Most Australian SMEs that try to use AI for content end up with the same problem: it sounds nothing like them. The blog post reads like every other AI-generated blog post. The email feels generic. The product description could have come from anywhere. The reason is almost always the same — there is no brand voice document, or the one they have is written for humans, not for the AI that will be doing most of the writing from now on.
A brand voice document for AI is a different beast to the brand guidelines you might have buried in a Dropbox folder somewhere. It is not about logos, colours, or font hierarchies. It is a working set of instructions that you can paste into ChatGPT, Claude, or whichever model you use, and that produces copy you would actually sign your name to. Built well, it cuts editing time by 60-80% and means anyone in your business — staff, contractors, agencies — can generate on-brand content without going through you.
Why your existing brand guidelines won't work
Most brand guidelines were written for designers and copywriters who would then use their judgement to apply them. A line like friendly but professional makes sense to a human marketer. To an AI, it is meaningless padding. The same goes for we speak with confidence or modern and forward-thinking. These phrases tell the model nothing it can act on.
The model needs concrete instructions: specific words you use, specific words you don't use, sentence rhythm, paragraph length, what your headlines look like, what your calls-to-action sound like, and ideally three to five examples of writing you would happily publish. Without those, the AI will fall back on its training data — which means your content will sound like the average of every business blog post on the internet. Which is exactly what you do not want.
The 6 things your AI brand voice document must include
A working AI brand voice document has six sections. The first is who you are writing for: not just demographics, but the actual person — their job, their pressure points, what they are sceptical of, what they are tired of hearing. The second is your point of view: what you believe that your competitors do not, the position you defend, the things you push back on. Without a point of view, AI-generated content turns to porridge.
The third is vocabulary: a list of 15-20 words and phrases you use often, plus 15-20 words and phrases you ban. ProjxAI bans words like unlock, leverage, cutting-edge, seamless, and revolutionise because they are AI-tells. The fourth is sentence and paragraph rhythm: do you write short and punchy, or long and flowing? Do you use one-line paragraphs for emphasis? Do you ever use rhetorical questions? Be specific.
The fifth is structure: how your blog posts open, how they close, how often you use sub-headings, whether you use lists or prose. The sixth — and the one most businesses skip — is 3 to 5 full examples of writing you love. Paste them in verbatim. The model will pattern-match off them harder than any abstract instruction you give it.
How to actually write it — the 90-minute method
Block out 90 minutes and open a fresh document. Spend the first 20 minutes pulling together every piece of writing you have ever published that you genuinely like — emails to clients, a LinkedIn post that landed, a paragraph from your About page, a quote you gave a journalist. Read them back and underline what is consistent across all of them. That is your voice in the wild.
Spend the next 30 minutes on the banned words list. Open three or four AI-generated drafts you have rejected and circle every phrase that made you wince. Revolutionise. Empower. Game-changing. In today's digital landscape. Robust. These go in the never-use column. Then list 15-20 words you actually do use. For ProjxAI it includes words like operator, workflow, practical, time-poor, honest, outcome. These are your always-available words.
Spend the final 40 minutes writing the document itself, structured around the six sections above, with your favourite three pieces of writing pasted in full at the end as examples. Save it as one Markdown or text file — not a PDF, not a Word doc — because you are going to paste it into AI tools constantly. Friction kills adoption.
How to actually use it day-to-day
Paste the entire document into the top of every AI conversation where you want on-brand output. In ChatGPT or Claude, create a Project or Custom GPT and load the document as system instructions so you do not have to paste it each time. In automation tools like Make or n8n, include the document as part of every prompt that generates customer-facing copy. This single move is what separates teams who get useful output from AI and teams who do not.
Test it properly. Generate the same blog post or email three times — once with no brand voice document, once with your old human-style brand guidelines, once with the new AI-ready version. The difference is usually obvious within the first two paragraphs. Iterate on the document monthly. Every time the AI produces something you have to heavily edit, ask yourself which rule in the document was missing. Add it. Over six months your document becomes a precision instrument.
The one thing you can do today
Open a new document right now and write your banned words list. Just that. Twenty words and phrases you never want your business to say. It will take you 15 minutes, and the next time you ask an AI to write anything, paste the list in with the instruction Avoid every word and phrase on this list. The output quality will jump immediately, and you will have started the brand voice document properly.
If you want a structured framework for thinking about how AI fits into your business — including content, brand, and the workflows underneath — start with the free ProjxAI Opportunity Audit. It maps out where AI will move the needle for your business specifically, and where it won't. Twelve questions, takes about six minutes, no fluff.
Or if you'd rather just talk through it, book a 15-minute conversation. We'll tell you in plain English where AI fits and where it doesn't for what you do.