Claimer
The internet currently contains two equally unhelpful positions on AI-assisted writing.
The first insists that anything touched by AI is fraudulent.
The second insists the human barely matters anymore.
Neither feels particularly fair to me.
The footer on every page of this site notes that it may contain traces of computational assistance. This page explains what that actually means.
Yes, AI is involved in the writing on gazza.net.nz.
More. AI is an integral part of what is finally published on this site.
But not in the way people imagine when they talk about AI as intellectual dishonesty.
The thoughts on these pages are not summoned into existence by my typing a prompt: “Write 800 words about dishwashers in the style of David Sedaris.”
The actual process is far messier, more conversational, more human, and considerably more fun than that.
Every piece begins with something I noticed, experienced, found absurd, or could not stop thinking about.
Sometimes it’s a phrase somebody used incorrectly.
Sometimes it’s a domestic ritual.
Sometimes it’s a strange emotional reaction to an appliance, a bureaucracy, or modern life generally.
AI does not have those experiences, and it does not generate them.
It does not live in my house.
It does not know my relationships.
It does not independently develop opinions about cupboard doors, phishing emails, seal road signs, or linguistic decay.
What it does extraordinarily well is help interrogate those observations.
The process resembles:
conversation,
refinement,
pattern recognition,
compression,
tone calibration,
and structural discovery.
I will often arrive to the conversation with
a sentence,
a screenshot,
a half-formed irritation,
or simply the sense that “there’s probably something in this.”
From there, the essay emerges through discussion.
The AI proposes.
I reject.
The AI reframes.
I redirect.
We gradually discover what the piece is actually about.
I do not use only one model.
Part of the process may include feeding a draft to Claude to see what it notices, whether it identifies weaknesses, whether an emotional turn feels earned, or whether it agrees with me that a paragraph is trying too hard.
Sometimes different systems notice different things.
This turns out to be surprisingly similar to how human editors work.
One editor notices structure.
Another notices rhythm.
Another notices sentimentality.
Another notices where the piece stopped trusting the reader.
The process also includes reading drafts aloud to Joy.
This is partly because she is often present in the stories themselves, but mostly because prose behaves differently in the air than it does on a screen.
A sentence can look elegant silently, and sound unbearable when spoken.
Another can appear slightly awkward visually and reveal perfect rhythm once read aloud.
So I listen for cadence.
For whether the joke lands naturally.
Whether the emotional shift feels honest.
And whether the prose sounds like a human being noticing something rather than a machine assembling observations.
And Joy’s reaction matters enormously.
Not because she is a formal editor, but because she is another intelligent (human) being hearing the piece in real time.
Confusion matters.
Laughter matters.
Silence matters.
A slightly delayed “oh…” matters.
The actual workflow often looks something like this:
- Human experiences thing.
- Human discusses thing with AI.
- Human refines through iteration.
- Human cross-checks against another AI.
- Human reads aloud.
- Another human reacts.
- Human revises for rhythm, restraint, and truth.
- Human publishes.
At no point does the machine decide:
- what matters,
- what is funny,
- what is cruel,
- what is too sentimental,
- what should be removed,
- what should remain implied,
- or what emotional truth the piece is actually circling.
Those decisions are mine.
So is the taste.
Which is perhaps the most important part.
Because taste is the actual creative act here.
Not the typing.
I hope the essays ultimately sound less like machine-generated prose, and more like “Gareth with unusually good editorial stamina.”
And that is probably a reasonably accurate description of the process.
I do not regard this process as dishonest any more than I would regard editing, collaboration, conversation, workshop feedback, or reading drafts aloud as dishonest.
What I am not claiming is that every word emerged directly from my solitary consciousness unaided. Almost nobody’s writing process works that way.
The tools are undeniably powerful.
But power is not authorship.
The posts are my observations. They draw from my relationships, use my humour, amplify my irritations, show my sensibilities, and reflect my worldview.
Anyway.
I use AI. I am not Stephen Fry.