Claude Projects: Build a Searchable Training Knowledge Base

What This Builds

A Claude Project is a persistent AI workspace where you upload your training library (existing courses, policy documents, style guides, past facilitator guides) and then use AI to search, reference, and build from that accumulated knowledge in every conversation. Instead of re-uploading files or re-explaining context each time, your Project holds everything. Every new course you build draws from your existing institutional knowledge automatically.

Think of it as giving Claude a permanent desk in your office, stocked with every file you've ever built, so it can help you without starting from zero every time.

Prerequisites

• Claude Pro account at claude.ai ($20/month, Projects require Pro tier)
• Existing training files to upload (PDFs, Word docs, past course outlines, facilitator guides)
• Clear idea of what work you want this Project to help with
• Comfortable using Claude for basic content generation (Level 3)

The Concept

Without Projects, every Claude conversation is like meeting a new contractor each time: you explain the same context over and over. With a Claude Project, you have a persistent collaboration space. The Project remembers:

• Your organization's context and audience
• Your training standards and formats
• All the files you've uploaded to it
• Any custom instructions you set

Every conversation within the Project starts from that foundation. You can also share the Project with colleagues, so your whole team works from the same knowledge base.

Build It Step by Step

Part 1: Plan what goes in your Project

A well-organized Project produces better results than one where you dump everything. Plan 3 categories of content:

1. Context documents (required):

• Your training design standards or style guide
• A sample facilitator guide in your preferred format
• A sample quiz question set
• Your organization overview (1-page description of your company, industry, and audience; no sensitive data)

2. Source content (the knowledge base):

• Existing course outlines or storyboards you've already built
• Policy documents that are frequently the basis for compliance training
• HR or People Operations documents (handbook sections, onboarding materials)
• Any SME-provided content libraries you regularly reference

3. Reference tools:

• Bloom's Taxonomy action verb list (or ask Claude to generate and store one for you)
• Kirkpatrick evaluation framework summary
• ADDIE or SAM process checklist

Keep source content focused. A lean, well-labeled knowledge base produces better results than a chaotic one with hundreds of files.

Part 2: Create the Project

Go to claude.ai → sign in → click the "Projects" option in the left sidebar (or look for a "New Project" button).

Click "New Project." Give it a name that's clear and bookmarkable:

• "L&D Design Hub — [Your Company Name]"
• "Training Content Library — [Your Team Name]"

Part 3: Write Project instructions

Before uploading files, set your Project instructions. Click "Set project instructions" or similar (the interface may show this as "Edit instructions" or a settings panel).

These instructions appear at the start of every conversation in this Project. Write them like your Custom GPT system prompt:

This Project is a corporate training design workspace for [YOUR NAME], a Training & Development Specialist at [COMPANY/ORGANIZATION TYPE].

AUDIENCE: [Describe your primary learner population]
FRAMEWORK: [ADDIE/SAM/backward design]
AUTHORING TOOL: [Articulate 360/Rise/etc.]
LMS: [Your LMS platform]

STANDARDS:
- Learning objectives: Bloom's action verbs, format "By the end of this [module], learners will be able to..."
- Quiz questions: 4-option multiple choice, answer key included, scenario-based where possible
- Facilitator guides: Timed column format (left: time; right: content)
- Writing tone: [Professional and direct / Conversational]

KNOWLEDGE BASE CONTENTS:
- Training style guide: [Filename if uploaded]
- Sample formats: [Filenames]
- Policy library: [List key documents uploaded]

When building new content, reference the uploaded materials for consistency with existing formats and company-specific language. Flag when content requires SME or legal review.

Part 4: Upload your knowledge base documents

In your Project's sidebar or file area, click "Add content" or the upload button. Upload your documents one by one.

Naming convention matters: Claude searches and references files by name during conversations. Use clear, descriptive names:

• style-guide-2026.pdf
• sample-facilitator-guide-leadership.docx
• onboarding-course-outline-v3.pdf
• hipaa-compliance-policy.pdf

Upload limit: Claude Pro supports multiple documents per Project. Keep individual files under 100,000 words for best performance.

After each upload: Run a quick test. Ask Claude "What format do I use for facilitator guides according to my style guide?" If it references the correct file and format, the document loaded correctly.

Part 5: Test your Project with real work tasks

Run these test prompts to confirm your Project is working as intended:

Test 1 — Format consistency: "Write a facilitator guide for a 60-minute customer service skills workshop. Use the format from my uploaded sample guide."

Verify: Does the output match your sample format?

Test 2 — Knowledge base reference: "I need to add a compliance module on [TOPIC FROM YOUR UPLOADED POLICY]. What does our policy say I should cover, and how should I structure it?"

Verify: Does Claude reference the uploaded policy document accurately?

Test 3 — Context continuity: Open a new conversation within the same Project (not a new chat; click "New conversation" inside the Project). Ask: "What's my standard format for learning objectives?"

Verify: Does Claude know your format without you re-explaining it?

Fix issues: If tests fail, refine your Project instructions to be more explicit, or re-upload files with clearer naming.

Part 6: Build your first course from the knowledge base

Now put it to work on a real project.

Workflow for a new course request:

1. Open your Project → New conversation
2. Describe the new course: "I need to build a [DURATION] course on [TOPIC] for [AUDIENCE]. This is similar to our existing [EXISTING COURSE NAME] (one you uploaded)."
3. Ask: "Using our standard course structure, draft the learning objectives and a 4-module outline."
4. The output should reflect your formats AND reference your existing courses as models.
5. For each module, continue in the same conversation: "Now expand Module 2 with content, activities, and assessment questions in our standard format."
6. The entire session accumulates context. Each response builds on the previous one.

Real Example: Using Your Project for a Monthly Content Review

Background: You have a Project containing 8 existing course outlines, your style guide, and 3 policy documents.

Monthly request: HR sends a new harassment prevention policy update and asks you to refresh the training.

Session in your Project:

Prompt 1: "HR has updated our harassment prevention policy. Here are the key changes: [paste 200 words of changes]. How does this affect our existing Harassment Prevention course?"

Claude cross-references your uploaded course outline and the policy changes. It identifies which modules need updating and which are unaffected.

Prompt 2: "Update Module 3 (Reporting Procedures) to reflect these policy changes. Keep the same format as the current module."

Claude rewrites the module matching your existing format, incorporating the new policy language, and flagging two points that need legal review.

Prompt 3: "Write 3 new quiz questions for the updated Module 3 content."

You now have a complete course refresh in one 20-minute session instead of a full day of document review and rewriting.

Setup time: 60 minutes (one time) Per-session input: Policy change summary Output: Targeted course update with format consistency and flagged review items Time saved per refresh: Full-day task → 20-minute session

What to Do When It Breaks

• Claude doesn't reference uploaded files → Use explicit file-reference language: "Based on the [filename] I uploaded, write..." Direct references work more reliably than hoping the AI searches automatically.
• Outputs don't match your format → Add a concrete example of your exact format into your Project instructions. Copy-paste one complete quiz question or facilitator guide section as a format example.
• Context bleeds between conversations → Each conversation within a Project starts fresh (Project instructions carry over, but conversation history doesn't). If you need to continue a complex build, stay in the same conversation thread rather than starting a new one.
• File contents seem to be ignored → Check that your files are not scanned PDFs (image-only). Claude needs text-readable PDFs. If a PDF is scanned, convert it to Word first using Adobe Acrobat or a PDF converter tool.

Variations

• Simpler version: Don't upload files at the start. Just write detailed Project instructions. You get context persistence without a full knowledge base: easier to set up, less reference capability.
• Extended version: Create separate Projects for different content domains (Compliance Library, Manager Development, Technical Skills). Each Project has its own specialized knowledge base and instructions.

What to Do Next

• This week: Create your Project, write instructions, and upload your 3–5 most important reference documents. Run the 3 test prompts.
• This month: Use the Project for your next 3 course builds and note which parts of the knowledge base actually get referenced. Prune what isn't used.
• Advanced: Share the Project link with your L&D team members (they need their own Claude Pro accounts). Establish team norms for what gets added to the shared knowledge base and how files are named for searchability.

Advanced guide for corporate trainer professionals. Claude Pro subscription required ($20/month). Projects are a Pro-tier feature.