Claude-Washed
I learned something new about my writing process. I’m late again on this blog post, because I’ve been buried in a coding project. But that compression of my time forced me to use dictation tools more, and the dictation tools let me pick at the scab of what I was already feeling.
I always have trouble with the blank page. I think everybody does, which is why I’ve always spent my time doing structured writing, building the outline first. But the real organic piece of the process is the brainstorming. You can talk much faster than you can write. You can say stuff out of order. You can have your circling thoughts, mess things around, and it doesn’t really matter, because you can always sort it out later.
When I’m asked for writing advice, it’s simple: write like you talk. Dictate your brain vomit, and the writing comes out sounding better for it. Getting AI to help here is kinda clutch. Not to rewrite your words. To resequence them, clean them up, organize them into something cogent for someone to bother reading.
I take all of this as an experiment. Whatever’s the correct process today is hot garbage tomorrow. So you have to stay willing to investigate and interrogate your own process with a critical lens. Not just writing. It’s everything about how you live your life, with a sense of experimentation. I kinda like to experiment on myself. Weird, I know.
I’m still learning how to write authentically with AI. In previous posts I was fighting the robot first drafts, constantly going through this anti-AI pass to make sure I scoured away the AI-isms and make sure it sounded like me. Too often I’d just give up and rewrite whole parts of it all the way through. But it was never a complete rewrite.
The friction I was feeling trying to write those posts is the real tell.
It really came home to me when Substack buddied up with the Pangram AI detector. I ran a couple of mine through it. Even though I know for a fact they weren’t, it said many of them were a 100% AI. Jeepers. Didn’t matter how much time I spent editing it and cleaning it up, running anti-AI passes, running my own voice pass through it. Didn’t change the indictment that the words that landed on paper, digital paper if you will, weren’t truly mine. They were AI’s interpretation of my own thoughts. Then Claude announced their AI watermarking for text: if Claude lightly touches the work, that watermarking goes in, and even editing and polishing it up afterward doesn’t take it back out.
I mean, if a detector can’t reliably tell genuinely human work from AI’s, that’s not just annoying for the writer, it’s a real big-ass problem for how post-secondary thinks about academic integrity cases and AI-use.
The Pangram thing proved to me that I couldn’t confidently wash AI away from AI-generated text. Claude watermarking confirmed it to be impossible anyway. I don’t actually care about the watermark itself. Or Pangram for that matter. I disclose my AI use. What irritates me is the false positives and the AI-slop witch hunts and human purity test bullshit. I know my writing process is authentic. But at the same time, I could be better. Last week’s post was an example of it being better, because I knew one hundred percent that those were my words going into the post, not Claude’s paraphrasing of my words.
The idea for that came from Julie Kaiser’s blog, where she introduced me to the term vibe writing. Essentially that’s what she was doing to maintain her author voice: she’d dictate or brain dump and brainstorm, then use AI to try and reconstitute and build a blog post out of it. I’m taking it a step further, trying very much to have the AI keep my language and my voice intact, and actually take what I’ve actually dictated, what I’ve actually spoken, what I’ve actually said and thought, and use that directly.
The thing that was interesting about how I wrote last week’s post was I had actually spent the time and brainstormed, three or four, maybe even five separate voice dictations, and sent them over to Claude to do the transcription. I directed it to use my own language from my dictation rather than the paraphrasing it does by default. Much the same as what I’m doing for this post right here, except doing it interview style.
I don’t think I’m changing my entire process. I’m just changing the order in which I do things. In the previous process I’d do the brainstorming, get an outline made and organized, build the post from there, and then edit it. This is different. Rather than taking my raw text, turning it into an outline, and generating a draft I have to hand-edit, I’m getting the raw text organized while making sure the word choices and the sentences and the ideas come from my dictated text instead of being Claude-washed.
I don’t think it’s actually a big change. It’s just recognizing and appreciating that my own words are better than Claude’s. That’s all.
