What changes when your guidelines become a BCP.
A brand book is written for a designer who flips to the right page. An assistant does not flip. It reads what it is given, in the order it is given. Restructuring the same content into addressable files lets a tool ask for the part it needs.
Where typical guideline sections land.
| Usual PDF section | BCP file | What gets added |
|---|---|---|
| Brand story, mission | brand.md, values.md | Short factual identity block an assistant can quote |
| Tone of voice | voice.md | Preferred and avoided vocabulary as lists, channel-by-channel tone |
| Words we do not use | voice.md, voice/anti-ai.md | AI-specific patterns to avoid, with examples |
| Legal-approved claims | claims.md | Each claim with evidence, caveat, and status: permitted, needs caveat, or forbidden |
| Brand safety, sensitive topics | boundaries.md | Hard and soft limits written as rules |
| Boilerplate, about us | representation.md | Preferred framing and descriptions to avoid |
| Logo, color, type | visual.md | Tokens, approved pairs, forbidden modifications, asset URLs |
Hypothetical example. Northwind Coffee, a fictional brand, has a PDF that says "We are warm but never cutesy." In voice.md that becomes a tone line plus an avoid list: no puns on "brew", no "coffee lovers", no exclamation points. Their PDF says the beans are "ethically sourced", and legal requires the certifier's name every time. In claims.md that claim is marked as needing a caveat, with the exact required wording. An assistant asked for an Instagram caption now has both rules in one place.
What the interview fills in.
Guidelines rarely say what the brand refuses to claim, or how to describe it when compared with a rival. Encoder asks. The answers go into claims.md, boundaries.md, and representation.md.