How to Write a Brand Voice Guide an AI Writer Can Actually Follow

If you've ever handed an AI writer three adjectives — "friendly, professional, confident" — and gotten back a draft that reads like every other company's About page, the problem isn't the tool. It's that adjectives aren't instructions. A brand voice guide for AI writing needs to give a model something it can actually check a sentence against, not a mood to guess at.

Key takeaways:

• Adjectives alone ("warm," "bold," "authentic") produce generic output because they have no fixed definition an AI can test a sentence against.
• A working one-page guide has 5 to 7 named sections, not a long brand book — anything longer gets skipped by both writers and AI tools.
• Every tone rule needs one before/after sentence pair; a rule without an example is just another adjective in disguise.
• A banned-words list of 10 to 20 terms (jargon, hedges, overused claims) removes more generic-sounding text than any positive instruction does.
• Testing the guide means running one real draft through it and counting how many sentences break a named rule — not re-reading for a general "feel."

Why Does AI-Written Content Sound Generic?

AI-written content sounds generic because most brand instructions are adjectives, and adjectives don't constrain word choice or sentence structure — they just describe a feeling the model has to invent on its own. Tell a model to sound "professional" and it will default to the most statistically common version of professional writing it has seen, which is the same version every other brand's AI draft defaults to.

The fix is to replace each adjective with a rule that can be checked against a sentence. "Confident" becomes "state the claim directly, no hedging words like 'we believe' or 'it's possible that.'" "Warm" becomes "use the reader's problem in the first sentence, not the company's product." A rule like this gives an AI writer — or a human editor — something to test a draft against, sentence by sentence.

What Belongs in a One-Page Brand Voice Guide?

A one-page guide needs five to seven named sections, each answering a single question a writer or AI tool would otherwise have to guess at. Longer documents get skipped; shorter ones leave too much room for interpretation.

The sample paragraph does more work than any rule on the page — an AI writer matching a real paragraph performs better than one matching a description of a paragraph.

How Do You Describe Tone With Examples Instead of Adjectives?

You describe tone by pairing each rule with one sentence that fails it and one that follows it, side by side. A rule without a before/after example is functionally the same adjective it was meant to replace.

Here are three rule types with worked pairs:

1. Direct claims, no hedging.

Before: "This might help improve your results over time."

After: "This cuts your first-draft time by half."

1. Reader's problem first, not the product.

Before: "Our platform offers an all-in-one content solution for businesses."

After: "If your blog hasn't been updated in three months, here's what to fix first."

1. Numbers instead of vague scale words.

Before: "Many small businesses struggle with content."

After: "7 of 10 small business sites we reviewed hadn't published in over 60 days."

Three to five pairs like these, sitting inside the "Tone rules" section of the one-page guide, do more to shape an AI draft than three paragraphs of description ever will.

Words and Claims Your Brand Should Never Use

A banned-words list should name 10 to 20 specific terms, not a vague instruction like "avoid jargon." Specificity is what makes the list enforceable by a human editor or an AI tool checking its own output.

Common categories worth banning outright:

• Empty intensifiers: "cutting-edge," "game-changing," "revolutionary," "world-class"
• Hedging phrases: "might help," "could potentially," "in many cases"
• Unverifiable superlatives: "the best," "the only," "unmatched" — unless you can back the claim with a number
• Filler transitions: "in today's fast-paced world," "at the end of the day"
• Vague scale words: "many," "most," "a lot of" — replace with an actual count or percentage whenever one exists

List the actual words, not the category. "Avoid buzzwords" is itself the kind of vague instruction this guide exists to eliminate — put the ten specific words your writers keep reaching for.

How Do You Test the Guide on a Real Draft?

You test the guide by running one real piece of content through it and marking every sentence that breaks a named rule — a general re-read for "does this feel right" defeats the purpose of having written rules at all.

A simple four-step test:

1. Take a draft the AI writer already produced — ideally one that felt generic to you.
2. Go rule by rule from the guide (sentence length, banned words, tone pairs, numbers policy) and mark every line that fails.
3. Count the failures. More than 3 to 4 broken rules in a 500-word draft means the guide's instructions are still too abstract — go back and turn the vaguest rule into a before/after pair.
4. Rewrite the marked sentences using the after-example pattern from the matching rule, then feed the corrected draft back to the AI writer as your first "gold standard" reference.

Once a draft passes with fewer than 2 rule breaks, that draft becomes the sample paragraph in your one-page guide for the next round. If you want to see how your current site content reads before you build this guide, you can run the free scan on your own pages and use the results as your starting "before" example.

A brand voice guide that works isn't a longer document — it's a shorter one made of rules a writer, human or AI, can actually check a sentence against.