Steven Johnson is a prolific non-fiction writer and thinker who has spent decades exploring how technology amplifies human thought, and who recently collaborated with Google Labs on Notebook LM, an AI-powered research and writing tool. This conversation covers his philosophy of writing structure, his creative process, his approach to editing and sentence-level craft, and the deeper intellectual themes that unite his body of work.
The surprising core skill behind working with AI
The most important skill for getting large language models to do useful things is not programming — it’s clear, persuasive prose.
Johnson was brought into the Notebook LM project partly because of his writing background, and he ended up writing many of the system’s prompts himself.
This reflects a broader shift: writing ability has become a critical tool for directing AI, a development few in the tech world anticipated.
Notebook LM: from search to genuine collaboration
Johnson has spent decades building a personal archive of digital quotes and notes from everything he read, originally just to make them searchable.
Notebook LM takes this further by letting users ask open-ended questions of their own source material — not just keyword searches, but queries like “what are the most surprising facts about dolphins in my reading notes?”
The tool is designed around the idea of grounding a language model in a curated body of sources, then letting the user focus the AI on specific passages, notes, or combinations of notes through a unified interface.
This avoids the cognitive load of switching between tabs and applications, keeping the writer in a flow state.
The system suggests follow-up questions, creating what Johnson calls “conversational hypertext” — a chain of exploration that wasn’t possible before.
Johnson describes the tool’s tagline, “from information to insight,” as capturing its core purpose: distilling large bodies of raw material into usable understanding, quickly.
Structure as invisible architecture
Johnson thinks deeply about structure, and argues that when it works, readers don’t notice it — it simply feels right.
He compares it to a well-designed door: you only notice bad design when it fights you.
Once the right structure is found, many other decisions in a piece fall into place.
The Ghost Map: structure disguised as chronology
In The Ghost Map, about a cholera outbreak in 1850s London, Johnson structured each chapter as a single day of the outbreak.
This allowed him to riff on a different thematic idea each day — waste recycling, public health, information design — in a way that felt organic to the events.
The structure was widely praised by reviewers, but no one ever explicitly mentioned the chapter-per-day architecture, which for Johnson was the point.
Enemy of All Mankind: structure preceding content
For Enemy of All Mankind, Johnson had the structure before he had the subject: he wanted a book that begins with a dramatic, compressed event, then goes back centuries to explain the forces that led to it, converges on the event, and then traces downstream consequences — an hourglass shape.
He sat on this structural idea for years before discovering the story of pirate Henry Every, which fit perfectly.
This illustrates a broader pattern: Johnson often develops structural ideas long before he knows what content will fill them.
The Infernal Machine: cinematic structure on the page
The Infernal Machine weaves together the birth of forensic science, anarchist bombings in early 20th-century New York, the invention of dynamite, and the origins of the FBI.
Johnson opens with a bomb going off in an NYPD basement in 1915, then pulls back to follow seemingly unrelated characters across Russia, London, Paris, and New York before they converge.
The preface is written to create a cinematic, thriller-like tension — pulling the reader “down the stairs” toward a suspicious briefcase — to signal the kind of narrative architecture the whole book will use.
The spark file: capturing hunches before you understand them
Johnson’s creative process centers on what he calls a “spark file” — a single, ever-growing document (eventually 120,000+ words) where he dumps every interesting quote, story idea, hunch, or observation.
He revisits it periodically, looking for connections between old hunches and new ideas.
The key principle is frictionless capture: don’t over-intellectualize or tag ideas as they come, just write them down.
This is inspired by his research into how innovation actually works — not as sudden eureka moments, but as slow hunches that incubate and connect over years.
He is now exploring how Notebook LM can augment this process, using AI to surface connections between old notes that he might not find on his own.
Editing: fighting your own familiarity
Johnson is skeptical of over-editing during drafting, especially for long projects.
His key technique: avoid rereading what you’ve written each day. Instead, leave yourself a short note about where to pick up, then move forward.
This prevents the text from becoming so familiar that you can no longer read it with fresh eyes.
By the time he finishes a book and reads the whole thing, there are sections he’s genuinely forgotten — which lets him experience it more like a new reader would.
He also changes the format and location when doing a final read-through (iPad instead of desk, different typeface) to create psychological distance.
Writing at the level of sentences
Johnson describes his sentence-level skills as functional rather than virtuosic — he prioritizes clarity and intelligibility.
He sometimes struggles to make sentences aesthetically interesting, particularly in books where he wants a more novelistic feel.
He is excited about a feature he calls “paragraph thesaurus” — an AI tool that rewrites a paragraph while preserving its meaning, offering different metaphors and phrasings.
The value isn’t in using the AI’s output directly, but in breaking out of habitual rhythms and seeing the idea from new angles — like a Cubist painting showing multiple perspectives at once.
He recounts testing a similar tool on a bland paragraph about Hawaii’s climate, which returned an over-the-top result; when he typed “dude that is a little over the top,” the model understood the feedback and produced a better version — a striking example of how natural language has become a legitimate interface for creative collaboration.
Long sentences as set pieces
Johnson is a self-described “recovering long sentence writer” who had to unlearn the baroque, parenthetical style of his academic training in critical theory and literary criticism.
He still sees value in the glorious long sentence as a tool, and is interested in how a writer can signal to the reader that a sentence is going to be a moment — building from short, punchy Anglo-Saxon words through increasingly romantic, French-influenced prose toward a crescendo.
He compares this to a long tracking shot in film: the reader senses something ambitious is happening and stays with it.
Articles vs. books: different headroom, similar thinking
Johnson has written several 10,000-word pieces for The New York Times Magazine, a format he loves because it allows complex, multi-threaded storytelling without the pressure of filling a 60,000-word book.
The real structural difference is between short-form (1,500 words) and long-form: in short pieces, you’re limited to making a single point with one story. At 10,000 or 100,000 words, you can weave together multiple threads and disciplines.
For magazine pieces, he sometimes uses non-chronological structures — like opening The Man Who Broke the World with the death of Thomas Midgley Jr. (strangled by a pulley system he invented to cope with polio), then working backward and forward from there.
Learning from other writers
Johnson consciously studies other writers’ techniques when he’s preparing to do something new.
Before writing The Ghost Map (his first historical narrative), he studied Erik Larson’s Devil in the White City to understand how to recreate a city in time within a thriller structure.
Before The Infernal Machine (which uses extensive dialogue drawn from historical records), he revisited Larson’s Splendid and the Vile to see how diary entries could be used to create novelistic scenes in non-fiction.
He also learned from Malcolm Gladwell’s technique of ending sections with very short, punchy sentences — something he’d never done in his academic writing.
The synthesis across Johnson’s work
Johnson’s books fall into two strains: narrative non-fiction (true crime, historical drama) and books about innovation and creativity.
The common theme across both is thinking across disciplines and across scales of time to uncover the underlying forces behind events.
To understand anarchist bombings, you need chemistry (dynamite), economics, political philosophy, and individual biography.
To understand a pirate’s heist in 1695, you need to go back centuries to explain how India became the richest country in the world.
When this works, it produces two kinds of surprise at once: the page-turner surprise of narrative tension, and the intellectual surprise of seeing unexpected connections between distant fields and eras.
