Developing Brand Voice: Foundation First, AI Second

The most common brand voice problem is not bad writing, it is inconsistency. The website uses casual "you", the newsletter uses formal "you". On LinkedIn, five different employees post in five different tones, none of which really fits the brand. This happens not because the writing is weak, but because no shared foundation exists.

Why inconsistency is the real brand voice problem

Inconsistency sounds like a small thing you can get around to later. In practice, it hollows out what brand communication is meant to do.

A concrete picture: the same website uses casual "you" in one section and formal "you" in another. The LinkedIn tone reads relaxed and easy, the website feels neutral and distant. Add five employees writing for the company on LinkedIn, all with different styles, none of which really matches the brand. That is the rule, not the exception.

This shows up most in email marketing. Anyone who reads newsletters regularly can tell quickly whether there is one consistent voice behind them or whether each issue feels a bit different. Recognizability does not come from the logo in the header. It comes from being able to tell after two sentences who wrote this. That can only be built over time when everyone who writes for the brand knows how it sounds.

There is one angle that rarely gets factored in: AI systems like ChatGPT, Google AI Overviews, and Perplexity form their picture of a brand from what they find on the web. If the same brand is described one way on the website and another way on LinkedIn, one way in press releases and another in blog posts, these systems struggle to form a coherent picture. Consistent communication, the same core statements, the same positioning, the same key phrases, makes it easier for AI to represent a brand correctly. This isn't theoretical. It shapes how a brand shows up in AI-assisted search results. Google describes consistency and authority as central signals for helpful content.

How that plays out in practice and why AI in marketing delivers far less without a consistent brand voice: concrete examples in the linked post.

Brand voice is not a character essay, it is decisions at sentence level

Brand voice often gets confused with something that sounds much more elaborate than it needs to be.

The textbook answer goes something like: the character of a brand, expressed in language. True, but it does not help if you do not have one yet. More concretely: brand voice is the sum of tone (formal or close?), stance on topics (direct or evasive?), and the phrasing patterns that repeat. A company that explains things factually in trade articles and then posts questions on LinkedIn nobody asked does not have a brand voice. It has two writing modes with no connection.

What I see regularly: 60-page brand voice guidelines. Heavy to produce, never used. Most of these documents do not get opened again after the briefing meeting. A document that big never gets used in daily work. What does get used is one page, maybe two. The rest goes unread.

Tone of voice is not the same as brand voice, even if the terms get mixed up. Brand voice is the base voice that stays constant. Tone of voice is how that voice shifts depending on the situation. A brand that explains things factually on LinkedIn and sounds more personal in the newsletter shifts tone of voice while keeping the same base voice across both channels.

See the current state before developing anything

Before you start building, take an honest look at where you are. AI is unexpectedly useful here, not as a tool that prescribes how you should sound, but as a mirror.

The method is simple: gather existing texts (website copy, two or three newsletters, a few LinkedIn posts), paste them into Claude or ChatGPT, and prompt something like: "How does this brand sound? Describe the tone, stance, and phrasing patterns that repeat." The result is usually sobering, but informative. No consistent picture. Sometimes neutral, sometimes salesy, sometimes casual. That is where the work begins.

But AI only describes, it does not judge. "The texts sound formal and distant" is an observation, not a recommendation. Whether that should stay or not is a decision a human makes. AI gives you the raw material for that decision.

If you want to start with the oldest available material: pull the three oldest texts from the website or the LinkedIn profile, paste them into Claude, and ask: "How does this brand sound? What repeats, what contradicts?" The output is a more honest starting point than any self-written description of your own voice.

How to build a voice guide people actually use

The goal is a one-page document that gets used in daily work. The path there has a clear order: start manually, bring in AI at the end.

Begin with a list of adjectives: which fit the brand, which do not? "Direct yes, arrogant no. Factual yes, dry no." Not a guide yet, but a starting point made of real decisions, not wishful thinking.

Then a "We say / We don't say" list, concrete instead of abstract. Not "we avoid jargon", but: "We write 'this takes time' instead of 'this requires temporary resource allocation'." This list is more useful in daily work than any adjective collection, because it gives you decisions at sentence level.

Once that is in rough form, AI comes in. Prompt: "Here is our voice draft. What is missing? Where are the contradictions? What is unclear?" The method works because you are too close to your own document to spot places that contradict each other. AI does not invent a new voice. It reads the draft against itself and flags what does not fit. You decide what to do with the feedback.

How this fits a fully automated content process is shown in the Content Machine case study.

Consistency in daily work: from document to real text

A guide that sits in a folder changes nothing. Brand voice does not work in the document. It works in the text written after. The step from document to daily work is where most people fail.

The simplest entry point: paste a finished text together with your voice principles into a prompt and ask: "What in this text does not fit these principles?" That surfaces spots to revise, even if it does not deliver finished corrections. Especially useful when multiple people write for the brand or when external agencies deliver text nobody has checked against the guide.

The scalable version: the voice guide sits as a .md file, and Claude Skills load it automatically as context when they generate or review text. Sub-agents working on content tasks then have the guide in context without anyone pasting it in manually each time. Not a big technical lift, but it makes the difference between a guide that gets used and one that sits in a folder. The post on context engineering goes deeper on how to steer AI context deliberately.

Not a perfect system. AI reads text, but it does not know the context, the tone of a specific channel, the expectations of a specific audience. It can flag rough deviations, it does not catch fine nuance reliably. Better than nothing, as long as you know that.

If you want to go deeper on authentic AI content, that post explores what "sounds human" actually means. How brand voice fits into a broader content strategy is covered in the post on strategic content marketing.