AI-Assisted Writing: Where I Draw the Line

I write almost every day. Emails to my team, threads in Slack, posts for my newsletter (12,500+ readers), blog posts like this one, pitch decks, even the occasional thread on Twitter. Writing matters in my world. And for the last eighteen months, I’ve been using AI to help me do it better.

But “using AI to help” is doing a lot of work in that sentence. There’s a meaningful difference between AI-assisted writing and just using AI to generate content. The first is a tool. The second is outsourcing your voice. I’ve learned to recognize the line between them, and I want to talk about where I draw it and why.

Key Points

AI is invaluable for brainstorming, outlining, drafting, and editing—but not for the core voice and perspective that makes writing yours
The real risk in the AI era isn’t plagiarism; it’s publishing mediocre, interchangeable content because the tool makes it too easy
Your unique perspective is now your most defensible competitive advantage as a writer

The Things AI Does Well

When I sit down to write my newsletter, I don’t start from a blank page anymore. I start with a structure. I’ll open Claude and outline what I want to say: the story I’m leading with, the 2-3 insights I’ve had this week, a contrarian take, maybe a book or tool recommendation. That outline takes fifteen minutes and saves me an hour of staring at the screen.

Here’s where AI genuinely shines: scaffolding. Creating structure. Taking a messy idea and saying, “What’s the logical flow here?” I’ll generate three different outlines, pick the one that resonates, and then discard the AI’s drafting entirely. I only wanted the skeleton.

The editing stage is where I get real value, too. I’ll have a draft—my draft, written in my voice—and I’ll ask Claude to tighten it. “Cut 20% of the words. Remove clichés. Make it punchier.” It’s like having a trusted editor who’s available at 11 p.m. on a Tuesday. The rewrites are often too aggressive (Claude loves removing nuance), but even a rejected suggestion makes me reconsider my original sentence.

Research synthesis is another win. If I’m writing about a trend I’ve noticed, I’ll ask AI to pull together what’s been written about it, what the arguments are, where the consensus breaks down. That would take me three hours of browser tabs and Google Scholar. AI does it in thirty seconds. Then I synthesize it in my own way.

Brainstorming, too. “I have this half-formed opinion about how companies misuse data. Give me five different angles to explore.” The AI’s angles are sometimes mediocre. But one of them usually cracks open the idea I was sitting with. That’s the job.

All of these are force multipliers. They compress time. They remove friction. They’re labor-saving, not mind-replacing.

The Things Only You Can Do

The opinions are yours. The stories are yours. The voice is yours. These are non-delegable.

Stephen King wrote that good writing is the word-by-word refusal to accept mediocrity. I think he’s right. And no AI is refusing anything. It’s suggesting. It’s generating. It’s optimizing. But it’s not standing at the line saying, “No, that’s not good enough. I believe something different.”

I use a simple test: If I removed the AI assistance from my workflow, would this piece still have my perspective? If the answer is no, I haven’t actually written it yet. I’ve just narrated a draft.

So the deep editing is mine. The final read-through where I’m checking not just for clarity but for truth—is this actually what I believe? Does this argument hold up? Am I bullshitting here?—that’s mine. The personal stories that anchor an essay are mine. I’m not asking Claude to invent a story about the time I almost lost a client because I over-promised. That happened to me. I own that. I write it.

And the voice—the rhythm and texture and personality that makes something feel like me rather than the AI—that’s mine to protect. I’ve developed a way of writing over fifteen years: short sentences sometimes. Contractions. A tendency toward directness even when it’s uncomfortable. Specific details over broad claims. These are choices I’ve made. They’re not generic.

When I see AI-generated content, it has a texture. It’s often correct. Rarely wrong. But it’s smooth in a way that human writing isn’t. It’s the absence of struggle. It’s consensus. And that’s exactly the problem.

The Real Risk: Mediocrity at Scale

I’m not worried about plagiarism. I’m not worried about AI stealing my voice. What I’m worried about is a world where everyone’s writing looks like everyone else’s because everyone’s using the same tools with the same settings and the same prompts. The easy path becomes the default.

A year ago, my inbox started filling with pitches from “founders” (I use quotes because many weren’t pitching ideas; they were pitching AI-generated content) who’d used ChatGPT to write their entire value proposition. The writing was clean. Sometimes witty. But it had no point of view. No risk. No specificity. It read like a newsletter template.

That’s the trap. Not that the writing is bad, but that it’s frictionless. It’s too easy to publish something that’s ninety percent good and call it done. And when everyone can do that, when the barrier to “publishable” drops to just being willing to press send, then volume becomes the currency, not quality.

George Orwell said that political speech and writing are largely the defense of the indefensible. I think he’s onto something broader: good writing is often about defending a position that costs you something to take. It’s risky. It’s specific. It’s not consensus.

AI doesn’t do risk well. It optimizes for agreement, for mainstream interpretation, for the interpretation that’s been written about most. If you want to say something that challenges that—something contrarian or personal or weird—you have to do it yourself.

My Actual Workflow

This is the meta part: this essay was written with AI assistance. Here’s how:

I pitched Claude the idea: “I want to write about using AI for writing without outsourcing voice. Here’s what I think…” Then I sent an outline. Claude expanded it, suggested subheadings and transitions. I rewrote the entire thing in my voice, using the outline. Then I asked Claude to tighten weak sentences and catch logical gaps. Then I rewrote again—the parts that felt impersonal got more specific. The parts that were too clever became clearer.

The Orwell quote, the King quote, the criticism of AI-generated pitches—those are all mine. The framework (what AI does well vs. what only humans do) is mine. The conviction underneath it is definitely mine. But the prose was tightened by an AI. The structure was suggested by an AI. Some of the transitions are better because of feedback from an AI.

Did I cheat? Did I plagiarize? Did I outsource my authenticity?

I don’t think so. But I also can’t claim I did this entirely on my own. And I think honesty matters here.

Why Voice Matters More Now, Not Less

There’s a paradox in the AI era: as the barrier to publishing falls, as anyone can generate text, the value of having an actual perspective goes up, not down.

Everyone can write a hot take now. Everyone can produce five hundred words on any topic. So the differentiator isn’t being able to write. It’s having something to say. It’s having a point of view that’s rooted enough in your actual experience that it can’t be generated.

I think about this with my newsletter. The writing is better because I use AI. The essays are tighter, the arguments clearer, the structure sharper. But what people subscribe for is the perspective. They want to know what I’m thinking about the intersection of business and technology and building in public. They can get better-written essays on those topics from plenty of places. But they can’t get my specific experience or my specific point of view anywhere else.

That’s always been true, but it’s more true now.

If you’re still learning to write, AI might be dangerous. You might optimize for efficiency over learning. You might use it as a crutch instead of a tool. But if you’re writing from a place of actual knowledge and experience, AI becomes a force multiplier.

I’m not convinced I’m drawing the line perfectly. I think about it a lot. But I know the direction I want to pull: toward using AI to amplify my voice, not replace it. Toward using it to do the work that’s less important so I can focus on the work that’s entirely mine.

The challenge, I think, is staying aware of the seam. Knowing where the machine ends and you begin. And fighting the constant pressure toward frictionlessness—the quiet whisper that says, “Why rewrite this? It’s already ninety percent good. Just ship it.”

That whisper kills voice. That’s the real line to watch.

Read more on writing and AI: