Why AI Content Never Quite Sounds Like You (And What to Do About It)
I’ve been using AI to write content for a couple of years now. Posts, emails, blog articles, client-facing copy. And for most of that time, the output was usable — technically correct, reasonably structured — but I’d spend as much time editing it as I would have spent writing the thing myself.
For a while I assumed I just wasn’t prompting well enough. That if I gave Claude better instructions, it would get closer. So I got more specific. I’d write longer prompts, more detailed briefs, examples of what I was going for. It helped, sometimes. But there was still this gap I couldn’t close — this version of my voice that was almost right but not quite, like hearing a cover of a song where all the notes are correct but the feeling is off.
What I eventually figured out is that the problem wasn’t the prompting. It was that AI doesn’t know how you write. Not really.
What “write in my voice” actually does
When most people tell Claude to write in their voice, they mean it the way you’d give a direction to a copywriter you just hired — a general sense of the vibe they’re going for. And Claude does what it can with that. It adjusts the formality level, maybe the sentence length, the overall register.
But that’s describing voice from the outside. What actually makes someone’s writing theirs is a lot more specific: the sentences they’d never write, the way they open a piece, the things they reflexively cut, how confident they sound on a topic where they have ten years of experience versus one where they’re still figuring it out. None of that is captured by “keep it conversational.”
At first it was emojis — everyone was scrubbing those out of ChatGPT output before it became too obvious. Then it got more subtle. For me it was phrases like “and honestly?” and “quietly…” — that specific kind of performed casualness that started showing up everywhere. And then came the wave of Medium posts and Substack newsletters and LinkedIn threads cataloguing every tell, every phrase to hunt down and delete if you didn’t want people to know you’d used AI. Which is its own kind of exhausting — editing your content against a checklist of red flags instead of just editing it to sound like you.
This is why AI output requires so much editing. It’s not that the AI is bad at writing — it’s that it’s writing from a generic baseline, and you’re spending all your editing time moving it from that baseline to your actual voice. Every time. For every piece.
What actually fixes it
The answer isn’t better prompts in the moment. It’s giving Claude a persistent set of instructions that it reads before it writes anything — a voice profile, essentially. A document that captures the real specifics: your patterns, your hard rules, what you’d say and what you’d never say, how you open and close a piece, examples from your actual writing.
When that’s in place, the baseline shifts. Claude starts from your voice instead of starting from a generic professional default and waiting for you to correct it back toward yourself.
The difference in editing time is real. The first time I used my own profile, what I noticed was that it sounded like me — like exactly what I would have written if I’d had all day to write and edit. I still reviewed it. But I wasn’t correcting voice anymore. I was just editing for content.
How to build one
If you want to do this yourself, here’s what actually needs to go into it.
Your patterns. How you open a piece. How you close it. Whether you use headers, how you use them. Your sentence rhythm — if you tend toward long explanatory sentences or short punchy ones, or some mix of both. Whether you think in lists or paragraphs.
Your hard rules. The phrases you’d never use. The framing you’d never take. The level of confidence you’d claim on different topics. This is often easier to articulate in the negative — “I don’t do hot takes,” “I’d never write a sentence that starts with ‘here’s the uncomfortable truth,’” “I don’t use exclamation points in professional writing.”
Your tone calibration. Who you’re writing for, what you actually know versus what you’re still learning, and how that should show up. The way someone sounds when they’re the expert in the room is different from how they sound when they’re figuring something out alongside the reader — and both are legitimate, but they shouldn’t be mixed up.
Writing samples. Real ones, that actually sound like you. This is the most important part. The AI can learn your patterns from description, but it learns them much faster and more accurately from examples of the real thing.
Once you’ve got all of that, you create a Project in Claude, paste the whole thing into the project instructions, and that’s it. Every conversation you start inside that project will use those instructions automatically.
The shortcut
I built a tool that does this for you — a short interview that asks the questions you’d need to answer to build a voice profile, and then generates the skill file automatically. It’s $37 and takes about fifteen minutes. You get the file by email and install it yourself.
It’s at myvoice.alignedxd.com if you want to try it.
If you’d rather build your own from scratch, everything above is enough to get started. The main thing is to be specific — the more concrete and particular the instructions, the less editing you’ll be doing on the other side.
Related reading:
The Hidden Cost of Editing AI Content You Did Not Train
What “Write in My Voice” Actually Means to Claude
How to Write a Claude System Prompt for Your Voice (With Real Examples)
8 Comments