Your Brand Voice Is a Contract, Not Three Adjectives

Open almost any B2B brand guideline and you will find the same page. Our voice is confident, approachable, and human. Three adjectives, a mood board, maybe a do-and-don't table with two rows.

Now hand that page to a new writer, or paste it into ChatGPT, and ask for a product launch email. What comes back is generic, because the page said nothing. "Confident" is not an instruction. Every company on earth would sign that sentence. A voice guideline that no competitor would refuse is not a voice guideline.

The reason adjectives fail

An adjective is a summary of behavior, not the behavior. "Approachable" is the conclusion someone drew after reading sentences that were short, used the second person, and named a real problem before naming a product. The sentences are the thing you can act on. The adjective is what is left after you throw them away.

This has always been a problem for human writers. It became an expensive problem the moment marketing teams started generating at volume, because an LLM given three adjectives will fill the gap with the average of everything it has ever read — which is exactly the beige, breathless, synergy-adjacent prose everybody now recognizes as machine-written. The model is not failing. It was handed no constraint, so it produced the mean.

So the useful question is not "what is our voice?" It is: what would a stranger have to be told, in writing, to produce a sentence we would ship?

That is a contract. It has clauses. It can be violated, and you can point at the clause.

What goes in the contract

A brand voice contract that actually holds writers — human or model — to something has four parts.

Words, with the evidence behind them. Not "we are direct" but "we say what this costs you today; we never say leverage synergies." A say-this / never-say-this list, each entry traced to a real sentence you have already published. Evidence matters because it settles arguments: nobody has to litigate whether the brand is playful when there are eleven examples of it being playful in your own shipped copy.

Mechanics. Sentence length. Paragraph length. Whether you address the reader as you. Whether you use contractions, em dashes, questions as headings, first person plural. These are the levers that make prose recognizably yours, and they are trivially checkable — which is the point.

Trade-offs, decided. Every real voice sits somewhere on a spectrum, and the interesting part is which side you take when the two sides conflict. Warm or precise, when you cannot be both in one sentence? Plainspoken or authoritative? A guideline that says "both" has decided nothing. The contract names the winner and says what to do at the tie.

A self-check. The clauses have to be testable against a draft, or the contract is decoration. Given a paragraph, does it violate a never-say? Is the average sentence twice as long as the standard? Did it open with the product instead of the reader's problem?

None of that is new thinking about brands. Good copy chiefs have worked this way for decades. What is new is that a contract in this shape is directly executable by the machine doing most of the drafting.

What we built, and what is different about how it works

The Brand Voice lab is the same module that runs inside T2D3 OS, opened up so anyone can run it without an account. You paste in something you have already written — a page, a post, a launch email, a few of them — and it works in four moves.

It reads you back to yourself. Rather than asking what you want to sound like, it tells you what you already sound like, quoting your own sentences as the evidence. This is the move that changes the conversation. Asking a founder to describe their voice produces aspiration. Showing a founder their own three-clause sentences and asking "is this you, or is this an accident?" produces judgment. The system drafts, you decide — and disagreeing is the most valuable thing you can do, because the disagreement is signal nothing else can supply.

It makes you choose, by forcing scarcity. The trade-offs come as a MaxDiff: from a set of voice attributes, which is most like you and which is least. Rating everything on a five-point scale lets you say yes to all of it, which is how guidelines end up claiming a brand is simultaneously bold and humble. Forced choice takes that escape away. What survives is a ranking you actually made.

It checks its own work. Once the contract exists, the voice is run against sample output and graded on its own clauses. A contract that cannot catch a violation in a paragraph it just generated is not finished, and the self-check is what makes that failure visible before it reaches your audience rather than after.

It leaves with you, in machine form. The output is not a PDF for a folder. It is one instruction pack per place you actually write: a system prompt, a Claude Project instruction set, ChatGPT custom instructions, a written guide for human writers, and structured JSON for anything else. Paste it wherever your team drafts. That is deliberate — a voice that only works inside our product is a voice we have taken hostage, and a guideline nobody can carry to the tool they already use is a guideline nobody will use.

The voice itself is never behind a form. You see the whole finished thing before anyone asks for an email address. A lead magnet that hides its own output is an advertisement.

Where it goes if you keep it

Inside T2D3 OS the contract stops being a document and becomes a standard. It is part of the locked foundation — alongside your ICP, personas, and value propositions — which every downstream module inherits: content drafts, campaign copy, website pages, sequences. A recurring sweep grades published content against the clauses and tells you where the work drifted. When the voice changes, everything grounded in it is told that it is now stale.

That is the difference between a voice you wrote down and a voice that holds. The document has no way to notice it is being ignored. The contract does.

Try it on your own writing

Take one page you are proud of and one you are not, paste both in, and see what comes back. It takes a few minutes, needs no account, and the sandbox erases itself in thirty days unless you save it into a workspace.

The test is simple: read the say-this / never-say-this list and see whether a competitor could sign it. If they could, the contract is not done yet — and now you have something specific to argue with, which is more than three adjectives ever gave you.

Build your brand voice contract, free →