How to Run a 20-Minute Founder Voice Audit Before Using AI

You want AI to draft your content, but you don't want it to sound generic or fake. Fair.

Ask yourself 20 questions in three groups: what you believe, how you speak, and what you will not say. Write specific answers. Don't hide behind words like "professional" or "friendly." Those labels are too weak.

Turn your answers into a short voice profile with clear rules and examples. That profile can guide any AI writing tool. It can also help you catch the parts of a draft that don't sound like you.

Why a tone label is not a voice profile

Words like "clear," "bold," and "friendly" aren't useless. They just aren't enough on their own.

An AI can write something "bold" that sounds nothing like you. It can sound friendly while saying nothing useful. It can sound professional because it stuffed the sentence with empty words.

A useful voice profile records actual choices:

Here's an illustrative example. Imagine a fictional founder of support software for small teams.

Without a voice profile, a product update might say:

"We are excited to announce our powerful new support workflow. It helps teams streamline communication and improve customer success."

With a real voice rule, the same update might say:

"A bigger support dashboard will not fix a messy process. This update gives small teams one clear place to see what needs an answer next."

Both versions describe the same update. Neither needs slang, jokes, or dramatic language.

The second version makes a decision. It says what will not solve the problem. Then it explains the update in plain words.

The selected founder belief here is simple: clear beats clever every single time. If I cannot explain my value in one sentence, I simplify it.

That belief gives the AI something useful. It changes the decision behind the sentence, not just the surface style.

The first seven questions: find your beliefs

Your beliefs give your content a point of view. They explain why you choose one idea over another.

Ask yourself these seven questions:

Don't answer with labels. "I care about quality" tells the AI almost nothing. Tell it what quality means to you.

A concrete answer might look like this:

"I believe clear beats clever every single time. If I cannot explain my value in one sentence, I simplify it."

This is the selected founder belief. It isn't a universal fact. It is a clear opinion the founder can use when making content decisions.

Each answer can become:

Make the answers specific enough that another person could tell when a draft follows them and when it doesn't.

The next seven questions: define how you speak

Now deal with the mechanics of your writing. This turns a general tone preference into rules an AI can follow.

Ask yourself:

Don't overthink the word "feel." You aren't promising a reaction. You're describing the kind of reading experience you want your writing to create.

An illustrative example for a fictional founder might look like this:

"Use short sentences. Name the problem before the feature. Explain technical terms with a concrete example. Remove inflated claims."

That's a real writing profile. "Be bold and friendly" isn't.

The rules tell the AI what to do. The examples show what those rules look like in practice. You need both.

If you say "use simple language," give an example. If you say "be direct," show a sentence you would actually write. Otherwise, the AI fills the gap with its own idea of simple or direct.

The final six questions: set your red lines

Red lines stop the AI from making claims or choices you would never stand behind.

These are your guardrails. Write them before you need them.

An illustrative red-line profile for a fictional founder might say:

"Do not promise a result without evidence. Do not invent customer stories. Do not publish a claim I would not defend in a direct conversation."

These rules give you a final check before approval. They don't replace fact-checking. They don't replace your judgment. They simply make it harder for a draft to slip past your standards.

Be specific about what you will not say. "Don't be dishonest" is too broad. "Don't promise a result without evidence" gives the AI and the founder a clear rule to check.

Turn the answers into an AI-ready profile

The audit becomes useful when you reduce it to three short blocks:

Keep the profile short. The full answers help you think. The short profile helps you use the result again and again.

Here's an illustrative completed profile for a fictional founder:

"I believe clear beats clever. I write in short sentences and explain technical ideas with examples. I do not make claims without proof or use inflated product language."

Give this profile to the AI before you give it a content task. Then give the AI the topic, audience, and goal.

You can also ask the AI to point out any line in its draft that breaks the profile. Treat that as a recommendation, not a guarantee. Read the draft yourself.

RollKind starts with one interview to learn how you think, what you believe, who you talk to, and how you sound. That interview captures the same beliefs, audience, and language choices covered in this voice audit. You still decide whether each draft is true, useful, and right for your voice.

The founder owns the final choice. The profile can guide the draft, but it cannot decide whether the draft is worth publishing.