Your brand voice, at the volume you actually publish.

AI can produce content at scale. Keeping it recognisably yours is a systems problem — and it's solvable.

Voice systems for content teamsBrisbane & remote

The problem

Readers have learned to recognise the machine.

Not consciously, and not by vocabulary. They notice a rhythm — an evenness in the sentences, a paragraph that resolves too neatly — and there is a half-second reaction before any judgement forms. Attention drops. Trust drops. They rarely say why.

I call it the wince. It is a fact about audiences, not a verdict on your writers.

The tells that trigger it are structural, which is why the usual fixes don't hold. Swapping words treats the surface. And don't sound like AI is a negative constraint: it tells a writer, or a model, what to avoid and nothing about what to hit.

The useful question isn't how to sound less like a machine. It's what your brand actually sounds like — specified precisely enough that something else can reproduce it.

Most brand voice guidelines describe a personality: friendly, expert, approachable. No model can act on that, and neither can a new writer in their first week. A specification describes behaviour — sentence architecture, how the brand handles a caveat, where it allows a fragment, what it never does.

Voice Architecture

Four stages, one system you own.

Corpus analysis

I read your strongest existing work closely and measure it — sentence-length variance, structural habits, the small function-word patterns where voice actually lives.

Voice specification

A platform-independent document that states, in operational terms, what your voice does and doesn't do. Usable by a person or a model.

Production prompts

The specification implemented for the tools your team already runs — Claude, ChatGPT, Copilot. Portable by design, so a product update isn't a crisis.

Editorial QA

A checking framework your editors can apply in minutes, plus a baseline so drift is something you can see rather than something you sense too late.

The output is a system you own and your team can run — whoever's writing that week.

Why this matters at volume

One piece is an editing problem. Forty is an architecture problem.

The fortieth piece this month. Four writers, three tools, one brand. At that volume nobody is reading everything, and consistency stops being something an editor can hold in their head. It has to live somewhere outside the people.

There's a second pressure. The default voice of every model is the voice every competitor is publishing in. Prompt the same tools the same way and messaging converges — quietly, over a quarter, until three companies in a category read interchangeably. Distinctiveness used to be a by-product of humans writing. It isn't automatic any more.

Who I am

A decade in marketing, then across the desk.

I spent ten years on the marketing side, so I know what a content lead is actually being measured on and why "just make it better" is not a brief anyone can act on.

Since then I've been building and running AI writing systems in live production — not evaluating them, running them, with deadlines attached and my own name on the output. That's where the method came from. It isn't a repackaging of older work; there was nothing here to repackage.

The measurement side draws on stylometry, the discipline literary scholars use to settle questions of authorship. It has spent forty years answering one question — what makes this writer's prose identifiably theirs — and almost none of it has ever been pointed at a brand.

I work with content teams and editorial leads, from Brisbane and remotely.

Where to start

Three ways in.

Diagnostic

Voice Audit

Close reading and measurement of your recent output against your best human work, and a written assessment of where the voice slips.

From $1,500Typically 1–2 weeks

Training

Team Workshop

Half a day with your writers and editors. The named tells, prompting for voice rather than topic, and a check your team runs without me.

From $3,500Half day, on-site or remote

Build

Voice System Build

The full four stages — specification, production prompts, QA framework, measured baseline — documented so it survives staff turnover.

From $8,000Typically 4–6 weeks

What each one involves