Most brand voice guides were written for a world where a small team produced a small amount of content and one senior person read everything before it went out. That world is gone. Content now comes from freelancers, agencies, sales, founders and AI tools, in a volume nobody can read line by line. The guide still says "confident, warm, human", and every one of those contributors reads it differently.
Voice does not survive scale as a set of adjectives. It survives as a set of rules that someone, or something, can check. This is how to write those rules and run the review so consistency holds when output grows.
Why adjective guides fail
"Confident" means one thing to a founder and another to a junior copywriter. One writes short declarative sentences; the other writes "we are the leading provider of...". Both believe they followed the guide. An adjective describes a feeling, and a reviewer cannot check a feeling. A rule describes a behaviour, and a reviewer can.
Step one: turn each adjective into rules
For each quality in your current guide, ask "what would I see on the page if this were true?" and write that down. For example:
- "Confident" becomes: state the claim in the first sentence; no hedging words such as "may help to potentially"; no superlatives we cannot prove.
- "Plain" becomes: prefer the short word; one idea per sentence; explain any acronym the reader might not know on first use.
- "Human" becomes: write to one reader as "you"; use examples from real situations; no stock phrases from the banned list.
Step two: keep a use and avoid list
A short list of words and phrases you always use, and a longer list you never use, does more for consistency than pages of description. Include your product and feature names spelled correctly, the terms your buyers use for their problem, and the jargon and filler that creep in when writers are in a hurry. Update it whenever a reviewer catches something new. It should be a living document, not a PDF from the rebrand.
Step three: separate claim rules from voice rules
Tone problems are embarrassing. Claim problems are expensive. Treat them as a separate check with its own rules:
- Which claims you can make, and the source for each.
- Which claims need a citation on the page.
- Which claims you never make: guarantees, invented numbers, customer names without permission, comparisons you cannot support.
A piece can be perfectly on-voice and still make a claim you cannot back up. Reviewing for voice and claims in one pass usually means one of them gets skimmed.
Step four: show before and after
For every rule, include one sentence that breaks it and the rewrite that fixes it. Writers and reviewers learn far faster from pairs than from principles, and the pairs make disagreements quick to settle: does this new sentence look more like the "before" or the "after"?
Step five: set the range per channel
Voice is not the same as sounding identical everywhere. A LinkedIn post can be more personal than a pricing page; a support reply can be warmer than a technical doc. Write down what flexes by channel, such as sentence length, first person, humour, and what never flexes, such as claim rules, product names and banned phrases.
Step six: run review as a system
With rules in place, review can scale:
- Check every piece against the rules, not against the reviewer's taste. Findings should cite the rule they break.
- Fix at the source. If the same rule is broken repeatedly, the brief, the template or the guide is unclear. Fix that, not only the sentence.
- Version the guide. When a rule changes, record what changed and why, so older content can be checked against the new rule if needed.
- Audit what is already live. Old pages drift out of voice as the guide evolves. A periodic check of published content catches the pages that now contradict the rest.
A note on AI-drafted content
AI drafting makes this more urgent, not less. A model given a vague voice description will produce fluent, generic copy with a recognisable set of stock phrases. A model given specific rules, word lists, claim rules and before-and-after pairs, plus your real source material, produces something much closer to your voice. Either way, a person should approve anything that goes public.
How the Brand Crew runs this
The Brand Crew needs a brand voice reference plus your current copy and assets to work from. If your brand docs live in Google Drive, the crew can read them as source material. Each week it checks content and media assets against that reference and produces brand-adherence findings, with review-ready rewrites and asset fixes for anything that does not match.
It does not publish or replace anything by itself. People approve any change to the voice rules, and any rewritten copy or replaced asset, before it is published. The Content Crew drafts new work from your positioning, and a person reviews the draft, its claims and its destination before it goes live. The Brand Crew is measured on findings resolved and content approved without rework, which is the practical definition of a voice that holds.
See the connections on the integrations page and the step-by-step jobs in the how-to hub.
Start with your current guide
Open your voice guide and highlight every adjective. For each one, write two rules a reviewer could check and one before-and-after pair. Then pick five recent pieces of content and review them against the new rules. You will find out quickly which rules are clear and which need another pass.
Want to see how consistent your site reads today? Run the free audit.