After 25 years of writing, I know that voice changes with the work.
An email to a colleague does not sound like a formal proposal. A message to an engineer can rely on shared technical context. End-user documentation should help someone complete a task without burying them in administrative detail, while administrator documentation must address permissions, dependencies, risks, and system behavior that an end user may never need to see.
Those shifts are part of the judgment I bring to the work.
When I began building the Structured Ink publishing system, I had to make that judgment explicit. The system needed to support two voices that come from the same person but serve different purposes: Katie Kearns and Structured Ink.
I decided to use AI for some of the work, both to save time and to gain more hands-on experience with different models and capabilities.
I Use AI More Like an Intern
The simplest way to describe AI’s role in this system is that I use it more like an intern than an author.
Before drafting, it can gather information, organize source material, collect my answers and reactions, find connections, and identify questions that still need attention. During drafting, it can prepare a first version, adapt an approved idea for another channel, and check the work against the recorded voice guidance.
That assistance is valuable. It gives me material to react to and lets me spend more time on decisions that require judgment.
But the ideas are mine, the voice is mine, the edits are mine, and the final approval is mine.
Every article, post, and public-facing piece is checked and edited by me before publication. AI-generated text is not considered finished simply because it is fluent or follows the requested structure.
The same boundary applies when the system suggests a change to a voice guide. AI can say, “You have made this correction several times; should it become part of the guidance?” It cannot make that decision for me.
I remain responsible for deciding what the writing means, whether it sounds like me or Structured Ink, whether the claims are supportable, and whether the work is ready to carry my name or the company’s.
Two Voices Can Share a Vocabulary and Still Be Different
Structured Ink grows directly out of my work. The company reflects my experience, standards, interests, and way of thinking. Naturally, the two voices share some characteristics. Both should be clear, specific, evidence-minded, and restrained. Neither should inflate a claim for effect or treat confidence as a substitute for support.
If I stopped there, however, the system produced two versions of the same generically polished professional voice. One might use “I” and the other “we,” but changing a pronoun is not the same as changing the speaker.
The real difference is what each voice is meant to do.
Katie can show more of the journey. I can write about what I tried, what surprised me, what went wrong, and why I changed my mind. The path to the conclusion can be part of the point.
Structured Ink has a more edited, systems-oriented role. It takes an observation and asks what another person or organization can use from it. The company voice should turn experience into a useful explanation, method, or recommendation.
I did not want the company to sound like a large institution speaking through a committee. I also did not want every company article to feel like a personal post with a logo attached.
Training the voices meant teaching that distinction.
I Started by Answering Questions
The process began with voice-calibration interviews.
Instead of starting with a list of adjectives or a mission statement, I answered questions about how I actually communicate. AI helped gather and organize my responses, identify areas that needed clarification, and turn the discussion into material I could review. That was useful because the distinctions were not always obvious at first. We had to draw them out before the system could apply them, and AI helped me brainstorm.
- What do I sound like when I am explaining something I understand well?
- How much of my reasoning should remain visible?
- When does personal detail strengthen the point?
- How do I express uncertainty?
- What makes a sentence feel too polished to be mine?
- What kinds of claims should Structured Ink make?
- What should the company avoid claiming?
The questions surfaced decisions that were already present in my writing. For example, my personal voice can say, “I tried this, it failed, and here is what the failure clarified.” Structured Ink might use the same experience differently: identifying the underlying problem, explaining why it matters, and giving the reader a way to recognize or address it.
The source material may be shared, but the editorial treatment should not be. The interviews gave me the first version of each voice guide. They were a starting point, not a finished solution.
The Voices Learn Through Correction
A voice guide becomes useful when it survives contact with actual drafts.
I read the generated copy and respond to what feels wrong:
- This sounds too polished.
- This sounds like a much larger company.
- Katie would show more of the reasoning here and talk about her background in science.
- Structured Ink should get to the principle sooner.
- This sentence makes the company sound certain about something that is really my personal interpretation.
- This version repeats the same post with different pronouns.
Each correction adds information that a label such as “professional” or “conversational” cannot capture.
As I edit, AI can recognize recurring patterns in my feedback and suggest changes to the recorded voice guidance. It might notice that I repeatedly remove a certain kind of phrasing, ask for more visible reasoning, or distinguish a personal observation from a company claim.
Those suggestions do not automatically become part of the voice. I review them and decide whether they reflect a genuine principle or only the needs of one draft. I approve, revise, or reject each proposed change.
That human-in-the-loop step matters because an edit is not always a universal rule. Sometimes I am correcting the voice. Sometimes I am making a choice for one audience, subject, or moment. The system can identify a possible pattern, but I decide whether it belongs in the lasting guidance.
Over time, the guides become less about surface style and more about editorial choices:
- who is speaking
- who they are speaking to
- what the reader already knows
- how much reasoning should remain visible
- where the authority for the claim comes from
- what the reader should understand or be able to do afterward
- who should be accountable for the published statement
That is what I mean by training a voice.
It is not choosing a few descriptive words and expecting every future draft to work. It is giving the system examples, gathering input, responding to the results, naming the difference, and deliberately deciding which corrections should carry forward.
Boundary Cases Provide the Most Useful Training
The easy examples do not teach the system very much. A personal account of a career experience clearly belongs to Katie. A service description clearly belongs to Structured Ink.
The useful training happens in the middle.
This website series is a good example. I am writing about choices I made while building the Structured Ink publishing system. The experience is personal, but the subjects—content architecture, governance, automation, publishing controls, and documentation practice—also belong to the company’s work.
Either voice could address them.
The question I use to separate the treatments is:
Where does the authority for this piece come from?
If the value comes primarily from something I experienced, tried, noticed, or learned, the piece probably belongs to Katie.
If the value comes from turning that experience into a method, explanation, or recommendation that can stand on its own, it probably belongs to Structured Ink.
Sometimes the right answer is to create both. Structured Ink can explain the system. I can explain the choices, doubts, mistakes, or experiences that shaped it.
Those versions should be related, but they should not be copies with the pronouns changed.
My Documentation Work Gave Me the Model
Training these voices felt familiar because it resembles work I have done throughout my career. Documentation for different audiences may describe the same product, but readers have different goals, permissions, risks, and responsibilities. Writing every version in exactly the same way would not make them consistent. It would make one or both less useful.
The writer has to understand what each reader needs, what they can act on, and what consequences they are responsible for.
I applied the same principle to Katie Kearns and Structured Ink.
The two voices can share subject matter and values without becoming interchangeable. Consistency does not require sameness. It requires a clear relationship among the speaker, audience, purpose, and claim.
The System Can Apply the Voice. It Cannot Own It.
The voice guides give the AI-assisted workflow something more useful than “sound like me.”
The system can compare a draft against the guidance. It can recognize when a personal post has become too institutional or when a company article spends too much time narrating my individual process. It can prepare distinct treatments for the website, the Structured Ink LinkedIn page, and my personal LinkedIn account.
But it does not own the experience behind a personal post, and it is not accountable for the company’s position. Producing a confident sentence does not give it the authority to choose its own speaker or publish its own conclusion.
Training the voices does not mean handing over editorial judgment. It means giving the system enough context to assist me more consistently, from gathering input through drafting and revision.
The approval boundary stays with me.
If you had to train someone—or a system—to write in your voice, which choices would you need to make explicit, and which decisions would you keep for yourself?
This is the second installment in The Website Is the Case Study, a series about the decisions, mistakes, safeguards, and small pieces of architecture behind Structured Ink’s publishing operation.
Next: Two Sources of Truth, One Publishing Flow.