Practical Guides

Workflow · 9 min

Build a brand voice system that AI tools can actually follow

Most brand-voice documents read well and produce nothing. A working system needs specifics, examples, and a way to load all of it into an AI tool that uses it.

  1. 1

    Start from the audience, not the brand

    Voice is a function of who you are speaking to. Name the primary reader — their role, their decisions, the vocabulary they use. A document that begins with the audience produces a voice that lands; one that begins with brand attributes ('innovative, bold, trusted') produces a voice that does not.

  2. 2

    Choose three axes and place yourself on each

    Pick three tone dimensions that matter for your category — for example, formal/casual, restrained/expressive, expert/accessible — and place your brand on each as a point, not a range. Specificity here is what makes the system usable.

  3. 3

    Write five before-and-after pairs

    For each pair, take a piece of generic writing and rewrite it in your voice. Cover an email, a marketing line, a support reply, a long-form opening, and an error message. These pairs do more work than any tone description because they show, rather than tell, what the voice means.

  4. 4

    Document the 'never' list

    List ten phrases or moves you do not make: 'we believe,' 'leverage,' 'cutting-edge,' starting with 'In today's,' ending with 'in conclusion,' over-explaining acronyms, exclamation marks in business contexts, whatever your context requires. Short and absolute beats long and qualified.

  5. 5

    Load it into the AI tool you actually use

    Move the voice document — audience, axes, before/after pairs, never list — into the system prompt or brand-voice profile of your AI tool. Most platforms support custom instructions or persistent style profiles. The document only works when the tool actually sees it on every request.

  6. 6

    Test on real outputs and refine the document

    Run the system on twenty real briefs from across your team. Where the output drifts, the document is incomplete — add a constraint, add an example, tighten the never list. Treat the document as a living asset; the first version is rarely the final one.

Common pitfalls

  • Vague attributes. 'Approachable but professional' produces nothing useful. Concrete pairs do.
  • Documents that live in Notion and never reach the model. The system prompt is where voice lives operationally.
  • Trying to capture every edge case. Cover the common ground; edit the long tail.
  • Forgetting to update the document when the brand evolves. Stale voice systems silently produce off-brand output for months.

Frequently asked

How long should a brand voice document be?

Working systems are typically 400-1,200 words. Longer documents are usually less effective because they are not read consistently.

Should each writer have their own voice profile?

Often yes — a brand voice plus a personal byline voice. Most tools support layering both.

How often should we update it?

Quarterly review; immediate update whenever positioning, audience, or category strategy changes meaningfully.

Run the workflow on a real draft

Open the rewrite engine and apply the steps to writing you actually have to ship today.

Start rewriting

Related guides