How I Created My Voice Reference With AI (And Why It's Not Cheating)

I've been writing online since 2011. Fourteen years of blog posts, technical tutorials, personal reflections, and half-finished drafts sitting in folders I'm afraid to open.

And yet, if you asked me to describe my "voice" as a writer, I'd stumble.

I know what I don't sound like. I'm not academic. I'm not a motivational poster. I'm not trying to sell you a course. But articulating what I am? That's harder.

So I did something that felt a little uncomfortable and exciting at the same time: I asked an AI to analyze my writing and help me create a voice reference document.

This isn't a new idea. Creators have been training AI on their voice for a few years now. Tiago Forte—whose PARA method heavily influences how I organize my own knowledge system—wrote about creating an AI style guide back in 2023. I independently dove into it back then while I was working with a company called Mach49, but never that deeply (the quality wasn't up to par IMO).

But in revisiting this again, what surprised me wasn't the process. It was what I discovered about myself.

The Setup

I've been building what I call my "digital self"—a system of interconnected projects, notes, and tools that externalize how I think and work. A big part of it stems from my original work from Self Engineer, but it's also built on Forte's PARA method (Projects, Areas, Resources, Archives) combined with the CODE framework (Capture, Organize, Distill, Express) and my own improvements. The whole thing lives in markdown files and git repos because—engineer brain—I want version control on my life.

One piece of this system is a voice reference file. The idea seemed simple: create a document that could guide AI tools when helping me draft or reformat content. A style guide for machines.

But I had a problem. The existing document was thin and generic. It said things like "be approachable" and "avoid jargon" without capturing what actually makes my writing mine.

I needed to debug my own voice.

The Process

The general approach is well-documented: feed AI your writing samples, ask it to analyze patterns, generate a style guide. Most guides recommend at least three samples totaling 1,000+ words.

Here's what made my session different:

1. I pulled up my actual published articles—not my "best" work.

Not the polished about page. The actual posts. Technical tutorials about JavaScript query strings. Personal essays about why creative people struggle. A piece about migrating from Evernote to Obsidian that somehow became a meditation on digital ownership.

2. I asked a specific question: do I have one voice or two?

I write about two main things: software engineering and personal growth. Most brand voice guides would tell you to create separate personas. I wanted to know if that was actually true for me.

3. I looked for the voice in the output, not my intentions.

Not what I thought I sounded like. What the words on the page revealed.

What we found was fascinating.

What I Discovered

I have one voice, not two.

My technical writing and personal growth writing share the same core identity. The same first-person perspective. The same tendency toward honesty about trade-offs and struggles. The same conversational asides.

What changes is the register—the formality level, the vulnerability depth, the use of metaphors versus code examples. But the person behind the words is consistent.

This matters because my whole "Self Engineer" concept is about applying engineering thinking to personal growth. If I had two separate voices, it would undermine the integration that makes my perspective useful. The engineer who philosophizes. The philosopher who builds. That's the value proposition.

Most voice guides miss this. They tell you to create different personas for different contexts. But for some of us, the power is in the integration.

I have strong opinions I wasn't fully articulating.

Buried in old notes from incomplete projects, I found a list of contrarian views I hold and already talk about:

Self-help is symptomatic, not systematic
Habits don't work alone—life's gravity pulls you back
The details of someone's success are not repeatable; only principles transfer
There's no "best self"—that phrase implies you're done evolving

These opinions were floating around in my head and in disparate notes but rarely made it into my public writing (though they existed privately in my Self Engineer book drafts). Seeing them listed out again here was reinforcing my roots and experience. Now they're documented as things I should be saying more often.

I have anti-patterns I fall into.

Phrases I use when I'm being lazy. Structures that bury the lede. A tendency toward long paragraphs when short ones would hit harder.

Having these documented means I can catch them. Or an AI can catch them for me.

The Mirror Effect

Most articles about AI voice training focus on the output—making AI sound like you. They miss something bigger: creating this document wasn't just useful for AI assistance. It was useful for me.

Reading my own writing analyzed back to me felt like a therapy session. "You're honest about your struggles. You invite dialogue rather than broadcasting. You apply engineering metaphors to everything, even when you're not writing about code."

Yeah. That's me.

There's something powerful about having your identity reflected back in a structured way. It's not narcissism—it's self-awareness. The kind of self-awareness that makes it easier to show up authentically because you actually know what "authentically" means for you.

I've journaled for years. I've done personality assessments (CliftonStrengths, Enneagram, all of it). But this felt different.

Those tools ask you to introspect. This process derives identity from output—what I've actually written, not what I think I am. There's a difference between believing you're a certain kind of writer and having evidence that you are.

Cal Newport talks about the importance of deep work and protecting your attention. I'd add this: sometimes you need to deeply look at what you've already produced. The artifacts of your focused work contain information about who you are.

How to Create Your Own Voice Reference

If you want to do this yourself, here's the process:

1. Gather your actual writing.

Not what you wish you'd written. Not your best work curated. Pull together a representative sample across different contexts. Emails count. Social posts count. That blog you abandoned counts.

Forte recommends choosing work that "sounds like you at your best." I'd modify that: choose work that sounds like you, period. Include the pieces where you were struggling. Your voice shows up there too.

2. Ask someone (or something) to analyze patterns.

You're too close to see yourself clearly. An AI works well for this because it has no agenda and won't spare your feelings. A trusted friend who's read your work can also help.

Prompt something like: "Analyze these pieces for consistent voice patterns. What stays the same across contexts? What changes? What phrases or structures repeat?"

According to HyperWrite's documentation, good AI analysis will identify your sentence structure, word choice, punctuation usage, and overall tone—then synthesize patterns across samples.

3. Identify what makes you, you.

Look for:

Sentence rhythms and lengths
How you open and close pieces
Your relationship with vulnerability
Recurring metaphors or frameworks
Strong opinions that show up implicitly

4. Document the anti-patterns.

What do you do when you're being lazy? What phrases creep in that don't sound like you? What structures fail to serve your ideas?

5. Create the reference document.

Structure it however works for you. Mine includes:

Core identity (who I am as a writer)
Voice attributes (what's always present)
Domain adjustments (how register changes by context)
Platform guidelines (LinkedIn vs Twitter vs long-form)
Anti-patterns to avoid
Exemplar passages from actual work

Taking It Further: Modular Voice Files

After creating the initial voice reference, I realized a single document wasn't enough.

I write differently for LinkedIn than I do for long-form blog posts. Twitter demands a different rhythm than Substack. The core identity stays the same, but the register—the formality level, the structure, the expectations of each platform—varies.

So I split my voice reference into modular files:

Core voice - Identity, attributes, anti-patterns (always loaded)
LinkedIn - Measured expertise, flowing sentences, balanced comparisons
Twitter - Punchy, direct, hook-driven
Substack - Personal, intimate, long-form depth
Website - Portfolio polish, evergreen focus, SEO-aware headings

Now when I'm formatting content, I load only what's relevant. Writing a LinkedIn post? Core + LinkedIn guidelines. Blog article? Core + Website guidelines.

This modular approach has a side benefit: it forces you to articulate why your voice shifts between platforms. Not just that it does, but why it should. LinkedIn readers expect professional insight; Twitter rewards contrarian takes; Substack subscribers want the intimate version of you they can't get elsewhere.

Making It Automatic with Claude Code Skills

Here's where it gets interesting for the technically inclined.

I use Claude Code as my primary AI coding assistant (as well as Cursor). It has a feature called "Skills"—essentially, specialized agents that get automatically invoked when certain conditions are met.

I created a voice-formatter skill. When Claude sees a markdown file with platforms: [linkedin] in the frontmatter, it automatically loads my voice guidelines and applies them during editing. No manual prompting required.

The skill lives in .claude/skills/voice-formatter/ and includes symlinks to my voice files (which live in my PARA-organized notes system). The source of truth stays where it belongs—in my knowledge management system—while the AI tool gets access through symlinks.

.claude/skills/voice-formatter/
├── SKILL.md          # Trigger conditions and process
├── core.md           # → symlink to 2-areas/content/voice/core.md
├── linkedin.md       # → symlink to 2-areas/content/voice/linkedin.md
├── twitter.md        # → etc.
├── substack.md
├── website.md
└── examples.md       # Calibration passages

The SKILL.md file tells Claude when to activate: "Use when editing markdown files with platforms: in frontmatter, when asked to format content, or when reviewing draft articles for voice consistency."

This is the kind of automation that compounds. Every piece of content I write now gets checked against my own documented voice patterns without me having to think about it.

On "Cheating" With AI

I know what some people are thinking.

Isn't this just outsourcing your personality? Isn't this fake? Shouldn't writing come from within, not from a document?

This debate is well-documented. A 2024 academic study found that young people are genuinely grappling with "how writers' authenticity and creativity are threatened when the labor of meaning-making can be delegated to machines."

It's a real concern. And here's how I think about it.

The document isn't my voice. I am my voice. The document is a mirror that helps me see myself more clearly and show up more consistently.

When I sit down to write, I still have to do the work. I still have to think the thoughts, feel the feelings, make the choices. The voice reference just helps me catch when I'm drifting into generic territory or falling into lazy patterns.

And when I use AI to help draft or reformat content, the voice reference ensures the output sounds like me—not like GPT's default helpful assistant tone.

One framework I've seen is the 80/20 rule: ethical AI use means 80% of the content reflects your authentic voice and experiences, with 20% being AI assistance for structure and polish.

I'd frame it differently: this isn't outsourcing authenticity. It's systemizing authenticity.

Making it easier to be yourself more often, across more contexts, with less friction. Isn't that what tools are for?

The Deeper Point

We spend so much time consuming other people's content that we forget to study our own.

Creating this voice reference forced me to sit with my own writing again. To notice what I actually care about, how I actually think, what I actually sound like when I'm being honest.

That's not cheating. That's self-engineering.

And now I have a document that serves two purposes: it helps AI tools assist me better, and it reminds me of who I am when I get lost or overwhelmed.

If you've ever felt disconnected from your own creative output, or struggled to articulate what makes your perspective unique, try this process. You might be surprised what you find.

The most valuable thing you bring to your content is your unique perspective. AI can replicate information, but it can't replicate your experiences, your insights, or your voice.

Unless you teach it to. And in teaching it, you teach yourself.

I'd love to hear if you create your own voice reference. What did you discover about yourself? What patterns showed up that you didn't expect?