Most brand books are long, beautiful and mostly ignored. They describe a voice, show some logos, and list a few words to avoid somewhere around page thirty. Editors learn the real rules by getting things wrong and being corrected.
That knowledge lives in people's heads, which is why quality drops when someone is on holiday and why onboarding a new writer takes months. If you want a system to check drafts against a brand, the first job is to get those rules out of heads and PDFs and into a form that can be checked.
Separate what's checkable from what's taste
Some brand guidance can be checked mechanically: banned terms, required elements, claims the brand is never allowed to make. Some of it is judgement: warm but not gushing, confident but not salesy. Both matter, but they need different treatment.
- Banned terms and phrases, with the approved alternatives.
- Required elements: product name, the standard call to action, legal lines.
- Claims policy: what can and can't be promised.
- Voice notes: a few sentences, with examples of right and wrong.
The first three can be checked strictly and quoted exactly. The voice notes give the AI something to score against, but a person should always make the final call on taste.
Get a brand lead to sign it off
Extracting rules from a brand book is quick. The step that matters is someone who knows the client reading the result and saying yes, that's right. A rule list nobody has approved is just a guess with a confident tone.
Let decisions write the next rules
The brand book is only the starting point. The rules that make a system genuinely useful come from the editor's decisions. Every time an editor keeps a flagged phrase, that's a rule to stop flagging it for this brand. Every time they send a draft back, the reason is a new rule.
After a few weeks, the rules a system checks against are closer to how your best editor thinks than anything in the original PDF.
That's the real value: not a tool that knows the brand book, but one that learns the brand the way your team does, and never forgets what it learned.