What “Write in My Voice” Actually Means to Claude

The first version of my own voice profile didn’t include anything about how to handle personal anecdotes. I’d covered tone, sentence rhythm, the phrases I’d never use, the confidence level I’d claim on different topics. What I hadn’t thought to say was: if you don’t have a real story to tell, flag it instead of inventing one.

So Claude invented stories. Plausible ones — the kind of thing that could have happened, described in the kind of detail that makes something feel real. For most content, that’s jarring but recoverable. For Live Bold, my travel blog, it was a different problem entirely. That whole publication is built on the premise that the stories are real — that the chaos and the wrong turns and the specific weird details are things that actually happened. An AI-fabricated travel anecdote isn’t just off-brand. It’s the opposite of the entire point.

That gap in my profile taught me something about what “write in my voice” actually does and doesn’t do — and what Claude actually needs to get this right.

What most people mean when they say it

When someone tells Claude to write in their voice, they usually mean something like: match my tone. Be casual but professional. Don’t use jargon. Sound like a person, not a press release.

That’s describing voice from the outside — the surface characteristics that distinguish your writing from generic output. And Claude can do a reasonable job with that from a brief description alone. It’ll adjust the formality level, vary the sentence length, avoid the most obvious corporate language.

What it can’t do from a brief description is the deeper stuff: how you actually open a piece, what you’d cut in the second pass, how confident you sound on a topic you know well versus one you’re still working out. And it definitely can’t know the things that are specific to your content — like that every story in your travel blog has to be something that actually happened, or that you’d never stake out a position on a topic where you don’t have real experience to back it up.

Those things don’t come through unless you tell it. Explicitly.

What Claude actually needs

A useful voice profile goes well past tone. Here’s what actually moves the needle.

Your opening patterns. How do you start a piece? With yourself in context, usually — a specific moment, a situation you were in, a question you were thinking about. Not a statistic, not a provocative claim. If Claude doesn’t know this, it’ll default to whatever opening pattern tends to perform well generically, which is almost never how you’d actually start something.

Your sentence rhythm. Whether you tend toward longer explanatory sentences or short punchy ones, and how those mix. This is harder to describe in the abstract, which is why writing samples are so useful — Claude can learn the rhythm from examples faster than from description.

Your hard rules. The things you’d never write, stated explicitly. “I don’t take positions on topics where I don’t have direct experience.” “I don’t use performed vulnerability as a device.” “I don’t write dramatic one-liner sentences for emphasis.” These rules do a lot of work once they’re in the profile, because they’re the things Claude would otherwise default to.

Your anecdote rules. This is the one I missed the first time. If your writing relies on personal stories — and a lot of the best writing does — Claude needs to know what to do when it doesn’t have one. The right instruction is something like: if this section would benefit from a personal example and you don’t have one, flag it with a bracketed note rather than inventing something. That one instruction would have saved me from the fabricated travel stories entirely.

Your confidence calibration. How certain you sound depends on how much experience you actually have. If you’ve spent fifteen years in UX research, you can state things flatly. If you’ve been working with AI for two years and are still learning, you frame things as observations rather than conclusions. Claude needs to know which topics fall into which category, or it’ll apply a uniform confidence level to everything — which reads as either arrogant or hedging, depending on the topic.

Writing samples. Real ones. This is the piece that makes everything else work better, because Claude can learn your patterns from examples much faster than from description. Three to five pieces you’ve actually written, that actually sound like you, will do more for the output quality than any amount of written instruction.

The thing the brief instruction misses

“Write in my voice” is really an instruction about surface texture. It’s useful as far as it goes. What it doesn’t do is tell Claude anything about the rules that are specific to you — the hard nos, the things that would be wrong regardless of how well-written they are, the gap between topics you have authority on and topics you’re still figuring out.

Those things have to be explicit. And they have to be persistent — meaning they need to live somewhere Claude reads before it writes anything, not just in the prompt you type at the start of each conversation.

That’s what a voice profile is, structurally: a document that sits in your Claude Project instructions and gets read automatically at the start of every conversation. You write it once. From that point on, you’re not re-explaining yourself every time.

Building it vs. using a tool

If you want to build one yourself, the sections above are a reasonable starting outline. Start with your hard rules and your anecdote policy — those are the things that prevent the worst failures. Add opening patterns and confidence calibration next. Then writing samples. The tone and rhythm stuff will mostly take care of itself once those foundations are in place.

If you’d rather not build it from scratch, the Aligned Voice Profile is a short interview that walks through all of this and generates the skill file automatically. The interview asks the questions you’d need to answer anyway — it just structures them so you’re not staring at a blank document trying to figure out where to start.

Either way, the key is specificity. The more concrete your instructions, the less Claude has to guess — and the less you have to fix.

---

Related reading:
Why AI Content Never Quite Sounds Like You (And What to Do About It)
How to Write a Claude System Prompt for Your Voice (With Real Examples)
How to Use AI for Content Without Losing Your Authentic Voice