How I Use AI — and How I Don't
A story called “The Serpent in the Grove,” by the writer Jamir Nazir, won this year’s overall Commonwealth short story prize after first going viral for the opposite reason — critics on social media accused it of showing “obvious markers” of AI use, and the literary magazine Granta withdrew from its long-standing agreement to publish the Commonwealth winners over the controversy. The Foundation reviewed the drafts, the time-stamped documents, the notes, and stood by the writer. Its director-general said they were satisfied the testimony from Nazir confirmed AI hadn’t written it, according to reporting in The Guardian.
I haven’t read the short story, but I’ve already unfairly judged the writer. I found myself doubting the legitimacy of the whole review, the Foundation included. I hope I’m wrong about all of that. But the fact that a prize-winning story now needs a forensic paper trail of drafts and timestamps to be believed tells you something about where we’ve landed, whether or not this particular writer is telling the truth.
Substack just introduced a new restriction around AI. They’ve brought in a software called Pangram that checks whether a piece of writing is fully human, or AI-assisted, or AI-written.
My first reaction was a mix of excitement and fear. The fear part is almost embarrassing to admit: some part of me felt like a kid who’d cheated on an exam, waiting to be caught out, because I have previously experimented with AI in my writing. Which is exactly why my quick judgment of Nazir sits uneasily with me.
The excitement comes from my belief that we do need some guardrails for using AI in writing, and I’m so glad Substack is doing so.
How I actually use it
Over the past few months, I’ve settled on how I use AI in my writing. When I have an idea that I’ve decided to write on, I riff on my phone for about five to ten minutes in a stream-of-consciousness manner. I then upload it to Claude and ask it to make sense of it with remarks and comments, but to use only my words with a reflection at the end.
Then using Scrivener (a word processing program)I rewrite what was my first messy draft. This is where I do the heavy lifting, and it’s always without AI.
Once I’ve got a proper, legible draft down, I ask AI to act as an editor on three levels. The first, which I’ve only started doing recently, is using it as a thought partner that argues against me. I’ll ask it to counter my own points, find the holes in my thinking, so I can go back and bulletproof the idea before anyone else pokes at it. The second is developmental — does this paragraph earn its place, does the piece actually flow, is there a hole in the structure? And the third is proofreading and line editing.
Sometimes, I can use Claude like Google on steroids for some research questions — faster access to the same kind of information I’d have hunted down anyway.
To me, AI is a great tool that we can use or abuse. It’s how we choose to use it. I choose to use it as a thought partner and an editor on hand at all hours that enhances my thinking.
Where it goes wrong
The danger is what happens when people skip thinking deeply about an idea, prompting and letting the AI do so. What we get is one generically competent voice, multiplied across every corner of the internet. It’s not that AI writes badly. It writes well. That’s almost the problem — it’s smooth, vanilla, competent, and nobody had to sweat to produce it.
The whole point of art, craft and the act of creativity is that there should be some human struggle behind it. We’re storytellers because we intuitively recognise each other in a story. We’re not going to feel that same recognition toward something that has no experience of its own to draw from.
Also importantly, when you shortcut the effort of writing something — an essay, a book — you miss out on what that effort was supposed to do to you. When I wrote about my Shrooms experience in my memoir, the act of remembering, collecting my notes from that time, putting it on paper, and making sense of the whole adventure was what gave me the true insights.
There was a moment I wrote about a cluster of white flowers in the garden: “They had been there for four years, but that was the first time I had noticed them.” I didn’t understand what that line was really about until I’d sat with it on the page — that the whole experience had been about finally noticing what was always there. No prompt would have handed me that. The sitting did.
If I’d just narrated my thoughts to Claude and asked it to write it out for me, then I wouldn’t have spent so many hours with my thoughts that finally marinated into my being.
When you outsource to AI the deep emotions we go through, then there’s no real change, no suffering, no pain in our writing. And most readers will catch that on the level of the whole piece, over time.
Finally, if we allow ourselves to keep using Claude to think deeply for us, not with us, then we let the muscle of thinking wither away — and that muscle is what makes us human. The internet, social media and consumerism have already spent years training us to think less.
We can’t allow AI to be the final nail in the coffin.
Where I land
I’m not a Buddhist, but I resonate with much of its ideology, especially the Middle Way concept. We can have enough discipline not to reach for whatever is easiest without tipping into the other trap of treating the struggle as some virtue in itself.
We keep treating the suffering as the toll, the bit you get through so you can reach the writing on the far side. But it isn’t the toll, it’s the writing. That’s where whatever meaning there is actually gets made. Take it away, and you’ll still have your essay, your book, your song. You just won’t have the person who was meant to come out the other end of making it.
So for me, I’m going to keep AI close, a thinking partner, an editor, that I can reach at any hour. What I won’t hand over is the one part that was the whole reason for sitting down. The thinking and writing that originates from my mind and heart, and that keeps me alive.
But I’m also wary of handing tools like Pangram the final word on what counts as human. They often flag real writing as fake, and let’s remember it was us humans who used the em dash long before AI ever did.