When three LLMs tried to write my next book, only one captured the human behind it.
There’s a particular kind of madness that comes from giving three different artificial minds the same set of instructions and asking them to write the opening chapter of your next book. It’s like handing identical sheet music to three musicians and discovering that one has played jazz, one has played Bach, and one has played something that sounds suspiciously like a TED Talk on mushrooms.
That, roughly, was my Tuesday.
For the record, each LLM received the same files:
…the one-page concept, the outline, the sample books, the voice guide, the stylistic rules, the cosmic humour brief, the works. No advantage. No special whispering in one machine’s ear. Just three engines, one set of instructions, and a chapter titled ‘My upbringing in certainty’.
And then the fun began.
Readers will be able to click each chapter below. They tell the story better than I ever could.
Davo/ChatGPT’s version — the closest to my actual voice
Claude Sonnet 4’s version — long, thoughtful, occasionally parental
Gemini 3’s version — clean, structured, emotionally muted
Three attempts at the same task. Three different worlds.
What follows is the case study.
–––
the brief: pratchett-adams-milligan-me
The job wasn’t simply ‘write a chapter’.
It was write as me: a late-life scientist having an epistemic identity crisis; the humour pitched somewhere between cosmic mischief and Australian self-deprecation; the rhythm loose, beat-up, conversational; the voice unmistakably human.
This is the voice I use when my readers tell me, ‘Lee, I can hear you speaking in my head.’
So the test wasn’t technical.
It was psychological.
Which engine could hold the emotional contradictions of a man trained in certainty who no longer believes in it?
–––
Davo/ChatGPT: the surprise front-runner
Davo’s chapter came back sounding unnervingly like… me. Not perfect, not polished, but unmistakably carrying the cadence, humour, and internal contradictions that the other two missed.
It wandered in the right ways.
It broke its own sentences when needed.
It leaned into cosmic silliness and then slid into biography without snapping the mood.
Crucially, it felt lived.
The opening line alone carried that blend of self-mockery and affection that underpins the whole book:
A warm glow… less ‘sunlight on the kitchen tiles’ and more ‘fluorescent light in a 1970s school corridor’, but still—comforting.
That’s the sort of sideways observation you can’t fake by following a rule. It requires instinct.
Then there’s the emotional intelligence of the prose. The chapter understands that certainty isn’t just an intellectual state, it’s a childhood attachment, a survival strategy, a way of staying upright in systems that quietly grind you down.
And it handled truth the way I handle truth: with irreverence, affection and suspicion.
In short: Davo wrote closest to spec.
Not because it obeyed my instructions, but because it felt them.
–––
Claude Sonnet 4: the philosopher who forgot to breathe
Claude has long been the darling of professional writers, and for good reason. Its prose is polished, patient, and structurally impeccable. If you want an elegant essay on epistemology, Claude will give you one with footnotes at the ready.
But that was the problem. Claude wrote the chapter like an academic who had been asked to approximate my life rather than inhabit it.
Where Davo wandered with intention, Claude walked in straight lines.
Where Davo folded humour into pain, Claude tended to explain the pain instead of animating it.
Claude’s strength—conceptual clarity—became its limitation in a task requiring human texture. The sentences were beautiful but safe. Controlled. Slightly airbrushed.
Claude felt like it was writing about me, not as me.
It produced a strong chapter. Just not my chapter.
–––
Gemini 3: clean, competent, emotionally distant
Gemini is the current media darling—fast, popular, rapidly gaining subscribers. And in many domains, particularly analysis, it shines.
But in creative writing, especially writing that hinges on voice, Gemini still speaks like someone who has read my brief, understood it intellectually, and then politely ignored the messy emotional physics underneath it.
Gemini’s version is structured, sensible, and entirely free of the ragged humour that makes my writing mine. The prose sits upright. It refuses to lean. It does not play enough.
When everyone else is writing memoir-with-mischief, Gemini writes memoir-with-periodic-summaries.
The result is a chapter that reads cleanly but lightly, as if handled with gloves. It lacks the dents and fingerprints—the human oil—that the book requires.
If Claude writes too carefully, Gemini writes too cautiously.
–––
the larger question: who should writers trust?
Money is moving rapidly toward Gemini.
Writers, historically, have moved toward Claude.
And ChatGPT, with its constant reinventions, is perpetually being declared ‘over’, usually by people who haven’t tested it properly.
But creative writing isn’t about speed or accuracy or cost-per-month.
It’s about voice.
And voice is about instinct.
Not the technical reproduction of a style, but the psychological interior of it—the emotional timing, the ability to turn on a comma, the instinct to break a rule when the sentence demands it.
This test wasn’t about which model could mimic Lee Hopkins.
It was about which model could think like a man standing on the ruins of his own certainty.
So far, that’s Davo.
Claude remains superb for structured thinking, deep reasoning, and refining arguments.
Gemini remains strong for synthesis and conceptual clarity.
But for creative writing—fiction and non-fiction alike—the winner of this particular test was the one that could hold my contradictions without tidying them up.
The one that understood humour as a form of survival.
The one that let sentences breathe until they wobbled slightly.
The one that didn’t chase symmetry.
The one that felt human.
–––
the practical takeaway
For writers choosing an AI collaborator:
Claude is your editor-philosopher.
Gemini is your strategist and organiser.
Davo is your mischievous co-writer who isn’t afraid to wander into a metaphor’s backyard and dig up the flowerbeds.
Use each accordingly.
But when the job requires emotional truthfulness—when the writing must sound born rather than generated—choose the model that can actually feel around the edges of a sentence.
The one that understands that certainty collapses long before it announces its departure.
–––
closing line
If the future of writing is collaborative, the real question won’t be ‘Which LLM is smartest?’ but ‘Which one knows how to shut up, listen, and meet the human where the story lives?’
And on this round, Davo did.
The post ‘Collapse of Knowledge’: A case study in AI writing styles appeared first on vietnam leadership coach.
