At a glance
A recognisable brand voice comes from choices: what you notice, how directly you make a point, which claims need evidence and what you refuse to dress up in impressive-sounding language. Teach those choices and the draft has a better chance of sounding like your publication.
Getting AI to write in your brand’s voice means teaching it the judgement behind your writing. A recognisable voice takes shape in the questions you ask, the evidence you trust and the way you explain an idea to your audience. A draft can be fluent and grammatically sound while missing all three.
That is the problem with instructions such as “write in a confident, friendly and professional tone”. Those words could describe thousands of businesses. They give an AI system little guidance about what makes yours distinct, leaving editors to put the character back into each draft.
A useful brand voice guide makes those editorial choices visible through examples. Start with passages that sound like your business, explain why they work, and show how an editor would improve a weaker version.
Collect examples that show a decision
Choose a small set of strong passages from your own work. Include an opening, an explanation, a criticism, a recommendation and a conclusion. Annotate why each works.
For example: “This opening starts with the reader’s problem rather than a general statement about the industry.” Or: “This paragraph distinguishes a measured result from the author’s judgement.” These instructions are more actionable than “be authentic”.
Anthropic’s prompting guidance recommends clear instructions and examples to shape output. The editorial application is to provide examples that demonstrate the behaviour you want, with enough context to explain the choice.
Define the publication’s point of view
A voice guide should include beliefs the publication can defend. For MarTech Logic, useful editorial questions include whether a tool improves a real decision, whether a metric earns its place and whether the customer experience matches the commercial promise.
These are not facts to repeat in every article. They are lenses for examining a subject. The writer should still reach a conclusion based on the specific evidence.
Keep opinion separate from reported research. “We think this is the better starting point” is a judgement. “The study found this result” is a factual claim that needs the source to support it.
Show the edit, not just the finished paragraph
Before: “In today’s rapidly evolving landscape, organisations must leverage innovative solutions to unlock efficiency.”
After: “If your team spends Friday fixing campaign reports, another dashboard may not be the answer.”
The improvement is not simply shorter wording. The second version gives the reader a situation, a point of tension and a reason to continue. It creates room for an argument.
Build a handful of these pairs using your actual editorial preferences. Include examples of overclaiming, vague transitions and unnecessary jargon. Avoid turning the guide into a long blacklist that leaves the writer with nothing positive to aim for.
Give the model a proper assignment
Supply the audience, the question, the intended argument, the source material and the desired next step. A voice guide cannot compensate for a missing brief.
Ask for a structure before a full draft when the subject is complex. Check whether the sections develop a useful argument or merely list familiar headings. Then review the prose against the sources and the reader’s needs.
For recurring work, measure the cost of usable output, including editing and fact-checking. A fast first draft that needs a complete rewrite is not necessarily a productivity gain.
Test with passages, not adjectives
Give editors a small sample without telling them which version used the revised guide. Ask concrete questions: Is the point clear? Does the opening feel relevant? Which sentence sounds unlike the publication? Which claim needs more support?
Collect recurring edits and improve the guide. If editors repeatedly remove grand introductions, add a stronger opening example. If the writing becomes blunt or repetitive, show how the publication varies pace and handles nuance.
Do not make every article sound like the same person delivering the same lesson. A comparison, an interview and an analytical essay need different structures even when they share editorial standards.
Keep the tools and responsibilities visible in your martech stack review. A writing workflow needs an accountable editor as much as it needs a model subscription.
Keep human ownership of the judgement
The final editor should decide whether the argument is fair, the evidence is sufficient and the language suits the audience. The brand voice guide is an aid to that work, not a substitute for it.
The goal is not to hide how a draft was produced. It is to publish writing with a clear mind behind it: something specific to say, a reason to believe it and enough care to make it worth reading.
