Arweave TX
James Campbell
It began, as these things so often do, with a message fired off into the digital void:
That was me, asking ChatGPT to help write an article about writing an article with ChatGPT. A bit like using a mirror to check if your other mirror is working — only with more existential dread and less toothpaste.
Why am I doing this? Because I want to write more. For other people and in public. But every time I sit down to start, I get a polite but firm "you must be new here" from my own brain. So I’ve turned to AI. Not because I think it’s the future of literature (though it might be), but because it’s available, non-judgemental, and doesn’t get distracted halfway through by the idea of making another cup of coffee.
This article is a record of our collaboration. It will be brief, self-aware, and hopefully not so meta it collapses into a singularity of smugness. We’ll look at:
The Beginning (this bit)
The Process (the meaty bit)
The Polish (the final tweaks before publishing it into the wild)
Every screenshot you see is real. Every line from ChatGPT is straight from the conversation. Every mistake is probably mine. You can even read the full chat transcript here.
Let’s get on with it.
One of the questions I’ve been turning over as we go is: how much of this is me, and how much is it the AI?
The answer — at least so far — is: it depends on the moment.
Sometimes I’m leading the direction, like when I opened with this:
"I'd like to write an article about AI-assisted writing. I'll be using screenshots from this conversation (including this initial prompt) to show the reader how we collaborated..."
That first prompt set the tone. I was clear about wanting it to be brief, meta, and light. From there, the AI helped with structure, tone, and phrasing. It suggested an outline. I approved it. Then it offered a draft intro, which I edited.
Later, I wrote:
“if you're going to use metaphor and simile (origami swan) — I'd like them to be a bit more douglas adams (who I'm sure would hate this).”
That was probably my most pointed bit of feedback so far. The response improved — funnier, weirder, more my style — but still recognisably AI-generated. I didn’t write the next draft, but I nudged it.
This process is iterative. I prompt, it replies. I critique, it adjusts. I shape the tone, but I’m not typing every word. At times, the line between author and assistant blurs — not in an eerie "who am I?" sort of way, just in a practical "this paragraph came from a conversation" way.
All of this — every prompt and reply — will be available in full, unedited, as an appendix. The article is built from that raw transcript, but polished for clarity and rhythm. Nothing hidden, nothing magic.
From here on out I'm flying solo. No AI allowed.
I'm writing this about a week after the initial experiment, and after reading it back - I'll be honest - I'm not impressed with the end result. However, the process was enjoyable and enlightening. Here are my key takeaways:
Curation is Key. The AI can produce almost endless content, and endless variations of that content. Knowing what to cut, what to keep, and what to rework, is crucial.
Limitations of AI. The humour is all over the place and screams AI - it feels forced and unnatural*.* This is definitely an area where human writers are just better.
Iterative Process. The back-and-forth nature of the collaboration is the core mechanic. Smaller and more focussed iterations work better. The ability to easily try out a different tone or structure is invaluable.
Future Potential. I'm definitely going to continue down this path. I'm learning what the AI is good at and what it's not. I'm learning how we can best assist each other. I want to refine my prompts and process into something more repeatable.
This experiment has been a valuable learning experience, and I've found it quite inspiring. I'm looking forward to creating more compelling and cohesive work in the future.
Full Conversation - including some bonus advice on from the AI about how to do this better.
