Writing That Sounds Like You (Not Like AI): A Practical Guide
To make AI-assisted writing sound like you, give Claude a small voice kit (three strong samples and a short description of your style), use it as your standing brief at the start of every session, then run a fast surgical edit that strips the AI tells and adds one thing only you could have written. That's usually enough to get to publishable in under 30 minutes.
Most people can hear it now. AI writing lands in a weird middle ground: technically fine, oddly polite, the same tone from top to bottom, padded with phrases no one says out loud. "It is important to note that..." "In this rapidly changing environment..." "This comprehensive overview..." If you've ever sent a draft and had someone ask, "Did you use AI for this?", they weren't guessing. They heard the register.
The useful news is that those tells are consistent, which makes them fixable. Using Claude isn't cutting corners; handing in generic, placeless prose is. If you're in your 40s, 50s, or early 60s, your track record is the product. Clients are buying the judgment in your head. The whole point of AI is to let you ship more of that judgment without sanding off the fingerprints that prove it's actually yours.
Why does AI writing sound so generic?
Because it's literally trained to aim for the middle. Claude has read millions of blog posts, memos, and decks. Ask it for a "professional" draft and it averages what those look like and hands you the median: clean structure, neutral tone, no sharp edges, minimal risk.
That's the opposite of how a seasoned practitioner writes. Your voice is the accumulation of 20 to 30 years of specific things: the deals that blew up, the board decks that went sideways, the bad incentives you watched play out. A 55-year-old GC, FP&A lead, or consultant isn't just stringing sentences together; they're trying to prevent the next avoidable mess. That lived intent leaks into word choice, rhythm, and what you refuse to say. AI can't guess it on its own. You have to hand it a starting point and keep it honest in the edit.
How do you calibrate Claude to your voice?
You calibrate Claude by giving it a small, reusable voice kit: three samples of writing you're proud of and a short plain-English description of how you actually sound. That becomes your default brief at the start of any serious writing session.
Here's the structure I see work with senior operators. First, pick the right three samples. Choose pieces that already sound like you on a good day: a client email that landed, a memo your team still forwards around, a post that drew real comments instead of polite likes. Length matters less than clarity of tone, and avoid heavily edited marketing copy, which is usually your voice with the life squeezed out.
Then hand Claude a clear brief. Paste the samples and say something like: "These are three samples of my writing. Study the voice, the sentence rhythm, and how I explain things. I'm a senior practitioner; I don't hedge much, I use concrete examples, and I prefer short, direct sentences. When I ask you to write, model this voice. Don't copy the content."
Then run a live correction loop. Ask for something short, a 200-word client note or a four-bullet summary of a deck, and read it the way a peer would. Then respond bluntly: "Too formal. I'd never call something 'best-in-class' or use 'leverage' as a verb. Make the sentences 20 to 30% shorter." Or: "This reads like a marketing blog. Strip the hype, keep only what a CFO would care about." Two or three cycles in one session usually shifts the baseline for good.
What's a simple framework to keep your voice consistent?
Use a dead-simple rule I call the 3P Voice Framework: Profile, Prompt, Pass. You can run it end-to-end in under half an hour.
- Profile (5 to 10 minutes): Build your voice kit once. Three samples plus a five-to-seven-line description of your style (direct or indirect, dry or warm, level of detail, tolerance for jargon). Save it in a notes app or as a Claude project instruction.
- Prompt (5 minutes): At the start of each new conversation, paste the profile and samples, then the task: "Using this voice, draft a 600-word note to the executive team on why we're moving from ad hoc reporting to a monthly FP&A pack."
- Pass (10 to 15 minutes): Run the fixed editing pass below to strip AI tells, reintroduce your quirks, and add one lived detail. This is where you earn your fee. Claude drafts; you author.
Once the profile exists you're not fiddling with prompts every time. You're applying the same small system to every memo, post, and slide note until it feels automatic.
What does a high-leverage editing pass look like?
A good edit isn't a rewrite. It's attacking the handful of patterns that make people smell "AI" in the first two paragraphs. Here's a lean pass that fits a 10-to-15-minute window and works whether you're 42 or 62.
- Kill the throat-clearers. Delete openings like "In this piece, we will explore..." Start with the actual point: "We're changing the comp plan because the current one rewards the wrong behavior." Your reader is busy. They'll thank you.
- Swap out AI-register vocabulary. Scan for the words you'd never say to a peer at a bar: "leverage" as a default verb, "holistic," "comprehensive," "best-in-class," "landscape," anything starting with "del-" that means "dig into." Replace with plain equivalents or just cut the modifier.
- Break the metronome. AI writes sentence-sentence-sentence, all in the same band of length. Read it aloud. Where it drones, change the rhythm. Make one sentence very short. Let the next one breathe.
- Delete templated pivots and recaps. "While X is true, it's also important to consider Y" and closing paragraphs that re-summarize the whole piece are boilerplate. If your reader made it to the end, they know what you said. Cut the recap and land on one concrete next step.
- Add one thing only you could know. The big one. Drop in a lived detail: "In a budget review for an $80M services firm, the only slide the CEO cared about was..." Claude can mimic professional tone; it can't fabricate your actual scar tissue.
A 52-year-old FP&A head I traded notes with tried this on a board-prep memo. Claude's first draft was fine but forgettable. He spent 12 minutes on the pass above and added one line: "The last time we chased a top-line target this way, we spent 14 months unwinding bad deals." That single sentence changed the room. No AI would have written it, because no AI was in that room four years ago.
What structural tells should you delete on sight?
Beyond individual words, a few structures reliably out the machine. Learn to spot and kill them:
- The three-part list where every item opens the same way ("First... Second... Third..."). Fine once; a crutch when it's everywhere.
- The false pivot: "While X is true, it's also worth considering Y." AI uses it as a template for adding nuance. Real writers just say Y.
- The recap paragraph that restates what you just said. Your reader was there. Cut it.
- The hedge stack: "It may be worth considering that in some cases it might be possible to..." Real professional writing takes a position.
I'll admit I was smug about this for a while, certain I could spot AI writing every time. Then I caught two of these tells in a memo I'd drafted by hand on a tired afternoon. The register is contagious. Read enough of it and you start producing it yourself, which is exactly why the deliberate edit matters even when no AI touched the draft.
| Tell | How it reads | The human fix |
|---|---|---|
| Scene-setting opener | "In this analysis, we will examine..." | Delete it; open with the point |
| AI-register vocabulary | "leverage," "holistic," "comprehensive," "best-in-class" | Plain words: "use," "look at" |
| Even cadence | Every sentence the same length | One short. Then a longer, looser one. |
| False pivot | "While X is true, consider Y..." | Just say Y |
| Recap ending | Restates the whole post | End on a concrete move, not a summary |
The Only-I test before you hit send
Before anything important goes out with AI help, run one check I call the Only-I test: does this draft contain at least one sentence only I could have written? A number from my own practice, a contrarian read of the situation, a detail from a client I've actually sat across from. If the honest answer is no, the piece isn't done, no matter how clean it reads. Clean and anonymous is precisely the failure mode. The 3P Framework gets you the draft; the Only-I test is what makes it yours.
Common questions, answered straight
How long until Claude sounds like me? With a few good samples and two or three explicit feedback rounds, you'll see real improvement in one session. Getting to where the output needs minimal editing usually takes three or four sessions across different writing types.
What if I don't have good samples? Start with what you have; voice patterns survive even in rough drafts. If you truly have nothing, describe your voice in words: "direct, dry, specific, low on abstraction, high on examples."
Should I disclose that writing is AI-assisted? In most professional contexts, no. You're the author and the AI is a drafting tool, like a ghost-writer. Still, check your industry's or publication's policy; some have explicit rules.
It keeps reverting to its default style. Why? Each conversation starts fresh. Your calibration doesn't persist unless you paste it back at the start. Keep your voice kit saved and lead every session with it.
What's hardest to make sound like me? Short-form posts and emails, because your voice is most concentrated there. Long-form is easier; there's more room for the specific content only you would include.
So here's the move you can make today: build the Profile once, three of your strongest samples plus a few honest lines about how you sound, and never start a writing session without pasting it in. Then run the Pass and the Only-I test on the way out. Do that and Claude becomes what it should be, a fast first-drafter that leaves the part that matters, your judgment and your voice, firmly to you.