Week 1
Check the voice exists
In writing, with examples. If it does not, that is the first job and configuring before it is done produces a confident wrong answer.
AI-assisted workflows · Brand-voice models · Sydney
Brand-voice models
A tool with no configuration writes like everyone else's tool with no configuration, which is the single reason so much published content now sounds identical. The fix is unglamorous and mostly consists of writing things down.
Grab the snowball. Give it a spin, and watch it grow as you scroll. Just like your voice.
4inputs a usable model needs
1person still writing and accountable
Starts closera better draft, not a finished one
Yoursthe configuration is handed over
The short answer
What is a brand-voice model, and what does it not do?
A brand-voice model is a set of instructions, examples and reference material given to a language tool so its drafts start closer to how your business actually writes. It is configuration rather than magic: rules, samples, a banned list and your own source material.
What it does not do is write your content. It shortens the distance between a blank page and a usable draft, which is a real saving. A person still writes, decides and is accountable, and the finished work is edited by somebody whose name is attached to it.
We are direct about that because the rest of this site argues that human-made work is the advantage as machine-made content floods every feed. A voice model is a tool for getting to good work faster; it is not a substitute for having something to say. It sits inside our wider AI-assisted workflows service, and because our Snowball SEO platform automates the search heavy lifting other agencies bill by the hour, more of your budget goes into the editing that matters.
The difference
What does a voice model need to be useful?
Four inputs. Most attempts supply one, the adjectives, and then conclude that the tools are not very good.
What goes into a working model
- Written voice rules, with before and after pairs rather than adjectives
- Twenty or so pieces of your own writing that genuinely sound right
- Your factual material: data, case detail, product truth
- A banned list: the claims, words and structures you never use
Why most attempts fail
- Three adjectives and a hope
- Examples chosen because they are recent rather than because they are right
- No factual grounding, so the output is confident and hollow
- Nobody editing, which is where the actual quality is added
If the voice itself is not defined yet, that is editorial voice and tone work, and it comes first.
The cost
What drives the price of setting one up?
How much of your voice and evidence already exists in a usable form. If the voice guide and the source material are there, this is configuration. If not, most of the work is upstream.
Configuration
Setup
The model configured from your existing voice guide and material, tested against real briefs and handed over.
Voice and configuration
Both
Defining the voice first where it does not exist, then configuring, which is the common case and the honest sequence.
Configured and maintained
Ongoing
Kept current as the business changes, because a model configured once drifts out of date exactly as guidelines do.
We quote after a free consult and a look at what exists. If the voice is not defined, we will say so: see editorial voice and tone.
What moves the number, in order
- Whether a written voice guide exists already
- How much of your own material is usable as reference
- How many content types the model has to serve
- Whether factual grounding has to be assembled from scattered sources
- How much testing against real briefs is included
- Whether maintenance is part of the arrangement
The saving is at the start, not the end
A voice model does not reduce editing time much. What it removes is the blank page and the first bad draft, which is where writers lose hours and momentum.
So the honest expectation is a faster route to a usable draft rather than finished content. Businesses that expect the second are always disappointed and usually publish something they should not have.
We would rather set that expectation now than sell a saving that does not exist.
The scope
What you receive
A configuration you own, and the evidence base behind it.
The voice configuration
The instructions, rules and examples, written down and handed over as text you own rather than locked into one tool.
A curated reference set
Twenty or so pieces of your own writing chosen because they sound right, which is a surprisingly hard selection to make well.
The factual base
Your data, case detail and product truth, assembled so drafts are grounded rather than plausible.
The banned list
Claims, words and structures never to use, which is what stops a tool wandering into things you cannot say.
Tested against real briefs
Proof it produces something usable on your actual work, before handover rather than after.
A maintenance note
What to update and when, since a model configured once ages exactly as brand guidelines do.
The return
Where does the model help, and where does it not?
It helps most where the work is structured and the facts are known. It helps least, and hurts, where the value is a genuine opinion or original experience.
Accelerate the routine so people can do the rest
The argument for this work is not that machines write acceptably. It is that a great deal of business writing is structured, repetitive and low-judgement, and clearing it faster leaves time for the writing that actually differentiates you.
A team that spends its week on product descriptions has no time left for the founder's argument nobody else could make. Moving the first of those does not devalue writing; it is the only way the second gets written at all.
That is consistent with our position that human work is the moat, and it is worth being precise about rather than vague: use the tools where judgement is not the product.
The process
How a voice model gets built
Two to four weeks, most of it selecting and writing down rather than configuring.
Week 1
Curate the reference set
Twenty pieces that genuinely sound right, chosen deliberately. Recency is the wrong selection criterion and the usual one.
Week 2
Assemble the evidence
Data, case detail and product truth, so drafts are grounded rather than merely fluent.
Week 2
Configure and test
Set up against real briefs from your actual backlog, not invented examples.
Week 3
Compare blind
Outputs read against your existing writing without labels, because a configuration everyone can pick out is not finished.
Week 4
Hand over
The configuration as text you own, with the maintenance note and clear limits on where it should not be used.
The brief
What to ask before you build a voice model
This is the easiest service in the estate to oversell, so the questions are mostly about what it will not do.
What will this not be used for?
A supplier who cannot name the work it should not touch has not thought about it. Original opinion and anything under a named byline should be on that list.
Do we own the configuration?
It should be handed over as text, not locked inside one supplier's tool or account. Ask before the work starts.
What grounds the facts?
Rules alone produce fluent, confident emptiness. Ask what factual material the model is given, because that is what separates usable from plausible.
Who edits, and whose name is on it?
A person, named, accountable. If the answer is that output goes out unedited, you are buying the thing the rest of the market is already drowning in.
We are happy to answer all four before you commit, including telling you when the honest first step is defining the voice rather than configuring anything.
Start here
Ready to stop sounding like everyone else?
Send us five things you have published that genuinely sound like you. The free consult is a working session, not a sales call, and you leave with three things whether or not you book us.
- An honest read on whether your voice is defined enough to configure anything
- A sample draft produced against your material, so you can judge the distance
- A fixed price, and a written list of what we would not use it for
Good questions
Brand-voice model FAQs
What is a brand-voice model?
It is a set of instructions, examples and reference material given to a language tool so its drafts start closer to how your business writes. It is configuration rather than magic: rules, samples, a banned list and your own source material.
Does this mean our content is written by a machine?
No. It shortens the distance between a blank page and a usable draft. A person still writes, decides and is accountable, and the published work is edited by somebody whose name is attached to it.
Why do most attempts at this fail?
Because they supply three adjectives and nothing else. A usable model needs written voice rules with examples, twenty or so pieces of your own writing that genuinely sound right, factual grounding, and a list of what you never say.
How much does it cost to set up?
We quote after seeing what exists. What moves it is whether a written voice guide already exists, how much of your material is usable as reference, how many content types it serves, and whether factual grounding has to be assembled from scattered sources.
Where should we not use it?
Original opinion, anything under a named author's byline, claims that require accountability, and anything drawing on experience only your people have. That work is the most valuable you produce and is the reason the routine work is worth accelerating.
Do we own the configuration?
You should, and we hand it over as text rather than leaving it inside one supplier's account. Ask this before any work starts, because it is a difficult conversation to have at the end.
How much time does it actually save?
Less than vendors claim and more than sceptics expect. It removes the blank page and the first bad draft, which is where writers lose hours. It does not reduce editing much, and anyone promising finished content is describing something you should not publish.
Does this conflict with your view that human content wins?
No, and it is worth being precise. Our position is that human judgement and real experience are the advantage. Using tools to clear structured, low-judgement work faster is what creates the time to do the work that actually differentiates you.