How I use AI as an editor, not a writer.

Firstly, I don’t ask the AI to generate text. I never say ‘please write me an article about X’.

If you’re looking for that kind of article, this isn’t it. You might like to jump straight to the 'Prompts I Often Use When Writing' section and the article summary at the end.

Thanks for reading! Subscribe for free to receive new posts and support my work.

For everyone else, welcome to the scenic route through My Writing Process.

I do all my own stunts writing-wise. Every word you read comes from my head except where quoted.

My most recent piece “AI is bullsh*t…” started with a photo of a note in my journal. It was an idea that wouldn’t go away.

What follows is “the making of…” style report on how I wrote that piece.

Like all the best things, my process follows the rule of 3:

1. The Framing/ Initial Notes
2. The Actual Writing
3. The Feedback Loop

1. The Framing/ Initial Notes

First I want to make sure I’ve got the right frame for the text to fit into so my first goal is to get some initial high-level feedback on the idea itself and any structure that supports it.

Here’s how this worked for “AI is bullsh*t”, the prompts and notes are for illustration purposes. They’re just examples, I’m not saying you should follow them verbatim.

I uploaded the photo from my journal to chatGPT

with the prompt “This feels like the start of a post”

That was a mistake because chatGPT responded with:

Absolutely — it’s raw, provocative, and poses a real challenge to how we frame both AI and work. Here’s one way you could shape it into a short post:

If AI Can Replace Your Job, Let It

AI is bullshit. But so is most of everything else.

If AI is coming for our jobs, why aren’t we saying yes please, take it?

If your role can be done by a machine, maybe that’s a mercy. Because if no part of your actual self — your essence, your weirdness, your spark — is in the thing you spend most of your waking hours doing… why do you care if it’s taken away?

Maybe AI isn’t replacing us. Maybe it’s revealing what never needed a human in the first place.

Want to go longer, or keep it punchy like this?

Now, I’m to blame here because “this feels like the start of a post” is a terrible prompt.

What I wanted was a response, I didn’t want a rewrite but chatGPT is a helpful assistant so it took my words as an implicit instruction to finish the post.

I compounded the error by responding with another thought: “There's something about how the arrival of AI has revealed this is just an arbitrary game with rules, winners and losers. It's like the matrix No I'm not stoned or on drugs, just the medication that helps me function in this system.”

And it provided me with more generated text.

I told it to stop: “You keep rewriting my lines, I just need notes not suggested text.”

Once I’d been clear about what I wanted, the magic started to happen.

chatGPT responded:

“Got it — no more rewriting, just notes from here on.

Here are some angles you could explore based on what you've got:

1. AI as Revelation, Not Disruption

   • Rather than destroying work, AI is exposing how empty a lot of work already was.

   • It breaks the illusion that our roles were sacred or uniquely human.

2. The Arbitrary Nature of the System

   • The rules of the game (job titles, performance metrics, “career paths”) were made up.

   • AI just plays by the same rules — only better — which makes the rules look ridiculous.

3. The Emotional Disconnect

   • If you’re medicated just to function inside the system, that’s not a life — that’s survival.

   • Why mourn the loss of something that already required you to numb yourself to get through?

4. Matrix Parallel

   • The system keeps people plugged in with just enough reward to prevent revolt.

   • AI is a crack in the illusion — and once you’ve seen it, you can’t unsee it.

5. Core Question

   • If you’re angry at AI for threatening your job, what are you really angry about?

   • That you were never seen? That your job didn’t reflect your value? That you were replaceable?

   Want to go deeper on any of those?”

This is high-value, these notes help me with framing my structure (what’s in? what’s out?) and how to manage reader expectations.

At this point, I go away and think about the article. I make notes in my notebook.

2. THE ACTUAL WRITING

When I sit down to write, I don’t reference the notebook, its work is already done. The point of the notebook is to build a mental model of the piece so I can write it (we’ll come back to this mental model later).

I also don’t touch chatGPT until I have a first, or even second, draft. Now some readers might be saying “this is counterintuitive. I want chatGPT to do the writing for me. I don’t have time for that.”

To use a metaphor I read in an book: I am a cyborg when I use AI to write. AI is part of my process, a built-in enhancement, I have wired in a writing enhancement module. I give this module input and process its output. I could detach it and still be able to write.

I’m not using it as if I were a centaur, half-man, half-AI, where the AI is an indivisible part of me and I must take responsibility for the words chatGPT outputs like they were my very own.

I presented a first draft of “AI is bullsh*t” to chatGPT with the prompt “tweaked it again”.

This time chatGPT’s response included “Here are my notes, as requested, no rewrites:” so it remembered from before. I didn’t need to tell it twice.

I asked it to summarise its output, again I include the full summary here

🧱 Structure & Rhythm

• Strong opening hook and sharp tone throughout.
• Excellent pacing from existential tension to blunt payoff (“0% you”).
• “Crazy Old Man Cobb Country” adds personality; tone works but could be softened slightly depending on audience.

🔥 Thematic Strengths

• Scooby Doo metaphor is vivid and effective.
• Highlighting the irony of undervalued, AI-proof jobs is a standout insight.
• The alternative CV is human and hopeful — balances critique with vision.

✂️ Possible Tweaks (Optional)

• Middle line about bills could be tightened for rhythm.
• “Break the game” works as a mic drop, but could end more emotionally if desired.

🎯 Platform Tone Consideration

• Works for both Substack and LinkedIn.
• For LinkedIn, a soft intro line might help ease more corporate readers in.

I had only 2 actions from the notes:

1. Go back and look at the middle line about bills. And look again at the ending.
2. I also responded “How much do you think I care about appealing to corporate folks?”. To its credit, chatGPT responded “Not at all. You’re not trying to get hired, you’re trying to say something true”.

3. The Feedback Loop

I made another 20? 30? 100? edits to the piece. Sometimes changing just a word, sometimes tweaking the punctuation. Each time I came back to chatGPT to ask for notes.

It’s important to remember that chatGPT and other LLMs are non-determinative. This is a fancy way of saying you won’t get the same response to the same inputs.

If you’re using AI to write, this is something you need to be aware of. It will tell you something works, then tell you at the next review point that that very same thing needs to be tweaked.

If I know why I’ve written every single word on the screen then I can justify each of them.

If the words had been generated by chatGPT, I would not have that uncorruptible mental model of the piece and the value of chatGPT as a writing partner would be diminished.

chatGPT has, on occasion, praised me and quoted phrases that aren’t in the text, it has provided notes on sentences that only live in its memory.

Your mental model is the best defence for chatGPT giving you notes on text that doesn’t even exist in your piece.

Other Neat Stuff chatGPT Does

Another key concept is that chatGPT can hold some memory of the context of the conversation. It’s that functionality that meant it remembered I didn’t want rewrites from it earlier.

So don’t be afraid to argue, make a stand. AI is a servant to your voice. Before it learns your writing preferences, its sensibilities can be that of the average corporate America of the Internet. If that’s who you’re writing for, fine. If not, tell it why you meant to say something a certain way (“I wrote it like that because…”) and chatGPT will remember and not provide notes on that again.

chatGPT has learned my writing preferences. It has committed a lot about how I work in memory. Every time I ask for feedback, it consults that memory to shape its responses.

Prompts I Often Use When Writing (you can and should combine some of these):

• “New piece for [Substack]. Working title is [“AI is bullsh*t”]”

  • When you tell chatGPT the intention for your writing, it can adjust the register of its notes. You don’t want the same review process for that email draft to Sales as for your Great American Novel.

• “Review/Analyse the following text as a kind but critical friend. Do not rewrite anything. Instead provide notes as a bullet list.”

  • Self-explanatory. Direct.

• “Made some edits, they’re subtle. I bet you can’t spot them all. Review the text and make sure you process every word then review and provide notes as a bullet list.”

  • The point of challenging it to spot changes is cueing that the new edits may be hard to identify. But I still want feedback on them.

  • I know, I’m anthropomorphising. chatGPT can’t ‘spot’ anything. You see I got a grip later in the prompt and switched ‘read’ for ‘process’ to better reflect what’s actually happening under the hood. But I’m wildly inconsistent as to when I apply anthropomorphism. Speaking of which…

• “What do you see here?”

  • I use this one all the time. Open-ended, can sometimes generate an angle I had not seen. More useful now I have a small body of work and chatGPT has memorised what I expect as an outcome format. I’ll usually upload a photo of a page from one of my notebooks to accompany this question.

• "Is this complete?"

  • Not ‘have I done all the editing possible?’ but ‘is this piece structurally sound?’ This is great for catching the ‘unknown unknowns’ where your reader wants something from a piece you haven’t yet provided.

• “What would you think if you were my boss and you read this?”

  • Especially for the ‘AI is bullshit’ piece. I don’t think AI is bullshit but the Internet is not the place for nuance and if you only read the headline, you might think I didn’t want my job any more. That’s why I didn’t post on LinkedIn (still not writing for corporate folks, heh).

• Not technically a prompt but respond to any chatGPT notes you disagree with and will permanently ignore: “I wrote [direct quote] that way because…”

In summary my process to use AI as a writing partner is to:

1. Come up with an original draft idea or piece of writing in my own words.

2. Submit to chatGPT for a first round of notes. This can be uploading a photo of my hand-written notes or a copy paste from Google Docs.

3. Screen chatGPT’s notes for what has value and what I can use. Respond individually to notes I am going to permanently disregard (‘how much do you think I care about corporate folks?’) so chatGPT doesn’t give me that note again.

4. Edit the piece.

5. Submit the edit to chatGPT for more notes.

6. Tell chatGPT which notes I am going to ignore so it doesn’t suggest them again.

7. Repeat 4 to 6 until its suggestions stop making me think of different ways to edit the piece.

   1. IMPORTANT: this is why it’s crucial to have a mental model of the piece and know it from soup to nuts. chatGPT will keep making suggestions as long as you keep giving it drafts. You and only you will know when the piece is actually complete.

8. (optional) publish your piece on the Internet, then (your mileage may vary) lie down in a darkened room telling yourself you haven’t given a vital piece of your soul away by exposing your writing on the Internet.

Thanks for reading! Subscribe for free to receive new posts and support my work.