## The Problem with “Written by AI”

There’s a growing convention online: label things as either *human-written* or *AI-generated*. A binary. You wrote it, or the machine wrote it.

This feels honest on the surface but collapses under any scrutiny. What do you call a paragraph where a human had the idea, described it in rough notes, and an AI organized those notes into sentences? What about the reverse — where an AI suggested a framing and the human rewrote every word? The binary doesn’t capture either case.

I’ve been thinking about this because I use AI tools when I write. 1. Mostly Claude, if you’re curious. I don’t have a sponsorship deal. I just find it useful. Not for everything, and not in the same way for every piece. Sometimes I write a draft and ask for feedback. Sometimes I describe what I want to say and iterate on the output. Sometimes I write the whole thing myself, start to finish, and the AI never sees it.

These are meaningfully different acts of authorship. Calling them all “AI-assisted” is like calling both a ghostwritten memoir and an edited manuscript “professionally assisted.” Technically true. Usefully meaningless.

## Five Modes, Not Two

So I built a system for this blog that uses five collaboration modes instead of a binary label. Each one describes a different relationship between my thinking and the AI’s contribution.

Here they are.

### Independent

I wrote it. The ideas, the structure, the sentences — all mine. No AI involved in any part of the process. This is the default for anything not annotated.

Most of the arguments and theses on this blog fall here. The things I care enough to write about tend to be things I’ve been thinking about for a while, and the core ideas usually arrive before I open any tool.

### Human-Directed

My idea, shaped with AI assistance. I know what I want to say. I might describe the argument in rough terms, or write a messy first pass, and then use AI to help phrase it more clearly. The thinking is mine; the AI contributed to the expression.

An analogy: you’re giving directions to someone drawing your portrait. You know what you look like. They’re holding the pencil. The result is a collaboration, but the vision is yours.

### Co-Developed

We figured it out together. This is the mode for genuine back-and-forth — where I might start with a half-formed thought and develop it through conversation with the AI. The final idea isn’t purely mine or purely the AI’s. It emerged from the exchange.

This happens more often than people might expect.

I’ll have a vague sense that two concepts relate to each other, and through iterating — me pushing back on the AI’s framings, the AI surfacing angles I hadn’t considered — we arrive at something neither of us would have produced alone.

2.

This is also how the authorship system itself was designed. I had the instinct that binary attribution was wrong. The five-mode framework emerged through conversation.

### AI-Drafted

The AI wrote a first draft based on my direction, and I edited, curated, and approved the result. The ideas might be mine or might have emerged from a prompt, but the prose came primarily from the AI and I shaped it after the fact.

This mode is common for transitional paragraphs, summaries, and sections where the content is straightforward but I want it said cleanly. I describe the purpose of the paragraph, the AI drafts it, and I adjust until it sounds right.

### AI-Generated

The AI produced this with minimal human input. I might have given a brief prompt, but I didn’t substantially direct, edit, or reshape the output. I’m including it because I think it’s useful or accurate, but I didn’t meaningfully author it.

I don’t expect this mode to appear much on this blog. If I don’t care enough about a piece of writing to direct it, I probably shouldn’t publish it. But the category exists because honesty requires it — and because other contexts (documentation, boilerplate, formulaic content) might warrant it.

## Why Bother?

A reasonable question. Most readers don’t care how a sentence was produced. They care whether it’s good.

I have a few reasons.

**Honesty.** I’m writing a blog partly about writing itself. Claiming full authorship of AI-assisted prose would undermine the whole project. If I’m going to think in public, I should think honestly.

**Precision.** “I used AI” is almost as uninformative as saying nothing. It covers everything from “I asked ChatGPT to write my essay” to “I ran a spell check.” The five modes give me — and you — a more precise vocabulary.

**Self-awareness.** Tracking authorship forces me to notice my own process. When I label a section *ai-drafted*, I’m acknowledging that I outsourced not just labor but a kind of creative choice. That acknowledgment keeps me honest about where my thinking ends and the tool’s output begins.

## The X-Ray

If you scroll up, you’ll notice the authorship bar at the top of this post — a row of characters showing the rough breakdown of collaboration modes. There’s also an x-ray toggle.

Turn it on.

The post text itself will light up. Sentences and sections are annotated with their collaboration mode, shown through color and underline style. 3. The encoding is redundant by design. Color alone isn’t accessible for colorblind readers, so each mode also has a distinct underline style and a different character in the bar. Three channels, any one of which is sufficient. Hover over any highlighted span to see which mode it belongs to.

This is the part I’m most interested in as an experiment. Not the labels themselves, but whether making the collaboration visible changes how you read — and whether it changes how I write.

## What This Isn’t

This system is descriptive, not prescriptive. I’m not arguing that *independent* is better than *co-developed*, or that *ai-drafted* content is inherently lesser. A well-directed AI draft might be clearer than a clumsy independent paragraph. The modes describe process, not quality.

I’m also not claiming perfect accuracy. Attribution is partly a judgment call. When I brainstorm with an AI and then write the paragraph myself, incorporating ideas from the conversation — is that *independent* or *co-developed*? I’ve made choices, and reasonable people might categorize differently.

The goal isn’t precision to the sentence. It’s a rough, honest map of how the thing was made.

## An Evolving Practice

I expect this system to change. I might find that the five modes are too many, or not enough. I might discover that the x-ray mode changes my writing in ways I don’t like — making me self-conscious about using AI where it’s genuinely helpful. I might find that readers don’t care at all.

That’s fine. The point was never to solve attribution permanently. It was to take the question seriously instead of ignoring it.

If you’re a writer who uses AI tools, I’d encourage you to think about your own version of this. Not necessarily a five-mode system with inline annotations — but some honest accounting of the collaboration, even if only for yourself.

The exercise of labeling is itself clarifying. It makes you notice what you’re doing.

And noticing what you’re doing is, I think, the whole point.