How to Prompt AI to Match Your Brand Voice

Here's the same sentence, written two ways.

Generic AI: "We're excited to announce our new feature that will help streamline your workflow and boost productivity."

Brand-voiced: "We shipped the thing you asked for six weeks ago. It cuts your weekly reporting time from an hour to about ten minutes. You're welcome."

Same feature announcement. Same underlying facts. One sounds like it was written by a corporate press release generator, because it was. The other sounds like a person who built the thing and is mildly proud of it. The difference isn't a better model or a longer prompt. It's that the second version was written with an actual voice on file, and the first wasn't.

Why AI defaults to generic voice

Large language models are trained on an enormous average of internet writing. Left alone, they regress to the mean: safe, upbeat, mildly corporate, full of words like "streamline," "elevate," and "excited to announce." That's not a flaw, it's the expected output of a system with no information about who you are. When you type "write a launch email for our new feature" with no other context, you get the statistical average of every launch email ever written. It's competent and forgettable.

The fix isn't asking the AI to "sound more human" or "be more casual." Those instructions are too vague to change anything meaningful, similar to the failure pattern covered in how to fix a bad AI prompt: a vague ask produces a vague, hedged answer. What actually works is giving the model a specific, structured description of your voice, the same way you'd brief a new freelance writer on their first day.

What a brand voice prompt actually needs

A style guide for AI prompts doesn't need to be a 20-page brand book. It needs four things, and all four should live in the same prompt or system message every time you ask for writing:

1. Three to five tone adjectives, ranked, not just listed. "Direct, warm, a little irreverent" tells the model something specific. "Professional and friendly" tells it almost nothing, because that describes 90% of business writing already.

2. A do list: concrete word choices, sentence patterns, or habits you want. Short sentences. Contractions. Specific numbers instead of "a lot." Second person.

3. A don't list: the exact words and patterns you're tired of seeing. This is the highest-leverage part of the whole prompt, because it directly counters the model's default habits. Ban "streamline," "seamless," "unlock," "game-changing," exclamation points, or whatever your team already rolls its eyes at.

4. A sample passage, 100 to 200 words of writing that already sounds like you. Real writing, not a description of writing. Models imitate examples far more reliably than they follow adjectives alone.

That fourth item matters more than people expect. Telling a model "be conversational" is an instruction it has to interpret. Showing it 150 words of actual conversational writing from your blog or your best support reply is a pattern it can match. If you've already put effort into giving AI context about your business, a voice sample is the writing-specific version of that same idea: concrete material beats abstract description every time.

The reusable brand voice prompt template

Copy this into a note, fill in the brackets once, and paste it at the top of any writing prompt, whether that's a blog draft, a marketing email, or a customer support reply.

```

BRAND VOICE

Tone (ranked, most important first): [e.g. direct, warm, a little irreverent]

Sentence style:

• [e.g. Short sentences. Average under 20 words.]

• [e.g. Contractions always. "We're," not "we are."]

• [e.g. Second person ("you"), rarely first person plural ("we").]

Words and phrases to use:

• [list 5-10 words that sound like you]

Words and phrases to avoid:

• [list every overused AI phrase and corporate filler you're sick of]

Sample passage to match (match the rhythm and word choice, not the topic):

"[Paste 100-200 words of your actual best writing here]"

TASK: [the actual thing you want written]

```

Once this block exists, reuse it verbatim for every writing task. It works as a one-off prompt in ChatGPT or Claude, and it works even better as the persistent instructions in a custom AI assistant you set up once and reuse, so nobody on the team has to re-paste it each time.

Worked example: before and after

Say a project management tool for freelancers wants a short paragraph explaining a new invoicing feature. Here's the plain prompt versus the same prompt with the brand voice block attached.

Plain prompt: "Write two sentences about our new automatic invoicing feature."

Output: "We're thrilled to introduce automatic invoicing, a powerful new feature designed to effortlessly transform your billing process. This exciting update will help you save time and get paid faster than ever before."

Same prompt with the brand voice block added:

```

Tone: direct, dry, allergic to hype

Sentence style: short, no exclamation points, no adjectives before "feature"

Avoid: "effortlessly," "powerful," "exciting update," "we're thrilled"

Use: specific numbers, plain verbs

TASK: Write two sentences about our new automatic invoicing feature.

```

Output: "Invoices used to take you 20 minutes to build and send. Now it's automatic, and you get paid on average four days faster."

Both answers describe the same feature. The second one sounds like it came from a company with an opinion instead of a template. Note that the "four days faster" number is illustrative for this example, not a real benchmark, which is exactly the kind of specific claim you'd want to swap for your own verified data before publishing anything with it.

Making it stick across a whole team

A brand voice prompt only solves the consistency problem if everyone who touches AI-generated content actually uses it. A few practical habits make that happen:

• Store it centrally. A shared doc, a pinned Slack message, or a custom GPT/project with the voice block baked in beats everyone keeping their own copy, which drifts within a month.

• Update the sample passage periodically. As your actual writing improves, swap in fresher examples. A stale sample from two years ago teaches the model an old voice.

• Pair it with business context, not just tone. Voice is how you sound; context is what you know. A prompt with strong tone instructions but no facts about your product will still write confident-sounding nonsense. This is worth pairing with the broader approach in prompt engineering, which covers structuring prompts for accuracy, not just style.

• Test it on your hardest content type first. If the voice block survives a technical support reply or a pricing objection email without collapsing back into generic phrasing, it's solid enough to trust everywhere else.

Common questions

How long should a brand voice prompt be?

Long enough to include all four elements, tone words, a do list, a don't list, and a sample passage, but rarely more than 200-300 words total. The sample passage does more work than extra adjectives, so if you have to cut something, cut adjectives before you cut the example.

Can I use my "About" page as the sample passage?

Usually not well. About pages are often written in a more formal, marketing-adjacent voice than your actual product writing or support replies. Pick a sample from the content type closest to what you're generating, a blog post for blog prompts, a support reply for support prompts.

Does this work the same way in ChatGPT and Claude?

Yes. Both models respond well to explicit tone instructions plus a sample to imitate. The main difference is where you put it: as a one-off prompt prefix in either tool, or as persistent system-level instructions if you're building a repeatable assistant.

Why does the AI drift back to generic phrasing in a long conversation?

Models weight recent context heavily, so after many back-and-forth exchanges, the original voice instructions can get diluted. Re-paste the brand voice block every few exchanges in long sessions, or better, put it in a system prompt that persists automatically rather than relying on it staying in the model's working memory.

Is a brand voice prompt the same as a style guide?

It's a compressed, machine-readable version of one. A full style guide might cover grammar rules, formatting, and legal language. A brand voice prompt only needs the parts that change how sentences sound: tone, banned words, preferred words, and an example. Keep the rest of your style guide separate.