software-architecture.aiYour voice. Even when you're not the one writing.
The moment someone else writes for you, the tone starts to drift. I build the layer in between: brand codex, touchpoint adapters, update loop. So every email, every reply, every product text sounds like your brand. Not like a language model.
Three places, one system.
Brand voice
Voice under scaling pressure
Where the tone starts to drift.
As long as you write every email yourself, the brand sounds like you. The moment someone else writes alongside you, the tone starts to spread out. Not suddenly, not loudly. More like: every other paragraph makes you wince inside, without quite being able to say why.
External writing hits the tone at seventy percent.
Brief understood, work done, delivered on time. You still step in every other piece. You explain twice, three times, then you let it run. What's missing isn't the writer. What's missing is the reference they could write against.
Mail tools suggest templates.
They're not wrong. They're interchangeable. You could swap the sender for any other brand. That's exactly what the language model does too.
Service macros sound the same everywhere.
Polite, correct, solution-oriented. For a complaint, that's often the wrong choice. Not because the macro should do something else, but because nobody noticed it was a choice in the first place.
Product copy turns into filler.
Premium build, sustainable materials, timeless design. Nobody who actually built the product would say it like that. It's there because there's no capacity for anything else.
The problem isn't delegation. The problem is that there's nothing to delegate to. No system that knows the brand. Just briefings, and the gut feeling that the texts are quietly losing substance.
What a brand voice engine is. And what it isn't.
The term is becoming a buzzword. It now covers very different things, none of which are what's meant here. So a quick clean-up.
- Not a brand voice PDF that dies in a Drive folder
- Not a generic prompt that says write in my brand's style and lets the language model decide what that means
- Not a Custom GPT that nobody feeds after three weeks
- Not a workshop deliverable that no longer fits your brand six months later
A living system of five components that work together.
- Defineswhat your brand is. And what it explicitly isn't.
- Showsthrough examples how your voice feels.
- Translatesyour brand per touchpoint. A welcome email sounds different from a WhatsApp reply. Both stay your brand.
- Updates itselfso the engine in twelve months doesn't look like a photo of you from 2021.
A brand voice engine isn't a document.
—It's infrastructure.
The five components
What's inside, in detail.
- 01
Brand codex
The foundation. What your brand means, what values it carries, what world it claims, and above all: what it explicitly isn't. Not a manifesto. A working document a new writer can read on day one and stop asking whether they're allowed to use the word cuddly.
- 02
Voice spectrum
Your brand isn't one tone, it's a corridor. It can become more formal, more informal, more serious, more playful. The spectrum defines how wide the corridor is and where the walls stand that nobody walks through.
- 03
Touchpoint adapters
A Klaviyo welcome email is a different genre from a Gorgias reply to a complaint. Both still need to sound like the same brand. The adapters translate codex and spectrum into concrete style rules per touchpoint. Tool-specific, brand-consistent.
- 04
Negative library
Here's how your brand explicitly does not sound. Concrete, with examples. AI phrases, bro-marketing tropes, greenwashing codes, interchangeable e-commerce clichés. A negative library makes a language model noticeably more useful, because it tells it what to avoid.
- 05
Update mechanism
Brands evolve. The engine has to grow with them. Concretely: a documented review rhythm, an escalation mechanism for edge cases, a clear owner. Without an update loop, the engine in six months is a language model imitating your brand from 2025.
How it runs
From first call to a system that ships.
- 01Week 0 to 1
Brand voice audit
A 60- to 90-minute audit workshop and an honest review of what you already have. Which texts are gold, which are filler, where the brand drifts. Outcome: a written stocktake and a clear recommendation on how deep the engine should go. If at the end I say you don't need an engine yet, I'll say it.
- 02Week 2 to 3
Codex workshop
Here I sit with you on what your brand is. Values, world, tone, what you're not. The one place where you really need to invest time. I can structure the rest for you, but I can't build the codex without you. Outcome: a documented brand codex and a defined voice spectrum.
- 03Week 4 to 7
Engine build
Technical implementation. Foundation layer, adapters per prioritised touchpoint, negative library, governance setup. We start with one prioritised touchpoint, not all at once.
- 04Week 8
Pilot live
One concrete touchpoint goes live. You see, you feel, you adjust. The critical phase, because it shows whether the engine really sounds like you or whether we need to sharpen the spectrum or codex.
- 05Week 9+
Roll-out and update loop
More touchpoints get connected to the engine in sequence. At the same time we set up the update mechanism: review rhythm, escalation path, clear owner. From here you can run the system yourself. I optionally stay on in a maintenance setup.
Brand voice audit
At the end you know where your brand goes quiet. Concretely, at which touchpoint, in which wording. Usable with or without follow-up engagement.
What you get
- Mirrored analysis of five of your current texts (mail, product, service)
- Voice spectrum (one page): the line from pole A to pole B, your target corridor marked, your real sentences plotted as points. What sits inside, what falls out.
- Mini-codex (ten pages): what your brand is, what it explicitly isn't
- 30-min walkthrough where I show you the cracks
Format
One week of work. Async, no workshops needed.
Fixed price
€1,490
plus VAT, one-time.
Frequent questions
A brand voice engine is a living brand module that language models use as a knowledge and style basis when they write for your brand. Not a passive PDF, not an autonomous tool. It's the structured layer between your brand and any LLM output: brand codex (what your brand is and what it explicitly isn't), voice spectrum, touchpoint adapters (a welcome mail sounds different from a service reply), negative library (what to avoid) and an update loop. The language model executes. The engine sets the corridor. What it costs depends on how deep it goes. A brand voice audit as entry is €1,490. A full engine build with codex and one first touchpoint sits in the mid four-figure to low five-figure range. A roll-out across multiple touchpoints and languages sits above. In the first call I'll tell you honestly what makes sense at your brand size and whether an audit as entry is enough.
If you want to know whether the engine fits your brand.
The entry is a 60-minute brand voice diagnosis for €99, no sales pitch. You speak, I listen, give concrete feedback on three of your current texts and tell you honestly whether I'm the right sparring partner. If not, I'll say it. If you want to go deeper afterwards, the brand voice audit is €1,490.