How to Create a Brand Knowledge Graph for AI

What is a brand knowledge graph?

A knowledge graph is a structured, machine-readable representation of brand facts and relationships. Instead of prose ("We are a B2B SaaS company that values clarity"), a graph says:

Organization(name: "Example Co")
  -> knownAs: ["B2B SaaS", "design systems"]
  -> hasValue: [Value(name: "Clarity"), Value(name: "Evidence")]
  -> serves: Industry(name: "Product teams")

Agents can query this graph: "What are our key values?" or "Which industries do we serve?" They get explicit answers, not guesses from embeddings.

Why brand needs a graph

Brands are networks of relationships:

• Voice shifts by context (support vs. marketing vs. social)
• Colors have roles (primary, accent, background)
• Messages ladder from general to specific
• Rules constrain behavior in different channels

A PDF flattens this to prose. A graph preserves structure and relationships.

JSON-LD: The standard format

JSON-LD is the standard for semantic structured data on the web. Google uses it, Schema.org standardizes it, and JSON-LD is readable by machines.

Here is a minimal brand graph:

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Example Co",
  "url": "https://example.co",
  "slogan": "Evidence over emotion.",
  "description": "B2B SaaS for product teams.",

  "knowsAbout": ["B2B SaaS", "product management", "design systems"],

  "hasValue": [
    {
      "@type": "Thing",
      "name": "Clarity",
      "description": "No jargon, no fluff. Ideas stated plainly."
    },
    {
      "@type": "Thing",
      "name": "Evidence",
      "description": "Every claim backed by data or user research."
    }
  ],

  "voiceContexts": [
    {
      "@type": "Thing",
      "name": "support",
      "tone": "empathetic, concise, action-oriented",
      "exampleMessage": "We apologize for the issue. Here is the fix."
    },
    {
      "@type": "Thing",
      "name": "marketing",
      "tone": "bold, evidence-led",
      "exampleMessage": "87% of teams ship faster with our tool."
    }
  ],

  "hasBrandColor": [
    {
      "@type": "Thing",
      "name": "Primary",
      "hex": "#9c4221",
      "usage": "CTAs, hero sections, key links",
      "doNot": "Text on light backgrounds"
    },
    {
      "@type": "Thing",
      "name": "Accent",
      "hex": "#d97706",
      "usage": "Secondary actions, highlights"
    }
  ],

  "typography": [
    {
      "@type": "Thing",
      "name": "Headlines",
      "font": "Instrument Serif",
      "weight": "regular",
      "tracking": "0.02em"
    },
    {
      "@type": "Thing",
      "name": "Body",
      "font": "DM Sans",
      "size": "16px",
      "lineHeight": "1.6"
    }
  ]
}

Building a graph step by step

Step 1: Identify entities

What are the nouns in your brand?

• Organization (your company)
• Values (clarity, evidence, craft)
• Industries (who you serve)
• Personas (user types)
• Voice contexts (support, marketing, social)
• Colors, typography, messaging

{
  "entities": [
    { "id": "org_example", "type": "Organization", "name": "Example Co" },
    { "id": "value_clarity", "type": "Value", "name": "Clarity" },
    { "id": "context_support", "type": "VoiceContext", "name": "support" },
    { "id": "color_primary", "type": "Color", "name": "Primary", "hex": "#9c4221" }
  ]
}

Step 2: Define relationships

How do entities connect?

• Organization -> has Value
• Organization -> serves Industry
• Value -> implies VoiceRule
• VoiceContext -> uses Tone
• Color -> has Usage rule

{
  "relations": [
    { "from": "org_example", "type": "hasValue", "to": "value_clarity" },
    { "from": "org_example", "type": "serves", "to": "industry_product" },
    { "from": "context_support", "type": "hasTone", "to": "empathetic" },
    { "from": "color_primary", "type": "usedFor", "to": "cta" }
  ]
}

Step 3: Add rules and constraints

Rules are the edges that govern behavior. They are queryable and testable.

{
  "rules": [
    {
      "id": "rule_support_apologize",
      "context": "support",
      "rule": "Apologize once, fix forward.",
      "negative": "Never apologize twice or dwell on the problem."
    },
    {
      "id": "rule_marketing_claims",
      "context": "marketing",
      "rule": "All claims must cite data.",
      "negative": "Never use vague adjectives like innovative or cutting-edge."
    }
  ]
}

How agents use graphs

Once you have a graph, agents can query it:

Query: "What tone should I use in a support email?" Graph response: Look up VoiceContext(name: "support") -> hasTone ["empathetic", "concise", "action-oriented"]

Query: "Can I use primary color for text?" Graph response: Look up Color(name: "Primary") -> doNot ["text on light backgrounds"]

Query: "What industries do we serve?" Graph response: Look up Organization -> serves [Industry("product teams"), Industry("design systems")]

No interpretation. No guessing. Explicit answers.

Integration with AI workflows

Your graph lives alongside CLAUDE.md and .cursorrules:

repo/
  CLAUDE.md              # narrative instructions
  .cursorrules           # editor-specific rules
  brand-graph.jsonld     # machine-queryable knowledge

When an agent needs to generate something:

1. Read CLAUDE.md for narrative context
2. Query brand-graph.jsonld for specific rules
3. Apply both to the task
4. Generate output that satisfies both

BrandMythos and knowledge graphs

When you upload a brand guide to BrandMythos, we:

1. Extract entities (colors, values, personas, rules)
2. Infer relationships (which color is for which use)
3. Generate JSON-LD in standard format
4. Validate against schema.org
5. Export alongside CLAUDE.md and .cursorrules

The graph is version-controlled, diffable, and synced to your repos. Learn more about how we model brand DNA from multiple sources.

Real example: Using a graph to score brand compliance

With a knowledge graph, you can automate compliance scoring:

Generated text: "Our innovative solution transforms your workflow."

Compliance check:
  1. Look up words in brand rules
  2. Find rule: VoiceContext("marketing") -> negative ["innovative"]
  3. Flag: "innovative is forbidden in marketing context"
  4. Compliance score: 92% (one forbidden word out of 8)

Recommendation: Replace "innovative" with "faster" (evidence-led)

Next steps

Start with a simple graph:

• 5-10 core values
• 3-5 voice contexts
• 2-3 color roles
• 5-10 key rules

Use JSON-LD format. Commit to Git. Query programmatically.

As the graph grows, agents get smarter. Your brand becomes infrastructure instead of documentation.

Try BrandMythos with your brand. We build the graph for you, automatically.