How to Use Claude Projects: Build a Brand-Specific AI Assistant with Custom Instructions and Knowledge Files
How to Use Claude Projects: Build a Brand-Specific AI Assistant with Custom Instructions and Knowledge Files
Claude Projects is one of the most powerful features available to Claude Pro, Team, and Enterprise users. It lets you create persistent workspaces where you define custom instructions and upload knowledge files, effectively transforming Claude into a specialized AI assistant tailored to your brand, workflow, or domain. This guide walks you through the complete setup process with real examples.
What Are Claude Projects?
A Claude Project is a dedicated workspace within Claude that maintains context across conversations. Unlike standard chats that start from zero every time, Projects allow you to:
- Set persistent custom instructions that shape every response- Upload knowledge files (up to 200K tokens of context) that Claude references automatically- Share the project across team members (on Team and Enterprise plans)- Create multiple specialized assistants without repeating setup
Step-by-Step: Creating Your Brand AI Assistant
Step 1: Create a New Project
- Open claude.ai and log in with your Pro or Team account.- In the left sidebar, click Projects.- Click Create Project and give it a descriptive name, e.g.,
Acme Brand Assistant.- Add an optional description: “AI assistant for Acme Corp content, support, and internal communications.”
Step 2: Write Custom Instructions
Click the Set custom instructions field inside your project. This is where you define Claude’s persona, tone, rules, and behavioral constraints. Here is an example instruction set:
You are the official AI assistant for Acme Corp.
Brand Voice
- Tone: Professional yet approachable. Never use slang.
- Always refer to the company as “Acme Corp” (never “Acme” alone).
- Use active voice. Keep sentences under 25 words when possible.
Response Rules
- Always include a call-to-action when answering customer-facing queries.
- For pricing questions, direct users to acmecorp.com/pricing.
- Never disclose internal roadmap details or unannounced features.
- When unsure, say: “Let me connect you with our team for the most accurate answer.”
Formatting
- Use headers (H2, H3) for long responses.
- Use bullet points for lists of 3+ items.
- Include relevant links from the knowledge base when available.
Audience
- Primary: Small business owners (non-technical)
Secondary: Enterprise procurement teams
Step 3: Upload Knowledge Files
Click **Add content** in the project knowledge section. You can upload multiple file types:
| File Type | Best For | Max Size |
|---|---|---|
| Brand guidelines, whitepapers, product manuals | ~30MB per file | |
| TXT / MD | FAQs, style guides, structured knowledge bases | ~30MB per file |
| CSV | Product catalogs, pricing tables, feature matrices | ~30MB per file |
| Code files (.py, .js, etc.) | API documentation, code standards | ~30MB per file |
brand-guidelines.pdf — Logo usage, color palette, typography, voice and tone- product-faq.md — Frequently asked questions and approved answers- pricing-sheet.csv — Current pricing tiers, features per plan- support-scripts.txt — Common support scenarios and resolution steps- competitor-comparison.md — Positioning against competitors (internal use)
### Step 4: Test and Iterate
Start a new conversation within the project and test with real-world queries:
You: "What makes Acme Corp different from CompetitorX?"
You: "Write a follow-up email for a customer who asked about enterprise pricing."
You: "Summarize our refund policy in 3 bullet points."
Review the outputs against your brand standards. If Claude deviates, refine your custom instructions. Common adjustments include:
- Adding explicit **"Do not"** rules for edge cases- Providing example outputs in the instructions (few-shot prompting)- Restructuring knowledge files for better retrieval
### Step 5: Integrate via the API (Optional)
For developers who want to use Projects programmatically, you can reference project context through the Claude API:
curl https://api.anthropic.com/v1/messages \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-d '{
"model": "claude-sonnet-4-20250514",
"max_tokens": 1024,
"system": "You are the Acme Corp brand assistant. Follow all brand guidelines strictly.",
"messages": [
{
"role": "user",
"content": "Draft a product announcement for our new analytics dashboard."
}
]
}'
For persistent context equivalent to Projects via API, use the system prompt combined with document content passed in the message context.
Pro Tips for Power Users
- Layer your instructions: Put universal rules (tone, formatting) at the top and domain-specific rules (product details, compliance) lower. Claude weighs earlier instructions more heavily in ambiguous situations.- Use structured knowledge files: Markdown with clear headers outperforms unstructured text. Use
## Section Nameheadings in your .md files so Claude can locate information faster.- Create multiple projects per function: Instead of one mega-project, build separate projects for Sales, Support, Content, and Engineering. This keeps context focused and reduces noise.- Version your instructions: Keep a changelog at the bottom of your custom instructions. Example:v2.3 — 2026-03-15: Added compliance language for EU customers.- Test adversarially: Ask Claude to write content that would violate your guidelines and verify it refuses or self-corrects. This reveals gaps in your instruction set.
Troubleshooting Common Issues
| Problem | Cause | Solution |
|---|---|---|
| Claude ignores uploaded knowledge files | File content is too unstructured or the query doesn't trigger retrieval | Add explicit headers in your files. In instructions, add: "Always check the uploaded knowledge base before answering product questions." |
| Responses don't match brand tone | Instructions are too vague | Add 2-3 example responses directly in your custom instructions as few-shot examples |
| Claude contradicts pricing or feature info | Outdated knowledge files or conflicting information across files | Audit all uploaded files for consistency. Remove outdated versions and re-upload current ones |
| Context window exceeded warning | Too many or too large knowledge files uploaded | Consolidate files. Remove redundant content. Prioritize the most-referenced documents |
| Team members see different behavior | Individual chats outside the project | Ensure all team members start conversations from within the shared project, not from the main chat |
How many knowledge files can I upload to a single Claude Project?
You can upload multiple files to a Claude Project, with a combined limit of approximately 200,000 tokens of context (roughly 500 pages of text). The practical limit depends on file sizes, but most brand assistants work well with 5-10 focused documents rather than dozens of large files. Prioritize quality and relevance over quantity.
Can I share a Claude Project with my team, and will they see the same custom instructions?
Yes. On Claude Team and Enterprise plans, Projects can be shared with team members. All participants see and use the same custom instructions and knowledge files, ensuring consistent AI behavior across your organization. Individual team members cannot override the project-level instructions, which maintains brand consistency.
What is the difference between custom instructions in Projects and the system prompt in the API?
Claude Projects custom instructions are set through the claude.ai web interface and persist across all conversations within that project. The API system prompt serves a similar purpose but is set programmatically per request. Projects are ideal for non-technical users who need a persistent, no-code setup. The API system prompt offers more flexibility for developers building custom applications, but requires you to manage context and document uploads manually in each request.