Logos and colors are mostly solved. The hard part — the part AI keeps missing — is voice. CRAFT captures your brand once, stores it in a dedicated folder, and turns it into a constraint your AI checks against rather than a guideline it tries to remember.
Last week we covered CRAFT’s project layout — six folders that make every project legible. This week: the optional seventh folder, where CRAFT stores the part of your brand that AI keeps missing.
Why AI Output Doesn’t Sound Like You
Most teams have a brand book somewhere. A Notion page, a PDF, a Figma file with the logo grid, the color palette, and a tone-of-voice section that ends with three example sentences. It exists. People reference it occasionally. The designer keeps it open.
When an AI session generates content, none of that is actually present. The model has a generic notion of “professional” or “friendly” — calibrated to the average of its training data, not to you. Every session produces output that’s fine and not yours. Logo placement might be right. The voice almost never is.
That gap — between what the brand book says and what the AI produces — is paid in editing time. You read the output, hear that it doesn’t sound like you, and rewrite the sentences. Or you ship it anyway because the deadline doesn’t care about voice. Either way, the brand is decoration, not enforcement.
The Brand Kit Builder
CRAFT’s answer is RCP-CWK-040: Brand Kit Builder, a recipe that captures brand identity as a structured file rather than a document. The output is a brand config — colors, typography, voice rules, messaging frameworks — that subsequent recipes can read.
The recipe runs in two modes.
Asks the essential questions: what colors, what fonts, what voice. You answer a Q&A and walk away with a usable config in fifteen minutes. The on-ramp.
Leverages CRAFT’s Brand ID Cookbook for deeper capture: logo usage rules, color systems with semantic roles, typography hierarchies, voice/tone matrices, messaging frameworks per audience.
The dual-mode design matters. Most projects don’t need the full identity system — they need a config file the AI can check against. Standalone is the on-ramp. Enhanced is for projects where brand is the product.
The Brand ID Cookbook
When projects do need the full system, the Brand ID Cookbook (CFT-FWK-COOKBK-BRAND-ID.txt) is a dedicated cookbook inside the CRAFT framework — same status as Cowork recipes, Studio recipes, or Admin recipes. Brand identity isn’t a footnote in the project documentation. It’s a recipe category.
Why a whole cookbook? Because brand identity is structurally different from one-off configuration. It has rules with exceptions (logo on dark vs light backgrounds), hierarchies (primary brand voice vs sub-brand voice for community channels), and dependencies (messaging frameworks reference voice rules; voice rules reference values). Treating it as a cookbook means brand decisions get the same versioning, the same recipe discipline, and the same reuse story as the rest of the framework.
The Enforcement Layer
A config file the AI can read is not the same as a constraint the AI does check. The enforcement layer is COM Branding Detection, an extension to the COM (CRAFT Operations Manager) recipe that monitors generated content for brand-guideline compliance.
When the AI produces a draft, COM scans it against the brand config and flags inconsistencies. Voice slipping toward generic-corporate? Flagged. Wrong color hex in a generated HTML block? Flagged. Logo treatment in a graphic prompt that violates the usage rules? Flagged.
This is the part that makes brand operational rather than aspirational. The PDF brand book in the shared drive doesn’t catch it. The COM extension does — at the moment of generation, before the human has to read every word and re-edit.
The Seventh Folder
The brand kit needs somewhere to live. CRAFT adds branding/ as the optional seventh folder in the standard project structure — sitting alongside the six required folders we covered last week.
Adding a top-level folder sounds expensive. In a less disciplined framework, it would be — every recipe that references the project layout would need to update. CRAFT’s design absorbed the change cleanly: ten files needed updating, none of them critical-path. The folder structure spec, the project setup recipe, a few file-path references. The rest of the framework didn’t notice.
A framework is extensible if you can add a major category without rewriting it. Branding was the test case. Ten files. (LL-090 in the lessons learned log, for anyone tracing the receipts.)
The Counterintuitive Part
Here’s what surprised us during testing: visual brand is mostly the easy part. Logos and colors are settled by the time the brand kit is built — they’re already in Figma, already in the launched product, already correct in ninety percent of generated content. Voice is where it happens.
Voice is where AI defaults to the average. Concise becomes thorough. Specific becomes generic. The em dash becomes a comma. The phrase you use deliberately becomes the phrase the training corpus uses. None of these are wrong. They’re just not you.
The brand kit’s most valuable section, it turns out, is the voice/tone matrix — the part that captures how you talk, what you don’t say, what your sentences sound like read out loud. That’s the section COM Branding Detection earns its keep on. The logo file is easy. The voice is the work.
The Recursive Proof
CRAFT used its own Brand Kit Builder on itself. The framework’s brand — Systematic Futurism, the color palette, typography, voice rules — was captured via RCP-CWK-040 in handoff H052, stored in branding/, and validated against COM Branding Detection in the same session. The Brand Kit Builder built CRAFT’s brand.
This isn’t a marketing flourish. It’s the test that proves the recipe works on a non-trivial subject. If a brand-capture recipe can produce a usable config for the framework that defines the recipe category, the recipe is doing its job.
Why This Matters
Brand identity has been the part of project work that survives by goodwill — the designer keeping the file open, the writer remembering the voice rules, the team catching off-brand output in review. Goodwill scales badly. Five sessions a week across three contributors means fifteen chances per week for the brand to slip, and the brand book has no way to intervene.
CRAFT turns the brand into a constraint the AI checks against. Captured once, referenced every session, enforced at generation. The decoration becomes the enforcement.
This is Week 7 of our 8-week capability spotlight. Each week we go deep on one part of CRAFT — how it works, what problems it solves, and how to use it. One week left.
