AI-mediated?

If publishing’s core function is matching readers with books they’ll love, then LLMs are potentially the most powerful tool ever created for that purpose. They can surface connections human curators might miss. They can serve niche audiences at scale. They can make serendipitous discovery feel personal.

But if publishing is about more than matching – if it’s about cultural curation, about saying “this matters” rather than “you might like this” – then publishers need to articulate that value clearly. The AI can optimise for satisfaction. Only humans can optimise for meaning.

The publishers who thrive in the LLM era will be those who understand they’re not in the keyword business, or even the book business. They’re in the meaning-making business. And meaning, thankfully, still requires humans.

Most Crucially: Reclaim the Conversation

Because in a world where discovery is dialogue, the most compelling voice wins.

The keyword is dead. The story – about books, about reading, about why any of this matters – is more alive than ever. Publishers who can tell that story, in ways both humans and machines can understand, won’t just survive the LLM revolution.

They’ll lead it.

This post first appeared in the TNPS LinkedIn newsletter.