Your Name Is Still on It
I scroll through my LinkedIn feed and I see a pattern. Sentences that break off mid-thought with that particular kind of dash. Paragraphs that open with "It's worth noting that." Closers that land with a neat philosophical symmetry no one actually speaks in. I recognize the fingerprints immediately.
When I do, I don't stop reading. I skim faster, or I lean in with a more critical eye. Sometimes I take the piece, drop it into a fresh chat with a completely neutral "what do you think," and wait for the verdict. Whatever substance is or isn't there comes back accurately assessed. The effect on how I read that author going forward is permanent.
I use AI every day. I've used it to write articles, think through problems, workshop ideas I couldn't shake, draft things I didn't have time to write from scratch. The output is mine. I can defend every claim, explain every choice, speak to every point - because the thinking behind it was mine before the tool touched the keyboard.
And right there - in an early draft of this article - I found a Claudism. The sentence read: "That's the distinction worth drawing." The word "worth" is a tell. It announces to the reader that something deserves their attention instead of just demonstrating it. It's on my kill list. And it slipped through anyway.
I caught it this time. I don't catch all of them. That's exactly what this article is about.
The two tests
There's a difference between asking AI to write for you and using AI to write better. The first produces text. The second produces thinking.
The first test: if someone asked you about a specific point in your article, could you go deeper? Not recite the article back - actually go deeper. Draw on the experience or reasoning that led you there. Connect it to something you know that didn't make the cut. If the answer is no, the AI wrote the article. If yes, the AI helped you write it.
The second test: if you lost access to the tool tomorrow and had an article due, could you write it yourself? It would take longer - maybe four or five times longer - but could you get there? If your honest answer is yes, you're a writer using AI. If the answer is no, you've been using it as a ghostwriter and calling the work yours.
The first test is about ideas. The second is about voice.
The ideas have to come from you
My articles come from the work I do - enterprise AI, the specific friction points and breakthroughs that happen inside a large organization. I'm not asking the AI what would resonate with my audience. I have a running list of things I've lived through that I want to write about. The ideas are there before the tool enters the picture.
The workshop
Once I have an idea worth developing, I spend time - sometimes a lot of it - in conversation before I write a word. On the iOS app, voice mode, walking around the house or the yard or commuting to work. Talking through what I want to say, what the key points are, what I want readers to take away.
And I'm deliberately verbose about it. Every usual filter is off. I drench the conversation in context - every detail, every tangent, every piece of background that might matter. It runs long and it covers everything, because the draft that comes out the other side is only as good as the thinking that went into the conversation.
The AI reflects back what I said - but how it articulates something is sometimes slightly different from how I was thinking about it, and that opens a new angle. I run at that angle, we go back and forth, and it either strengthens the original point, adds a new one, changes my mind, or gets cut. It's iterative in a way that would exhaust a person and doesn't get impatient.
There's useful friction in the dynamic too. I'm a verbose thinker by nature. The AI, left to its own instincts, wants to wrap up and move on. When I push back and say "no, let's stay here a minute," I'm usually doing it because something real is left to mine. That collision is where a lot of the clarity comes from.
That collision is also configurable in a way no human collaborator could commit to. You can configure the AI as a specific counterbalance to your own tendencies. Rush past details - have it draw them out. Over-qualify everything - have it push you toward directness. Whatever the gap in your process, the tool can fill it without ego and without forgetting. Understand your own writing weaknesses well enough to instruct against them.
Mid-conversation, you can also drop in research - tell it to go verify something, find supporting data - and pick up where you left off. Drop in a screenshot when describing something visual would take longer than showing it. The workshop isn't linear and doesn't have to be.
Build a style guide and a kill list
When the AI drafts something in my voice, it works against a style reference trained on real samples of my writing - emails, documents, articles, anything that captures how I actually sound. That guide gets refined over time: every time a draft sounds wrong and I catch it, the guide updates. Living document, not a one-time setup.
The companion to that is a kill list of the tool's own verbal tics - things it reaches for by default that I would never say. There are particular kinds of dashes. Words and phrases like "it's worth noting that," "load-bearing," "what I keep coming back to." Also "surface" - which I'll admit I actually like. It's a hip version of "interface." I still put it on the list. I've collected more than seventy of these, cross-referenced against research including Will Francis's field guide on AI writing tells and Wikipedia's Signs of AI Writing page. Every draft gets scrubbed against that list before I read it.
And there are almost certainly some in this article. Probably a few in previous ones too. That's not hypocrisy - that's how I built the list. Every entry came from finding it in something I'd already published or was about to. I don't call it failure. I call it experience.
It isn't a perfect filter. It's constant, iterative pressure so that my voice stands out and the tool's voice disappears.
The draft is never done on the first pass
A first draft, after a long workshop conversation, usually lands somewhere between forty-five and sixty percent of the way there. It's a distillation of the conversation, not a transcript of it. It picks angles, makes choices about emphasis, and it doesn't always get those right.
The revision loop is where authorship actually happens. Reading it as if you're hearing yourself talk. Catching the sentences that don't land like you. Correcting anything factually wrong - the tool fills gaps by making reasonable assumptions, and sometimes those assumptions miss. You have to know your own story well enough to catch it.
And if you're citing anything - a statistic, a quote, a reference - verify it yourself. The embarrassment of getting that wrong in public is permanent.
What it actually is
The AI doesn't make you a writer. It makes a writer faster, and it covers the distance between good and great - that last stretch that takes the most time and is hardest to sustain on a consistent basis. Decades of compounding interest finally paying out. What separates a professional from someone who's almost there. The ending that makes a movie worth the two hours. The catalyst that gets you to your best self not just when conditions are right, but every time.
Without it, you're still you. You can still write. It just takes longer and costs more energy to get there.
But that's also what exposes the counterfeit version. If the gap you're bridging goes from bad to mediocre, or mediocre to decent, the AI gives you speed and a polish pass. The rest of the distance has to be covered by a voice and a perspective that's actually there. If they aren't, there's nothing to amplify. What comes out is technically clean and entirely hollow - and people feel it, even when they can't name what's wrong.
Your name is still on it. Make sure you're in it.