Brand Voice Guide for AI: How to Write Down a Voice a Machine Can Follow

A brand voice guide for AI replaces adjectives with rules a machine can follow: typical sentence length, words and phrases you use and never use, punctuation habits, how you open and close, and real samples of your writing with notes on why they work. "Friendly but professional" tells AI nothing; "short sentences, no exclamation marks" tells it plenty.

Why don't normal brand guidelines work for AI?

Because most brand guidelines describe a feeling, not a behaviour. "Warm, confident, approachable" helps a human writer who already knows the company. An AI reads those words and produces the average of every warm, confident, approachable piece of writing it has seen, which sounds like everyone else.

AI writing drifts towards the probable. A February 2026 article in The Register, shared by Kottke.org, called the effect "semantic ablation": the distinctive parts of a text get replaced with the most probable, generic wording. A voice guide is how you push back. It gives the model specific, unusual choices to make instead of default ones.

There is also a people problem. According to Microsoft and LinkedIn's 2024 Work Trend Index, only 39% of people using AI at work have had AI training from their employer. Without a shared guide, each person describes the voice to their AI in their own words, and the brand fragments.

What should a brand voice guide for AI include?

It should include six things: sentence mechanics, vocabulary and signature phrases, banned words and phrases, punctuation habits, structural patterns such as how pieces open and close, and annotated samples of real writing. Each should be specific enough that you could check a draft against it line by line.

Punctuation deserves attention. Long dashes, for example, are rare in most people's writing and common in AI output, so a reader may treat them as a sign of machine writing. If your company never uses a mark, the guide should say so, and say what to use instead.

How do you capture the voice you already have?

Collect your best writing, such as the pages, emails and posts your team is proudest of, and study what they do consistently: sentence length, recurring phrases, punctuation, openings and closings. Write those patterns down as rules, then add the samples themselves. The voice exists already; the job is describing it precisely.

A practical method:

1. Gather 10 to 20 strong pieces across channels, written by people, not AI.
2. Measure the mechanics. Average sentence length, use of questions, paragraph length.
3. List repeated phrases and words, and the ones conspicuously absent.
4. Note punctuation habits, including marks never used.
5. Annotate three samples with why each works.
6. Test by asking AI to write a piece, then comparing it with the samples.

Your customers' language belongs nearby too. A voice is not only how you talk; it is how close you sit to how your buyers talk. The guide to a customer quote bank covers collecting their words.

How is a person's voice different from the company's?

A company voice is how the business sounds when the business is the author. A person's voice is how a named individual sounds when something goes out under their name: their stories, opinions, humour and habits. They need separate guides, because a founder writing on LinkedIn should not sound like the website.

The byline decides which guide applies. The company name on a piece means the brand voice; a person's name means their own voice file. Each named person should confirm their own file, because only they can say whether it sounds like them.

The same method works for both. For a person, though, the sources are their own posts, talks and recorded calls rather than company pages, and the guide records their positions as well as their style.

How do you check AI followed the voice guide?

Check drafts against the guide's measurable rules, such as sentence length, banned words and punctuation, using a simple script or checklist, then read for tone against the samples. Mechanical checks catch most drift; a human read catches the rest. Fold anything that slips through back into the guide.

Mechanical checks are easy to automate: flag banned phrases, count long sentences and search for forbidden punctuation. They catch the problems people stop noticing after the tenth draft.

People notice when AI lacks this context. According to Atlassian's 2026 Pulse survey, only 12% of workers say their AI understands their organisation like a tenured senior colleague. A voice guide is part of what that colleague knows without being told.

Brand voice is one of the core sets in a company brain.

Skein builds voice guides like this as part of the shared knowledge layer of a Company OS, with separate files for named people. That suits a team where several people publish with AI. A solo writer may only need a short list of rules and three samples.

Frequently asked questions

How do I make AI write in my brand voice?

Give it a voice guide with specific rules rather than adjectives: typical sentence length, words and phrases to use and avoid, punctuation habits, how pieces open and close, and annotated samples of your best writing. Then check drafts against those rules and add anything that slips through to the guide.

What should a brand voice guide include?

For AI use, include sentence mechanics, vocabulary and signature phrases, banned words, punctuation rules, structural patterns, and three or more annotated samples. Each rule should be specific enough to check a draft against. Vague descriptions like "friendly" or "professional" should be replaced with observable behaviours.

Why does AI ignore my brand guidelines?

Usually because the guidelines describe feelings rather than behaviours. Words like "warm" or "confident" give AI no specific instruction, so it falls back on average writing. Rules about sentence length, banned phrases and punctuation, plus real samples, give it concrete choices to follow.

Should a founder have a separate voice guide?

Yes, if pieces go out under the founder's name. A person's voice includes their stories, opinions and habits, which differ from the company's. The byline decides which guide applies, and only the founder can confirm their own voice file is accurate.

How long should a brand voice guide for AI be?

Short enough to be read in full before every task: usually two to three pages of rules plus a few annotated samples. A long brand book can sit behind it for designers and agencies, but the AI-facing guide should be concise and specific.

Related reading