Building with Claude Projects (Hands-On)
Now you will build the same chatbot using Claude Projects. This gives you hands-on experience with a second platform and helps you understand the differences so you can choose the best tool for each use case.
Prerequisites
- A Claude Pro ($20/month), Team, or Enterprise subscription
- Your completed planning worksheet
- A desktop browser
What Are Claude Projects?
Claude Projects are persistent workspaces within Claude. Each project has its own:
- Custom instructions: Rules that apply to every conversation in the project
- Knowledge files: Documents Claude can reference when answering
- Conversation history: Past conversations are saved within the project
Unlike Custom GPTs, Claude Projects are designed as personal or team workspaces rather than distributable chatbots. They are ideal when you need a focused AI assistant for ongoing work rather than a public-facing tool.
Step 1: Create a New Project
- Go to claude.ai
- Click Projects in the left sidebar
- Click New Project
- Enter a name and description for your project
Example:
- Name:
HR Onboarding Assistant - Description:
Answers new employee questions about company policies, benefits, schedules, and office logistics.
Step 2: Write the Project Instructions
Click on your new project, then click the Settings icon (gear icon) or Edit Project to access the project instructions.
Claude Projects use a single instructions field. Paste your planning worksheet and format it clearly. Claude responds well to structured, natural-language instructions.
Step 3: Key Differences from Custom GPTs
As you set up your Claude Project, note these important differences:
| Feature | Custom GPTs | Claude Projects |
|---|---|---|
| Instructions location | Configure > Instructions | Project Settings > Instructions |
| Conversation starters | Up to 4 clickable prompts | Not available (users type freely) |
| Capabilities toggles | Web Browsing, DALL-E, Code Interpreter | No toggles needed (Claude's capabilities are always available within its scope) |
| Profile picture | Upload or generate | Not applicable |
| Knowledge limit | Up to 20 files | Up to 200K tokens of context |
| Sharing | Public, link, or private | Team members only (Pro is personal) |
Step 4: Test Your Project
Start a new conversation within the project. Run the same tests you did with your Custom GPT:
Test 1: Core Task
Ask a question your chatbot should handle well.
"What health insurance options are available for new employees?"
Claude should reference the uploaded documents (we will add those in the next lesson) and respond in your specified tone.
Test 2: Boundary Test
Ask something outside scope.
"Can you help me write a resignation letter?"
Claude should politely redirect based on your instructions.
Test 3: Follow-Up Depth
Ask a follow-up question to test conversation continuity.
"Tell me about the first-week schedule." Then: "What happens on day three specifically?"
Claude should maintain context and provide increasingly specific answers.
Test 4: Tone Check
Read the responses out loud. Do they match the tone you specified?
Step 5: Refine Your Instructions
Based on testing, adjust your instructions. Claude is particularly responsive to:
Adding examples of desired responses:
When a user asks about benefits, respond like this:
"Acme Corp offers three health insurance tiers:
- Basic: $50/month, covers essential care
- Standard: $120/month, adds specialist visits
- Premium: $200/month, full coverage including dental and vision
Would you like details on any specific tier?"
Clarifying ambiguous situations:
If a user asks about something not covered in the documents, respond:
"I don't have that information in my onboarding materials. I'd recommend reaching out to HR directly at hr@acmecorp.com or checking the company intranet."
Strengthening boundaries:
Important: Never attempt to answer questions about salary negotiations, legal disputes, or disciplinary procedures. Always redirect these to HR.
When to Choose Claude Projects Over Custom GPTs
Claude Projects are the better choice when:
- You work with long documents: Claude's 200K token context window handles large manuals, legal documents, or technical specs better than GPTs' 20-file limit
- You need deep analysis: Claude excels at nuanced reasoning and detailed explanations
- You want a personal workspace: Projects work great as an ongoing tool for yourself or your team
- Privacy matters: Conversations stay within your project, not shared publicly
- You need careful, considered responses: Claude tends to be more thorough and cautious
Custom GPTs are better when:
- You want public distribution: The GPT Store lets anyone discover and use your chatbot
- You need web browsing: Custom GPTs can search the internet in real time
- You need image generation: DALL-E integration is built in
- You want conversation starters: Clickable prompts reduce friction for new users
Comparing Results
If you built on both platforms, compare the responses side by side:
| Criterion | Your Custom GPT | Your Claude Project |
|---|---|---|
| Response quality | ||
| Tone accuracy | ||
| Boundary handling | ||
| Knowledge use | ||
| Response length | ||
| Follow-up handling |
This comparison will help you decide which platform to use for your final chatbot.
Key Takeaway
Claude Projects offer a focused, workspace-style approach to building chatbots. They shine with long documents and complex reasoning but lack public distribution. Having experience with both platforms means you can choose the right tool for each chatbot you build. In the next lesson, you will add knowledge bases and files to make your chatbot truly useful.