How to Build a Brand Context Folder for Claude Code: Voice, Visual Identity, and Positioning
A brand context folder gives Claude your voice, design tokens, and ICP so every output sounds like you. Here's how to build one in 30 minutes.
Why Claude Keeps Getting Your Brand Wrong (And How to Fix It)
If you’ve used Claude for content work, you’ve probably hit this wall: the output is competent, but it doesn’t sound like you. The tone is slightly off. The colors it describes don’t match your palette. The messaging hits generic beats instead of your actual positioning. You end up editing everything — which defeats the purpose.
The fix isn’t prompting harder. It’s giving Claude a brand context folder.
A brand context folder is a structured set of reference files you drop into your project that tells Claude everything it needs to know about your brand before it writes a single word. Voice guidelines. Design tokens. Ideal customer profile. Positioning statements. Once it’s in place, Claude stops guessing and starts producing output that actually sounds like you.
This guide walks you through building one from scratch in about 30 minutes. No design degree or brand strategy background required.
What a Brand Context Folder Actually Is
Think of it as the briefing document you’d hand to a new freelance writer or designer on day one. Except instead of a PDF that gets skimmed once and forgotten, it lives in your project directory where Claude reads it every time.
In Claude Code, you can place context files in your project root and reference them explicitly in your prompts, or use a CLAUDE.md file that Claude automatically picks up as persistent context. Either approach works — what matters is that the information is structured, specific, and consistent.
A complete brand context folder typically contains:
- A voice and tone document — how you write, what you avoid, and examples of both good and bad output
- A design token file — your colors, typography, spacing, and other visual variables in a format Claude can reference
- A positioning doc — your ICP, value proposition, competitive differentiation, and messaging hierarchy
- An examples file — real copy samples that demonstrate your brand in action
Each file is plain text or Markdown. No special format required. The goal is to make implicit brand knowledge explicit enough that an AI (or a new human hire) can act on it without guessing.
Step 1: Build Your Voice and Tone Document
Voice is the hardest part to get right, so start here.
Define Your Core Voice Attributes
Pick three to five adjectives that describe how your brand sounds — not how you want it to be perceived, but how your best writing actually reads. Be honest. “Professional” is nearly useless. “Direct and a little dry, like a senior engineer explaining something” is actionable.
For each attribute, write two examples: one that captures it, one that fails it.
Example format:
VOICE ATTRIBUTE: Direct
GOOD: "You'll need to add an API key before this works."
BAD: "In order to proceed with this process, you'll want to make sure you've taken the necessary steps to configure your API key settings."
VOICE ATTRIBUTE: Technically specific
GOOD: "The agent runs on a 15-minute cron schedule."
BAD: "The agent runs regularly in the background."This before/after structure is more useful than a paragraph of description because Claude can pattern-match against concrete examples.
Document What You Never Say
Brand voice is as much about exclusion as inclusion. Make a list of words, phrases, and structural patterns your brand avoids.
Common categories to cover:
- Banned words — jargon you hate, words that feel off-brand, competitor terminology
- Banned structures — never use passive voice, never open with a question, never use bullet points for emotional content
- Banned tones — don’t be sycophantic, don’t use exclamation marks, don’t write like a press release
If you’ve already written a banned words list for your team, paste it in. If not, spend five minutes listing the things that make you cringe when you see them in copy.
Set Formality and Reading Level
Be specific about register. “Conversational but professional” is vague. Try something like:
READING LEVEL: Aim for 8th–10th grade readability (Hemingway App score ~70)
SENTENCE LENGTH: Average 14–18 words. Mix short punchy sentences with slightly longer ones.
PARAGRAPH LENGTH: 2–4 sentences maximum in body copy. One sentence is fine.
PRONOUNS: Use "you" and "we." Avoid "one" and third-person constructions.Claude responds well to numerical constraints. “Short paragraphs” is vague. “2–4 sentences” is executable.
Add Real Examples
Pull five to ten pieces of your best existing content — an email, a landing page section, a product description, a social post. Paste them in verbatim under a section called CANONICAL EXAMPLES.
These examples do more work than any description. Claude can infer voice patterns from samples in ways it can’t from abstract guidelines.
Step 2: Translate Your Visual Identity into Design Tokens
This is where most people skip a step. They add voice guidelines but forget that Claude often needs to generate code, write design specs, or describe visual treatments — and without visual identity information, it defaults to generic choices.
Design tokens are the answer. They’re just named variables for your visual decisions.
Colors
Document your palette in hex codes, with labels that reflect usage rather than color names:
COLORS:
--color-primary: #1A1A2E
--color-secondary: #16213E
--color-accent: #0F3460
--color-highlight: #E94560
--color-background: #FAFAFA
--color-text-primary: #1A1A2E
--color-text-secondary: #6B7280
--color-success: #10B981
--color-warning: #F59E0B
--color-error: #EF4444If you have a dark mode palette, document that separately with a --dark prefix convention.
Also note usage rules:
COLOR USAGE RULES:
- Accent color is for CTAs only. Never use for decorative purposes.
- Never place text directly on the highlight color.
- Background must always be light in onboarding flows.Typography
Document your type system with enough specificity that Claude can generate CSS or write accurate design specs:
TYPOGRAPHY:
Font families:
--font-heading: 'Inter', sans-serif
--font-body: 'Inter', sans-serif
--font-mono: 'JetBrains Mono', monospace
Scale (desktop):
--text-xs: 12px / 1.5
--text-sm: 14px / 1.5
--text-base: 16px / 1.6
--text-lg: 18px / 1.5
--text-xl: 20px / 1.4
--text-2xl: 24px / 1.3
--text-3xl: 30px / 1.2
--text-4xl: 36px / 1.1
Weight usage:
Headings: 700 (bold)
Subheadings: 600 (semibold)
Body: 400 (regular)
Labels/captions: 500 (medium)Spacing, Radius, and Shadows
These matter more than most people think. Without them, Claude generates components that feel visually inconsistent with your actual product.
SPACING (base unit: 4px):
--space-1: 4px
--space-2: 8px
--space-3: 12px
--space-4: 16px
--space-6: 24px
--space-8: 32px
--space-12: 48px
--space-16: 64px
BORDER RADIUS:
--radius-sm: 4px
--radius-md: 8px
--radius-lg: 12px
--radius-full: 9999px
SHADOWS:
--shadow-sm: 0 1px 2px rgba(0,0,0,0.05)
--shadow-md: 0 4px 6px rgba(0,0,0,0.07)
--shadow-lg: 0 10px 15px rgba(0,0,0,0.10)Logo and Asset Usage Rules
Claude can’t see your logo, but it can follow rules about how to reference or describe it in copy and code:
LOGO USAGE:
- Primary logo: horizontal lockup (logo + wordmark)
- Icon-only version: use when width is < 120px
- Minimum clear space: equal to the height of the logo mark on all sides
- Never place on busy backgrounds
- Approved backgrounds: white, --color-primary, --color-background
- Never stretch, rotate, or recolorStep 3: Document Your Positioning and ICP
Positioning is where Claude tends to get generic fast. Without explicit guidance, it writes for a hypothetical average customer — not your actual buyer.
Define Your Ideal Customer Profile
Be specific. Not “B2B SaaS companies” but:
IDEAL CUSTOMER PROFILE:
PRIMARY PERSONA: "The Ops-Heavy Founder"
- Company size: 5–50 employees
- Industry: Professional services, agency, or SaaS
- Role: Founder or Head of Operations
- Core pain: Spending 15+ hours/week on tasks that should be automated
- Tech comfort: Uses tools like Notion, Airtable, Zapier. Not a developer.
- What they fear: Losing control of quality when they scale
- What they want: To stop being the bottleneck without hiring another person
- How they talk: Outcome-focused. Impatient with jargon. Skeptical of hype.Remy is new. The platform isn't.
Remy is the latest expression of years of platform work. Not a hastily wrapped LLM.
If you have multiple personas, document each one with the same structure. Then add a note about which persona is primary for which content type (e.g., “Landing page copy targets Persona A. Email sequences target Persona B”).
Write Your Positioning Statement
Use a structured format that Claude can reference directly:
POSITIONING:
For [ops-heavy founders at small service businesses]
Who struggle with [spending too much time on repetitive work that slows down their growth]
[MindStudio] is a [no-code AI agent builder]
That [lets non-technical people create automated workflows in under an hour]
Unlike [traditional automation tools like Zapier or Make]
We [connect AI reasoning to your actual business tools, so agents can handle complex multi-step tasks, not just simple triggers]This format forces clarity. If you can’t fill it in cleanly, your positioning isn’t ready yet — which is useful to know before you hand it to Claude.
Add Messaging Hierarchy
Not every message deserves equal prominence. Tell Claude which claims to lead with and which to save:
MESSAGING HIERARCHY:
Tier 1 (lead with these):
- Speed to build: most workflows take 15–60 minutes
- No code required: built for non-technical users
- Real AI reasoning: not just trigger-action rules
Tier 2 (support claims):
- 200+ AI models available
- 1,000+ integrations
- Used by teams at TikTok, Microsoft, Adobe
Tier 3 (mention when relevant):
- Free to start
- Supports custom JavaScript/Python for technical users
- Supports local models
CLAIMS TO AVOID:
- Never say "the best" or "the only"
- Never make direct comparisons by name unless asked
- Don't lead with pricingCompetitive Context
Give Claude enough context to write differentiated copy without being disparaging:
COMPETITIVE CONTEXT:
We often get compared to Zapier, Make, and n8n.
Key differences to acknowledge if asked:
- Those tools are trigger-action based. We're designed for multi-step AI reasoning.
- They require separate AI subscriptions and API setup. We have it built in.
- We're faster to build on for non-technical users.
Tone when addressing competitors: respectful, factual, not dismissive.Step 4: Assemble the Folder Structure
Now put it together. Here’s a clean folder layout:
/brand-context/
CLAUDE.md ← Master index file Claude reads first
voice-and-tone.md ← Voice attributes, rules, examples
design-tokens.md ← Colors, typography, spacing, shadows
positioning.md ← ICP, positioning statement, messaging hierarchy
copy-examples.md ← 5–10 canonical copy samples
competitive-context.md ← Competitive landscape and differentiationWrite a Master CLAUDE.md File
This is the file Claude picks up automatically in Claude Code. Use it as an index that tells Claude what’s in each file and when to reference it:
# Brand Context Index
This folder contains the brand guidelines for [Company Name].
Always reference these files before generating any content, copy, or UI.
## Files in this folder:
- `voice-and-tone.md` — How we write. Read this before generating any copy.
- `design-tokens.md` — Visual variables. Use these for all UI and design work.
- `positioning.md` — ICP, positioning, and messaging hierarchy. Reference when writing marketing content.
- `copy-examples.md` — Real examples of our brand voice in action.
- `competitive-context.md` — How to handle competitor comparisons.
## Quick reference:
Primary brand color: #1A1A2E
Primary font: Inter
Primary CTA text: "Start building free"
Brand voice in three words: Direct. Specific. Credible.
The quick reference section at the bottom is useful for tasks where Claude doesn’t need to read the full files — it gets the essentials without loading everything.
Step 5: Test, Refine, and Maintain
Building the folder is 80% of the work. The other 20% is iteration.
Run a Calibration Prompt
Before using the folder in real work, run a test. Ask Claude to generate a piece of content it hasn’t seen before — a product description, a short email, a UI microcopy string — and compare it to your canonical examples.
Ask yourself:
- Does the tone match?
- Are the visual decisions consistent with the design tokens?
- Does the messaging lead with the right tier?
- What feels off?
Then update the relevant file to address whatever was missing.
Add Edge Case Rules as You Find Them
Every time Claude does something that feels wrong, document the rule that would have prevented it. Over time, your folder becomes a knowledge base of every brand decision that’s ever come up.
This is especially useful for:
- Content types you haven’t covered yet (FAQs, error messages, notification copy)
- Edge cases in tone (how to write about pricing, how to handle bad news)
- Technical contexts (how to name variables, how to write code comments)
Keep It Current
Brand guidelines drift. When you update your messaging or refresh your design system, update the context folder at the same time. A stale brand context folder is worse than none — Claude will confidently produce outdated output.
Set a quarterly reminder to review each file. It takes 20 minutes if you’ve been maintaining it, and it’s worth it.
How MindStudio Extends This Further
A brand context folder solves the consistency problem for Claude Code sessions. But if you’re producing a high volume of content — social posts, email sequences, product descriptions, support responses — manually prompting Claude every time still creates friction.
This is where MindStudio fits in. You can build an AI agent that has your brand context baked in permanently, then trigger it automatically based on a schedule, an email, a form submission, or a webhook.
For example: you could build a content agent in MindStudio that reads your brand context files from a connected Notion database, accepts a topic input, and generates a first-draft blog post that already follows your voice guidelines — no manual prompting required. It runs on whatever schedule you set, or on demand through a simple web interface you share with your team.
The agent can call any of the 200+ AI models available in MindStudio, including Claude, and connects to 1,000+ tools through built-in integrations. So if you want it to pull your latest product changelog from Airtable, generate copy based on your brand context, and post a draft to your content calendar in Notion — that’s a single workflow you build once.
If you’re already doing serious content work with Claude Code, pairing it with a MindStudio agent layer means your brand context isn’t just a reference file — it becomes the foundation for fully automated, on-brand content production. You can start building for free at mindstudio.ai.
Common Mistakes to Avoid
Making It Too Abstract
Remy doesn't build the plumbing. It inherits it.
Other agents wire up auth, databases, models, and integrations from scratch every time you ask them to build something.
Remy ships with all of it from MindStudio — so every cycle goes into the app you actually want.
“We’re a warm, human brand” tells Claude almost nothing. Every attribute needs at least one concrete example. The more abstract your guidelines, the more Claude defaults to its own interpretation.
Skipping the Examples File
Voice guidelines without examples are like a recipe without the photo. Claude performs significantly better when it can see what “right” looks like. Five real examples are worth more than two pages of description.
Forgetting Non-Copy Contexts
Most brand context folders cover marketing copy and then stop. But Claude also generates code comments, UI strings, error messages, documentation, and emails. Each of these has its own microcopy conventions. Add a section covering at least the most common non-copy contexts your team encounters.
Not Telling Claude When to Read What
If your CLAUDE.md doesn’t tell Claude which file to read for which task, it may not read all of them for every task — especially in longer sessions. Be explicit: “For any UI work, always read design-tokens.md first. For any marketing copy, read voice-and-tone.md and positioning.md.”
Treating It as a One-Time Setup
The folder is a living document. If it doesn’t get updated when your brand evolves, it becomes a liability rather than an asset. Assign someone ownership of it and put it in version control.
Frequently Asked Questions
What is a brand context folder for Claude?
A brand context folder is a collection of structured Markdown files that contain your brand’s voice guidelines, design tokens, positioning statements, and copy examples. When placed in a Claude Code project, these files give Claude the reference material it needs to produce on-brand output without repeated prompting. It replaces the guesswork with explicit, reusable documentation.
Does Claude automatically read files in my project folder?
In Claude Code, a file named CLAUDE.md in your project root is picked up automatically as persistent context. Other files need to be referenced explicitly in your prompts or pulled in through your CLAUDE.md index. The most reliable approach is to use CLAUDE.md as an index that points Claude to the right file for each task type.
How long does it take to build a brand context folder?
For most small to mid-sized brands, about 30 minutes for a first version. The voice and tone document takes longest — typically 15–20 minutes if you write it from scratch. Design tokens go faster if you already have a design system or Figma file to reference. Positioning usually takes 10–15 minutes if you’ve thought through it before, longer if you haven’t.
Do I need a design system before I can document design tokens?
No. You can document whatever visual decisions you’ve already made, even informally. If you know your primary button color is #E94560 and your font is Inter, that’s enough to start. A partial token file is better than nothing — you can expand it over time as you make more visual decisions.
How is this different from a system prompt?
A system prompt is a single-session instruction set. A brand context folder is persistent documentation that lives in your project and can be read across many sessions and tasks. System prompts are good for task-specific instructions. Brand context folders are better for the brand baseline that should apply everywhere. In practice, you’d use both: the context folder for enduring brand knowledge, the system prompt for task-specific parameters.
Can I use the same brand context folder for different AI models?
Yes, with minor adjustments. The files are plain text and work with any model that accepts file context. Claude Code, GPT-4 with file context, and others can all benefit from the same documentation. You may find that some models need slightly more explicit instructions than others, but the core files transfer well. The design tokens and positioning documents especially translate cleanly across models.
Key Takeaways
- A brand context folder is a set of structured reference files that gives Claude your voice, visual identity, and positioning so it can produce consistent output without being re-briefed every session.
- The four core files are: voice and tone, design tokens, positioning, and copy examples. A CLAUDE.md index ties them together.
- Voice documents work best when they include concrete before/after examples, not just abstract descriptions.
- Design tokens translate your visual decisions into named variables Claude can reference when generating code, UI specs, or design descriptions.
- Positioning documents should include a structured ICP, a positioning statement, and a messaging hierarchy that tells Claude what to lead with.
- The folder needs to be maintained — update it when your brand evolves, and add edge case rules as you encounter them.
- For high-volume content production, pairing your brand context with a MindStudio agent lets you automate on-brand content generation at scale, rather than manually prompting Claude each time.
