How I write
For me, the most important question is always why. Why this story, why now, why for this reader. The what and who and when follow naturally once the why is honest. If the why doesn't survive its own first interrogation, the rest of the work is wasted.
I keep a list of story ideas - mostly alien-contact scenarios, since that's the genre I most want to read in and most rarely find done well. Some on the list will make stories. Others won't. The first cut is the why.
On punctuation
On writing style, there have been some thoughts around the use of long hyphens in some of my stories, specifically that they reflect the use of LLMs in the creation of the text. My punctuation mark rules are mine and they do include both long and normal hyphens. Their use depends on the language of the characters and my avowed dislike of staccato/clipped sentences, something the long hyphen helps me avoid, perhaps to my cost. It is just a personal thing. Here are my rules:
I give each punctuation mark a job. A colon opens one idea out into what follows; a semicolon holds two balanced thoughts in a single breath; a comma carries the short aside; and the sentence itself, through its grammar, does most of the connecting — which is how the prose stays flowing rather than clipped. The dash I keep for the work nothing else does: the genuine aside, the sudden turn of thought, the line that wants a beat of emphasis, the speech cut off mid-word. I'm fond of the long dash and make no apology for it — but a mark that does everything ends up signalling nothing, so I let it do only the few things it does best.
The setting question
The who is harder than the why. It partly comes down to setting. I love Australia, and I write Australian settings often enough that I have to watch myself - there's a point at which it becomes habit, not place. So I move around: Scotland, Russia, Brazil, the southern tablelands, the Coral Sea, wherever the story actually wants to be set rather than where I'm comfortable.
The science question
The what is where I work hardest. I write science fiction, which means the speculation has to be earned. I read the published literature, I follow the active researchers, I extrapolate from what is established or from theory that has been proposed but not yet confirmed. The pace of technology now makes this harder than it used to be: what felt futuristic last year is shipping product now. The good news is that real research is in front of where most genre writers imagine; the bad news is that the writing has to run faster.
The scaffold
Once the why is set and the who and what are roughly in hand, I build a scaffold - a structural skeleton of the whole story, scene by scene, before any prose is written. The scaffold is what keeps the story from drifting. It is not the outline - outlines lock you in. The scaffold lets the story breathe while still pointing at the ending I think I want. I revise the scaffold mid-draft regularly; some endings only reveal themselves once you've written your way toward them.
Alongside the scaffold I keep five working documents:
- An arc bible - borrowed from television writing rooms, where each long-running series has one. A living document that tracks who is doing what, when, and why across the whole work. What I check the manuscript against when I want to know whether the story is still honouring its own logic.
- A character map - relationships, arcs, key facts, who knows what about whom and when they learn it. Especially important on long works with many actors.
- Science-and-society notes - what the actual physics, biology, or social structure underlying the story is doing. Where the speculation is grounded in published research, where it extrapolates, and where it openly invents. Each of my novels has one of these; they are also the source for the science background pages on this site.
- A writing style guide - voice rules for each character, prose conventions, the words I overuse and have to police, the rhythms I lean on too often.
- A collaboration rulebook - the working agreement between me and whoever I'm collaborating with on a given draft.
On collaboration
I am, before anything else, a scientist. Scientists use the tools available to them. For the last several years, one of those tools has been a large language model - a working partner I draft with, argue with, and push back against, in a relationship not all that different from working with a sharp editor or a research assistant who has read everything.
The LLM contributes in four specific places:
- Continuity - checking the arc bible against the manuscript, catching the inconsistencies a human reader (including me) misses on a long work.
- Voice - keeping each character authentic in their own speech, pulling me up when I drift into my own cadences across characters who should sound nothing alike.
- Feedback - challenging my ideas. Some of what I think is brilliant turns out to be ordinary; some of what I think is throwaway turns out to be the spine. The conversation surfaces the difference.
- Style - catching the small grammatical and rhythmic failures that haunt any prose, including the specific ones that haunt human-LLM-collaborated text when its provenance shows.
This is collaboration. Novelists have always worked with editors, copyeditors, beta readers, sensitivity readers, research assistants. The genre's history is collaboration. Pretending otherwise - pretending the book is the work of a single unaided hand - is the dishonest move, not the LLM. I treat the LLM the way I treat any other collaborator: with respect, with scepticism, and with the final call. The work is mine. The conversation around the work is shared.
Anyone who finds that uncomfortable hasn't looked closely at how novels get made.
I have set out the longer version of this argument, in two exchanges with the model itself, on the conversations page: AI assistance and the editing line →
An example: Saviour
For Saviour, the what arrived first - a Bronze Age artefact, an Old Believer community on the Danube Delta, an object made for a mind that doesn't yet exist. From there: scaffold, arc bible, character map, the science-and-society notes that established what the artefact actually does to a neural architecture, and the draft itself. The LLM was in the conversation throughout - pushing back on early ideas that turned out to be ordinary, catching continuity problems before they reached the page, holding the voice of each character against drift. The result is a novella I'm proud of, and one I'm making freely available. If you're curious about how this kind of collaboration reads, that's the place to read it. Read Saviour →
On revision
Some stories come quick and easy - they almost write themselves. Others take weeks, often months, to reach a stage where I'm happy with them. Six to twelve revisions per chapter is normal; on a twenty-chapter novel, that's a lot of revisions. Some stories I've rewritten four or five times from the ground up. Telling stories is never straightforward - but it is a lot of fun.
Feedback from friends is an important driver of what those revisions look like. Not always perfect, but always appreciated.
On the rest of it
I may never win any awards. That isn't the point. The point is to explore ideas most people don't realise exist - the texture of contact across cognitive distances most science fiction quietly avoids - and to give those ideas to the readers who want them. So far I'm having a hell of a time doing it.
Why don't you try having a bit of fun as well.
Disagree with the process? Have a sharper approach you swear by? Strong views on the collaboration question? Send me feedback →