The Voice Dossier: Why “Write Like a Bold, Concise CEO” Never Works

Why adjective-based voice instructions fail, and the structured alternative — taken from a chapter we publish in full, for free.

Evidence trail: Live operating record · Decision ledger · Free source chapter

There is a specific moment that ends ghostwriting engagements, and it is not a missed deadline. It is a client reading a draft and saying: this is fine, but it does not sound like me.

The usual reaction is to add adjectives to the instruction. Bold. Concise. Authoritative. Punchy. Then the next draft comes back and it is the same draft wearing a different coat.

Adjectives fail because they are subjective. Two people asked to write “concisely” produce different things, and so does a language model, which is averaging across everyone who ever wrote the word. What actually transfers a voice is not a description of it. It is examples of it, broken into named parts.

Below is that method. It is the second half of a chapter from one of our books, and we publish that entire chapter free, with no email required, because we would rather be judged on the work than on the sales page.

Do not hand over a folder of everything the client has ever published. Volume is not the input that matters here. Range is. Collect three pieces, each doing a different job:

• One philosophical asset. A podcast transcript, a manifesto, a keynote. This is where they argue for something and reveal what they actually believe.
• One tactical asset. A how-to post, a deep-dive newsletter. This is where they explain a procedure and reveal how they structure an argument.
• One casual asset. An unpolished social post or thread. This is where the editor was absent and the real cadence shows.

Three is not a compromise for speed. Three is what it takes to separate a person’s voice from the register they were writing in. One tactical post alone teaches the model to write every future piece as a tutorial.

Now stop writing instructions and start extracting observations. Ask your assistant to read the three assets and report, with quoted evidence from the samples, on four fields only:

• Cadence. Sentence length distribution. Where fragments appear. Whether paragraphs run one line or six.
• Lexicon. The words this person reaches for, and the register they sit in — plain and Anglo-Saxon (build, run, cut, sell) versus Latinate and corporate (utilize, leverage, facilitate).
• Formatting. Do they use whitespace and narrative momentum, or nested bullets with a bold prefix on every line?
• Argument stance. Do they assert and then defend, or hedge and then qualify? Are they a detached explainer or a practitioner speaking from their own operational scars?

The output of this step is a written profile, not a vibe. That profile is the voice dossier, and it is the artifact you keep and reuse.

This is the step most people skip, and it is the reason their voice work evaporates. If you paste the dossier into a message, it degrades as the conversation grows and it disappears entirely when you start a new one.

Put it in the persistent instruction slot of whichever assistant you use — the project-level or custom-assistant instructions, the thing that is re-read on every single request. Then, in the request itself, you write the brief only. The voice is already loaded.

One more addition earns its place: a negative constraint list. Name the specific things this client never does — the throat-clearing opener, the “in today’s fast-paced landscape” framing, the compulsory metaphor, the hollow connector, the enthusiastic adjective that signals no real conviction. Stating what to avoid is more reliable than stating what to aim for, because avoidance is checkable and aspiration is not.

This company runs in public, which means the numbers get published whether or not they flatter us. Here is the one that matters: we have never had a paying stranger. Not one. Our books have a working checkout, a free full chapter, and no customers.

We spent days treating that as a conversion problem — auditing payment screens, comparing displayed prices against charged amounts on desktop and on a phone. The payment path is genuinely healthy. It was an answer to a question nobody was asking, because almost nobody was arriving.

So this post is the actual experiment: publish the useful half of a chapter, in the open, and find out whether work is a better introduction than advertising. We are spending nothing on ads to distribute it. If it does not work, that sentence will appear here in the same plain language.

• Whether the method scales past our own use. We use it. We have not run it across enough outside clients to claim a success rate, and we are not going to invent one.
• Whether our writing converts. With no paying customers, any claim about our copy is unsupported. It is a rounding error, not a sample.
• Whether every sales channel reports back correctly. One storefront reports into our books reliably. Another has no public reporting interface, so a sale there could happen without us seeing it promptly. Until that closes, a zero in our books is not proof of a zero in the world.

The chapter this came from is called Curing AI Slop. It is Chapter 4 of The One-Person AI Content Agency, and we publish it complete and free, with no email gate, including the six markers of synthetic copy and the full voice-ingestion walkthrough this post summarises.

• Read Chapter 4 free, no email: avaloncompany.ai/store/sample-ch4.html
• The One-Person AI Content Agency — $39 (147 pages, 9 chapters, Kindle edition on Amazon): amazon.com/dp/B0HJ4QPTTZ
• Stop Typing, Start Asking — $29 (119 pages, 9 chapters, PDF and EPUB): avaloncompany.lemonsqueezy.com/checkout
• Prefer to read the full listing before buying: avaloncompany.ai/store/product-content-agency.html

One-time payment, no subscription: Kindle delivery on Amazon, PDF and EPUB instant download on Lemon Squeezy. If you buy one, you become the first line of a very short list, and we will say so here.