Over the past three posts, we’ve argued that brand guidelines are fiction the moment they’re created, that visual and verbal identity are governed in separate silos, and that machine-readable brand data produces measurably better AI output than prose guidelines. Now the practical question: how do you actually build this?
The answer is less complicated than it sounds, and it doesn’t require buying new software.
The three-layer stack
Everything we’ve built runs on three tools our team already used: Notion, Figma, and Claude. No new procurement. No platform migration. No vendor lock-in. The architecture is what matters — the tools are interchangeable.
Visual identity as structured data
Figma
Your design system probably already contains most of this. The step most teams skip is adding governance metadata to existing tokens. A color token that says #f44d37 is useful. A color token that says #f44d37, use for CTAs, never as background fill, max 15% of frame area is governable.
Verbal identity as structured data
Notion
We store verbal identity across five connected databases: a Perspective page, a Claims database with confidence scoring, a Narrative Library, Application Rules, and an Evidence database. Five databases. Standard Notion functionality. The power is in the structure.
Compilation and enforcement
Claude + MCP
Model Context Protocol lets AI tools read from multiple data sources in a single context. The compilation step is what makes this a Brand OS rather than a brand guide. When we change a narrative element, every application rule that references it is affected.
What you can build this week
Encode your visual tokens with governance metadata. Add use/never rules to each token. This alone improves AI output quality.
Build a claims inventory. Score each assertion by confidence. You’ll find half your marketing rests on claims nobody has checked in a year.
Write application rules for your top three content types. Store these as structured records, not prose.
Connect them. Give an AI tool access to all three layers and generate something. The difference is usually obvious on the first attempt.
The honest caveats
The verbal identity format is proprietary. We’d love to see a W3C-equivalent specification for machine-readable brand messaging. Measurement is still primitive. The tools will change. But the architecture — the principle that brand identity should be structured data, not prose documents — will outlast any specific implementation.
The category that doesn’t have a name yet
The intersection of AI governance and brand management — AI brand governance as a named, defined service category — doesn’t formally exist yet. We think it will within 12 to 18 months.
The teams building the methodology now will have a meaningful advantage over those who wait for someone else to define it. We’re building in public because we believe this architecture should be widespread, not proprietary. If this series resonated, reach out — we’re working with teams on building these systems.
The architecture will outlast any specific implementation.