Who wrote this, really?

New skill unlocked

I've always wanted to write more. I've always had ideas. I have more ideas than I know what to do with. But getting the ideas out of my head onto a page in a form that I'm happy with always took me forever. I write so slowly. The time it would take me to write one of these articles was time I never have had between my personal life, work, family, numerous other creative interests, etc. It would take me at least a full week to write something that I might want to share, and most of the time, those writings never got published. For years, my longer-form ideas (longer for me, anyway) mostly just stayed in my head.

But now, thanks to AI, I dictate my first draft, and even some of my edits, using Wispr Flow, which I love. Then I work with Claude on the shape and drafting, and go back-and-forth to get my voice and style right until I can call it my article. It's still not fast; I spend hours on every one of these. But a few hours I can find. Weeks, not so much.

So I work with AI to write these articles. It's not a secret, and I want to address and acknowledge how AI is helping me get them out to the world.

"I use AI to write" is a loaded thing to say out loud right now. Most people might hear that and think that what I'm writing is generic, useless slop. But that's not what I'm doing. And as I started publishing every week, a question I've been carrying is, "how much of this is actually my writing?"

I asked Claude to do the math. Give me a number. 80/20 me to AI? 37/63? But how would it even figure that out? When I ask, it usually just tells me that the article is mine because they're my ideas.

So I decided to get specific. There's no single number until you define what you think authorship actually is.

Consider this: a legendary musician dictates their memoir to a ghostwriter. The ghostwriter types every word, smooths every sentence, orders the chapters, makes it into a cohesive narrative. Whose book is it? Nobody hesitates: it's the musician's. The life is theirs, the stories are theirs, the final say over every line is theirs. The ghostwriter produced 100% of the literal words and authored none of the book. So if who-typed-the-words isn't authorship, the question becomes what is? Answering that requires a rubric. So I made one.

How I score it

My rubric scores six layers of how an article comes together, then weights them by how much each one amounts to authorship of a personal essay.

Concept is the biggest: 40%

Who originated the core idea, the lived story, the reason the piece exists? On nearly everything I publish, that's me—endless ideas, remember?

Direction: 20%

Who drove? Who made the choices, killed weak fluff, redirected when a draft drifted? Rejecting and refining is real authorship, not passive approval, and it's almost entirely me.

Voice: 15%

Rhythm, humor, the texture of the sentences, but only the part beyond following my own rules. When the AI avoids a word because I've banned it, that's me, not the machine; only genuinely new texture I choose to keep counts on its side. And AI is still not funny, so if you're laughing, that's probably not Claude's joke.

Research: 10%

Who decided what needed outside support and what to look for. When I send the AI to find a study backing up something I already worked out by living it, that's me directing a research assistant; the retrieval is its labor, the idea was already mine. On most personal pieces there's no research at all, so this layer drops off.

Structure: 10%

The arrangement: openings, order, section breaks. AI is good at proposing these, but proposing options I pick from is closer to formatting than to direction, so it's weighted lightly.

Wording: 5%

Who produced the literal words that survived into the final version. This is the cleanest thing to measure and simultaneously the least meaningful, which is why ghostwriters are generally accepted and not credited. It's the floor, not the foundation.

And then there's a layer underneath all of it, which I call Foundation. Everything AI generates runs through a system I built over time within it: a memory of my voice and my banned words, a second brain full of my people and projects and history, the prior drafts, standing rules about how I like to work. The machine isn't writing from nowhere; it's writing from inside an accumulated record of my choices. So Foundation doesn't sit in the production math competing for points. It applies afterward, applying a little bit of additional credit to me, because that substrate is mine even when the AI is the one using it to generate.

What the number means

I ran some test pieces through this and they came out around 80 – 90% attributed to me. The first three I tested landed at 88, 87, and 84. AI's share sits mostly in drafting and arrangement, with a little in the research I sent it to do.

That AI makes up a small percentage—fifteen or maybe twenty points—but it's the entire difference between an article existing and not existing. At a 15% contribution, the machine isn't writing my articles—it's the reason I've written anything at all. Without it, I'm back to ideas that never leave my head. The percentage is small, but the consequence is huge.

This isn't a defense; I'm not out to prove the robot didn't do it. I just want to keep a clear record of a collaboration I'm glad exists, with a set of tools I think are genuinely wonderful, in a way that respects you enough to show the math. From here on, every Natural Frequencies piece will carry a score showing the George/AI split, a one-line note on what the AI actually did, and a link back to this page. You can always see how the sausage got made. Starting with this one.