Dominic Baggott

Specimen #E8431C · Opinion

Thoughts on recording thoughts

Part of what I’m trying to do with this site is get my thinking written down. Mostly so folks who might be interested in working with me can get a sense of how I think, and decide whether that would suit them. The trouble is, I don’t think in writing. I can get there in the end, but it’s hard work. So I thought it might be interesting to write about how I use AI tools to make that easier.

I know there’s a lot of backlash against AI. Brain off, GPU go brr, thinking strictly not required. But I think there’s a way to use these tools to do quite the opposite, to help us clarify our thinking, rather than having our thinking atrophy. First though, story time. I promise it’ll get to the point.

Back when I worked on the NHS.UK alpha, one of the things we wanted to do was write openly about the work we were doing. We had some help from the excellent folks in the GOV.UK comms team, and Matthew Sheret in particular. I don’t think the technique originated with him, but he’s the one who helped us with it.

The thinking was that on teams you often have really smart, insightful people with lots of good stuff worth sharing, but who struggle to get it down on paper. Partly that’s a problem of the written word. A lot of us don’t have great experiences with writing. We start and stop in secondary school English lessons, writing essays, staring at a blank page where you have to write a thing. It’s really hard. And you lose a lot of your natural voice, because you write the way you think you’re supposed to write, not the way you talk. You lose the flavour, the intonation, the thing that makes your voice yours.

So Matt came along, popped a little tape recorder on the table and interviewed us. Not formally. He just started a chat about the things we wanted to write about. It turns out that if you spend a lot of time working on something, you have lots of thoughts about it. And for some people, given the chance to talk about it, those thoughts will just tumble out. So he gave those thoughts the nudge they need to start tumbling, then he took the recording, transcribed it, turned it into an outline and said: there you go, you’re on your way. The block, that stumbling block of the blank page, is gone. Instead of a blank page, we got our own thoughts reflected back at us.

I loved this. It worked so well for me, because I struggle with blank pages. I don’t know how to start, and my first drafts don’t feel remotely natural, remotely like me. They’re stilted and formal and just a bit too business-y. I’ve gotten better at getting started over the years, but I’ve always found speaking easier, and I think I always will.

So that’s what I’ve been trying with the pieces on this site. I put my AirPods in, head out for a walk and just talk. I monologue, recording myself talking about a topic like this one. When I get home I copy the transcript out of Voice Memos1, give that transcript to an AI model (Claude at the moment, through habit more than anything) and say: here are my words, here’s the raw material, pull out what makes sense for a little post. It produces an outline. I read it and go, no, that’s not right, that’s got the wrong sort of stuff. We go back and forth a bit, the way I might debate it with a person, to thrash out what I’m trying to say and whether the words on the page actually say it.

Then I typeset it, and in that process I’m reading the words a lot more, massaging it, stewing it. Eventually I feel like I’ve found my voice in the written word, not just the spoken one. It gets much easier to drop in and rewrite paragraphs here and there and shift things around, until I’m like: yeah, this is right. It sounds like me. It feels like me. I could imagine standing up and saying it. No ugh, no squirming.

And it’s still my own thinking. I’m not saying “I need an article on using AI tools to write articles, please write one for me.” The raw material is mine. The tools just help me over my own blocks and hurdles, to something that’s genuinely in my voice and that I can stand behind.

None of this needs to be a big serious thing with a big takeaway. These tools work for me. They might work for you. They’re practically free. Give it a try.

Side B

1 The recorded notes for this piece actually say “run it through Whisper, locally on my MacBook Pro”. But in following this process, I realised Voice Memos transcribes the recording entirely locally on the phone, and you can just copy it out directly. The wonders of technology, eh? ↑