How I Use Claude Projects to Build Entire Training Courses

The Problem With Using AI for Course Design

If you've tried using ChatGPT or Claude to help build a training course, you've probably hit the same wall I did. You paste in your content, get a decent outline, then start a new chat and have to re-explain everything from scratch. The AI forgets your SME notes, your brand voice, your learner audience. Every conversation starts at zero.

That changed when I started using Claude Projects. It's not a new model or a plugin. It's a feature built into Claude Pro ($20/month) that lets you create a persistent workspace where Claude remembers everything you upload and follows custom instructions across every conversation in that project.

What Claude Projects Actually Does

A Claude Project is a container with three things: uploaded files (your knowledge base), custom instructions (your rules), and conversations (your chats). Here's what makes it different from just pasting stuff into a chat window.

Uploaded files stick around. You can upload PDFs, Word docs, spreadsheets, HTML files, even images. Claude can reference all of them in any conversation within that project. The limit is 200,000 tokens of combined text, which is roughly 150,000 words, or about 300 pages. For most training projects, that's more than enough to hold your SME interviews, policy docs, existing course materials, and style guides.

Custom instructions shape every response. You write a set of instructions that Claude follows in every conversation. This is where the magic happens for instructional design. I tell Claude things like: "You are helping me build a 4-module compliance training course for healthcare workers. All content should be written at an 8th-grade reading level. Use scenario-based examples set in a hospital environment. Follow Bloom's taxonomy for learning objectives."

Multiple conversations, one brain. You can have separate chats for Module 1 outline, Module 2 quiz questions, assessment rubric. Claude has access to all your files and follows your instructions in each one.

My Actual Workflow: Policy Doc to Course Outline

Here's the exact process I used last month to turn a 47-page workplace safety policy into a 4-module eLearning course. Total time from start to finish: about 55 minutes.

Step 1: Create the project and upload your source docs. I created a new project called "Workplace Safety 2026" and uploaded three files: the full policy PDF, an incident report spreadsheet from the past year (anonymized), and our company style guide. Took about 2 minutes.

Step 2: Write your custom instructions. This is the most important step. I wrote instructions that covered the target audience (warehouse workers, mixed reading levels), the tone (direct and practical, not corporate), the format I wanted (scenario-based with decision points), and specific constraints (each module should be completable in 15 minutes or less). About 5 minutes to write.

Step 3: Extract the key learning points. First conversation: "Based on the policy document, identify the 10 most critical safety rules that employees need to follow daily. Cross-reference with the incident report data to find which rules are most frequently violated?" Claude pulled out the rules AND ranked them by how often they actually caused incidents. That's something I couldn't have done manually without hours of cross-referencing.

Step 4: Generate the course structure. Second conversation: "Create a 4-module course outline. Each module should cover 2-3 related safety rules, start with a realistic scenario, include 2 knowledge checks, and end with a practical application exercise." Claude gave me a detailed outline with scenario setups, question stems, and even suggested which types of interactions would work best for each topic.

Step 5: Build out the scenarios. Third conversation: "Write the opening scenario for Module 1. The learner is a new warehouse employee on their first solo shift. They encounter three safety decisions in the first hour." Because Claude had the incident data, it based the scenario on things that actually happen, not made-up situations.

Three Things That Make This Better Than Regular Chat

1. Consistency. Every conversation in the project follows the same rules. When I ask Claude to write quiz questions in conversation 4, it automatically uses the same reading level, tone, and scenario format I specified in my instructions. No re-prompting.

2. Source accuracy. Claude pulls directly from the documents I uploaded. When it writes a scenario about chemical handling, it references the actual procedures from the policy doc, not generic internet knowledge. You can ask it to cite which document a piece of information came from.

3. Iteration without repetition. When my SME reviewed Module 2 and said the forklift safety section needed more detail, I didn't have to re-upload anything. I just opened a new conversation: "The SME says Module 2 needs more detail on forklift pre-operation inspections. Expand that section using pages 23-25 of the policy doc." Claude knew exactly what I meant.

The Limitations You Should Know About

It's not perfect. The 200K token context means very large projects with dozens of source documents might not all fit. In that case, I create separate projects for different modules and keep the most relevant docs in each one.

Claude can also occasionally "hallucinate" details that aren't in your source docs, especially for procedural steps. I always fact-check anything that sounds too specific against the original documents. The good news is that having the source docs in the project makes this easier because you can ask Claude to quote exactly where it got a piece of information.

File types matter too. Claude handles PDFs, Word docs, text files, HTML, JSON, and CSV well. It can read images but won't extract text from scanned PDFs reliably, so use OCR first if your source docs are scanned.

Try This Today

What's Next

I'm working on a deeper walkthrough of how I use Projects specifically for building branching scenarios, the kind where one decision leads to different consequences. That workflow is a little more complex but Projects makes it genuinely manageable. Stay tuned for that one on Wednesday.