“It’s lovely. It just sounds a bit like ChatGPT.”
I’d written every word of that scene myself.
And I had no way to prove it.
I pasted the scene into a chatbot and asked, “Does this sound AI-generated?” It told me some passages “show features commonly associated with AI-generated text, such as balanced sentence structures and a reflective tone.”
So, a maybe. About my own sentences. I slept badly.
It isn’t only me, either. In June the Commonwealth Short Story Prize had to investigate accusations of AI use against its winners. What settled it wasn’t a detector (the Foundation itself said these tools “cannot provide conclusive evidence on their own”). It was the writers’ working drafts, time-stamped documents and notes.
The fix: keep the receipts
Here’s what I took from that: don’t ask AI to judge whether you sound human.
Ask it to help you keep the receipts.
At the end of every session where I use AI, I now ask it for one short, dated entry in what I call my decision diary.
What I asked for, what it offered, what I kept, what I threw out, and why, in my words. Three minutes.
Funny thing: it does more for my writing than for my defense, because it shows me where the AI helps and where I keep overruling it.
The prompt
Paste this at the end of any AI writing session, in the same chat:
Before I close this session, help me keep my decision diary.
Project: [TITLE] / Today’s date: [DATE] / What I worked on: [SCENE, CHAPTER OR SECTION]
Look back at our conversation and write one diary entry. Follow these rules:
Keep what I wrote and what you suggested strictly separate. Never blur the two.
When I rejected or changed a suggestion, quote my exact words.
Record decisions, not quality. No praise and no verdicts on the writing.
If you can’t tell who wrote something, ask me before logging it.
Keep the entry under 150 words.
Use this format:
What I asked you for
What you offered
What I kept, and how I changed it
What I rejected, and why
Decisions I made on my own today
Before writing the entry, ask me one question: which decision from today am I proudest of?
Save each entry in one running document, newest at the top. If you wrote without AI, open a new chat, describe your session in a few lines and use the same prompt. And keep dated versions of your drafts instead of overwriting “final_FINAL_v3”.
What happened at the next meeting
Before: my word against a vague hunch, plus a chatbot’s “maybe”.
After: a page like this, one for every session on that story.
Oct 3, “The Biscuit Tin,” scene 4
Asked AI for: five ways the brother could react when he finds the tin.
It offered: anger, laughter, silence, a joke about their mother, keeping one receipt.
Kept: the receipt, changed to him tearing it in half. My words: “Keeping it is too tender. Tearing it is him.”
Rejected: all of its dialogue. “Too explained. He’d never say this out loud.”
On my own: cut the last line of the scene. Moved the tin to the car.
I brought the diary and my version history to the group. She read two entries, laughed, and said, “Okay, that’s more process than I have.”
Your turn (10 minutes, I promise)
Open the last thing you wrote with AI.
Run the prompt at the end of that chat.
Read the entry
Under the entry, write 100 words on why you made the decision you’re proudest of. In your own words, please. That paragraph is the most convincing proof in the whole diary.
Watch for the entries where you rejected almost everything. They say more about your voice than the ones where you kept a lot.
A few last thoughts
If you don’t use AI at all, this still applies to you.
If you write non-fiction, the diary does double duty: it’s also where you log which facts came from AI and which you checked yourself.
And please don’t start sanding your sentences down so they sound “less like AI”. That’s the one way to really lose your voice. Keep writing like you, and keep the receipts.
Has anyone ever told you your writing sounds like AI?
Hit reply and tell me what happened. I might share the best answers (anonymously) in a future issue!
Until next week,
Wilma
It kills me that when I run my copy through Panagram, it tells me my copy is like 67% AI generated, when 0% of it is. And I know that because I’m physically writing the words.
That being said, it’s stunning how much everything sounds the same nowadays. Having a unique voice will really stick out.
I’ve actually been struggling with AI detectors because I, too, had a friend tell me that my stuff sounded like AI. Since then, I’ve run my work through a detector, and the detector used here on Substack specifically tells me that every single poem I write is 100% AI.
I just did a rewrite on one of my poems to do exactly what you said to try not to do—and that is, sound more human. And just as you said, I did, in fact, lose my voice in some parts because of it.
The worst part is that after the rewrite, I intentionally didn’t do any grammatical editing before running my poem through the detector again, and yet I still got a 100% AI result.
I write most of my work off the top of my head, as the feelings hit. Most of my poems don’t have inspiration notes, whereas my essays are more likely to have some point-form notes and at least one revision. I do, however, use AI for feedback and grammatical editing before finalizing my work. So it’s frustrating to see that result.
I am most definitely going to try the prompt you suggested. Maybe it can give me some peace of mind!