Using AI to Clean Up Your Thinking (Not Replace It)
The most useful thing AI has done for my creative work is help me hear what I’m actually saying.
I’ve been experimenting with using AI to process my voice memo transcriptions, and something unexpected has come out of it.
Not what you might think. Not “the AI generates my content” — if that’s what you want, there are tools for it, and it produces something that sounds superficially fine and is fundamentally empty in the specific way that things produced without genuine thought are always empty. That’s not the use case I’m interested in.
What I’ve found actually useful is something more specific: AI as a clarifying mirror. A tool that takes the slightly rambling, cross-cutting, half-completed way I actually think when I’m walking and talking, and helps me see the structure that was there all along.
Here’s the problem with voice memos as a capture medium.
They’re excellent at catching the thought as it arrives — raw, unedited, in the voice it arrived in, with all the connective tissue and the false starts and the moments where you’re clearly working something out mid-sentence. This is their great strength. You don’t edit on the way in; you just catch it.
But the output is not article-ready. The output is a transcript that reads the way live thinking sounds: repetitive in places, circular in others, with insights buried inside longer passages of working-toward-them that aren’t needed once you’ve got there.
The traditional editing process is time-intensive. You transcribe, you read, you identify the core idea, you pull it out, you construct around it. This can take four or five times as long as the recording itself. For someone who captures a lot — and I do, because capturing is cheap and forgetting is expensive — the volume of unprocessed material accumulates faster than the processing can keep up.
What I started doing, partly out of necessity and partly out of curiosity, was asking an AI assistant to help with the first pass.
Not to write the article. To read the transcript and tell me what the transcript was about.
The question I ask is something like: “Here’s a rough voice memo transcript. What are the main ideas in here? What is this actually saying when you strip away the processing?”
And what comes back — at least when the original thinking was genuine — is a useful compression. Not a replacement. A compression. The ideas I’d been circling around, stated cleanly. The connections I was making but not quite landing, made explicit. The thread that runs through the apparently disparate observations, named.
It’s like having someone listen to you think out loud and then say “I think what you’re actually saying is…”
Which is enormously useful.
There’s an interesting dimension to this that I want to name carefully, because I think it matters.
The AI is only as good as the thinking that went into the voice memo. If I recorded something that was genuinely exploratory — working something out, with real ideas underneath the rambling — the compression is valuable because the ideas are real. They were there. The AI is just finding them and cleaning up the path to them.
If I recorded something that was thin to begin with — a half-hearted attempt to generate content without actually having something to say — the AI produces a compression of nothing. Polished, well-structured nothing. Which is in some ways worse than the rambling original, because the rambling at least is honest about what it is.
The tool reveals the quality of the thinking. It doesn’t create quality where none exists.
This is why I’m not interested in AI as a content replacement. I’m interested in it as a mirror and an accelerant for thinking that already happened. The walk, the observation, the genuine attempt to work something out — that part cannot be delegated. The cleanup and the structural pass is where AI earns its keep.
There’s another dimension that I find genuinely interesting: AI as a playful collaborator.
One of the ideas I’ve been experimenting with — and this is more playful, less rigorous — is inserting structured creative constraints into the process. What if you took the raw voice memo transcript and asked an AI to identify three very different article angles it could support? Not to write the articles — to name the angles. To show you the different shapes the same material could take.
This is the kind of thing a good editor does for a writer. “This material could go left or it could go right — which direction do you actually want?” Having that question available quickly, without needing to wait for a human editor’s availability, is genuinely useful.
The honest summary of where I’ve landed on AI and creative work:
It’s a good tool for the stages of the work that are about organisation, compression, and structure. It’s a bad tool for the stages that are about genuine thinking, personal voice, and the specific resonance that comes from someone actually having experienced something.
Use it to find your thread. Not to spin it.
Use it to compress. Not to generate.
Use it to hear yourself more clearly. Not to replace the hearing.
The voice in the voice memo is yours. That’s the irreplaceable part. Everything else is just the path from thought to publication, and on that path, any tool that makes the journey faster is worth having.
If you have a backlog of recorded thoughts, notes, or voice memos that you haven’t done anything with — try feeding one to an AI and asking it what you were actually saying. You might find there’s an article in there that you’d already written, just not yet tidied up.
CP52 Stage: Stage 3 — The Setup (building the infrastructure)
Series: Making It Work Online
Image note: A phone with a voice memo, and beside it a clean printed page. The same idea in two states.
