Most people have a pretty clear sense of what good writing looks like in their organisation.
The standard exists and is maintained by experienced writers who have been applying it for years.
The problem is that it is rarely written down in a form that AI can use.
When someone prompts a model with ‘draft a staff note about this new process’ or ‘write a short client update’, the person giving the instruction often knows exactly what the result should sound and read like.
But the model doesn’t — and instead it fills the gaps from its own generic training.
The fix is giving the model a clear written standard to work from.
Ironically, we’ve never been big fans of voice guides. They tend to describe brands as practical yet inspiring, relevant but accessible. They get given to a new writer but otherwise sit unused — a bit too obvious to be useful to experienced staff and a bit too vague to be useful for anyone else.
But using a voice guide as specific instructions for a model — setting out how you want to say things, how you want to be perceived, and who your audience is — turns out to be a terrific idea.
Getting it usable is as simple as dropping the guide into the AI and asking for it to be turned into a working instruction set.
Often people rely on negative instructions to tune AI — ‘don’t use jargon’, ‘don’t be promotional’.
But positive instructions are stronger.
Tell the model: use short paragraphs, sound natural when read aloud, use contractions where appropriate, emphasise benefits over features, lead with the point, address the reader as ‘you’ … and so on — whatever your voice guide says your organisation’s voice is, that’s what gets fed into the model’s context.
It gives the model a working brief and, perhaps more importantly, it gives whoever reviews the draft something concrete to test against — not just ‘does this feel right?’ but ‘did it do what we asked?’