What is Claude, really?
Before you can use Claude well, you need to understand what it actually is — not the hype, the reality.
Large Language Model
Claude predicts the best next word/sentence based on everything it was trained on — billions of examples of human writing.
Context is Memory
Claude only "remembers" what's inside your current conversation window. New chat = blank slate unless you give it context.
Output = Input Quality
The richness of Claude's response is directly proportional to how clearly you frame your request.
Multiple Access Points
Claude.ai (chat), API (code/apps), Claude Code (terminal), Projects (organised workspaces).
How to access Claude
There are four main ways to work with Claude. You've already created agents — here's how each entry point works and when to use it.
🌐 Claude.ai — The Chat Interface
The simplest access point. Go to claude.ai, log in, and start chatting. No code required. Best for: writing, research, brainstorming, one-off tasks. This is what you're using right now.
📁 Projects — Organised Workspaces
Projects let you give Claude persistent instructions and upload documents. Think of a Project as a "briefed employee" — it knows your context, your tone, your rules. Available on Claude.ai Pro and Team plans.
⚙️ API — Claude in Your Apps
The API lets you embed Claude into your own software — WordPress plugins, websites, automations. This is how JoeBot works. You send a message, Claude responds, your app uses the response.
💻 Claude Code — Terminal Power
A command-line tool for developers. Runs Claude directly in your terminal to write, edit, and debug code files autonomously. Advanced use — not needed until you're building complex systems.
Go to Claude.ai and click "New Project"
In the left sidebar on claude.ai, look for "Projects" and click it. Then hit "Create Project".
Write a Project System Prompt
This is the "briefing" Claude reads before every message. Tell it who you are, what it's helping with, the tone, the rules. Example below ↓
Upload relevant documents
Drag in PDFs, text files, or paste content. Claude will use this as a knowledge base throughout the project.
Start chatting inside the Project
Every conversation within the Project automatically has access to your system prompt and uploaded documents. Claude stays "briefed".
Draft your own Project System Prompt
A great system prompt answers: Who am I? What are you helping me with? What tone/style? What should you never do?
Prompting that actually works
This is where most people fail. Prompting is a skill — and there's a clear framework that separates average results from expert-level output.
Context — Set the scene
Who you are, what project this is for, why you need this. "I'm building an email sequence for Groundhogg CRM targeting EdtrepreneurIgnite users who chose the Founder role."
Role — Give Claude a persona
Tell Claude who to be. "Act as an expert email copywriter who specialises in education-tech funnels." This shifts the entire style and depth of response.
Action — Be specific about the task
Not "write an email" but "write a 200-word onboarding email, subject line included, with a single CTA to visit the Tools page."
Format — Specify the output
Do you want bullet points? A table? HTML code? A numbered list? Markdown? Tell Claude exactly how you want it structured.
Tone — Lock the voice
"Write in a confident, mentor-like tone. No filler. No AI-speak. Direct and warm." This prevents generic corporate language.
❌ Weak Prompt
"Write me an email about my course."
Claude will guess everything — the audience, the goal, the tone, the length. Result: generic, flat, probably unusable.
✓ Strong Prompt
"Act as an expert email copywriter for education-tech. I'm Zeph Ginsberg, founder of EdtrepreneurIgnite — an identity-first entrepreneurship program for Grades R–12. Write Email 1 of a 5-email welcome sequence for a new subscriber who selected the 'Founder' role in JoeBot. The email should: welcome them, confirm what EdtrepreneurIgnite is about, introduce the Starter Map quest system, and end with a single CTA to visit edtrepreneurignite.com/starter-map. Tone: warm, direct, mentor energy. Length: 180–220 words. Format: subject line + body."
Build a CRAFT prompt for your next task
Think of something you need from Claude right now. Apply CRAFT — Context, Role, Action, Format, Tone.
Agents & Projects — Claude with memory
You've already built agents. Now understand exactly how they work, how to access them, and how to make them dramatically more effective.
System Prompt
The "job description" Claude reads before every conversation. This is where you define identity, rules, tone, and goals.
Tools
Web search, code execution, document reading. Tools let Claude take action, not just generate text.
Memory / Context
Projects give Claude persistent document access. For app-based agents, you pass memory in the API call.
Via Projects (most common)
Click "Projects" in the left sidebar → select your project → start a new conversation. Claude is automatically briefed by your system prompt.
Via API calls (for embedded agents like JoeBot)
Your agent lives in code — the system prompt is passed with every API call. To test it: open your WordPress plugin's PHP file and look for the system parameter in the API request body.
Via Claude.ai with a custom system prompt (beta)
On some plans, you can set a default system prompt in Settings → Custom Instructions. This applies to all new chats, effectively making Claude.ai itself your agent.
🤖 How JoeBot v2 works as an agent
When a user interacts with JoeBot on your WordPress site, the plugin sends an API call to Claude. That call includes: (1) the system prompt with JoeBot's identity and rules, (2) the user's role selection as context, (3) the user's message. Claude responds, JoeBot formats and displays it. The "agent" is your system prompt — it's the brain behind every response.
Write a system prompt for a specialist agent
Pick one of your use cases: curriculum assistant, CRM helper, funder proposal writer, or affiliate tools guide. Write its system prompt below.
API & Integrations
This is how JoeBot, Groundhogg integrations, and any future automated Claude tool work. You don't need to code it all yourself — but you do need to understand the structure.
📡 What is the API?
API stands for Application Programming Interface. It's a way for your code to talk to Claude. You send a structured message (JSON), Claude processes it and sends back a structured response. Your app reads that response and does something with it.
API Key
Your secret credential from console.anthropic.com. Never paste it in public code. In WordPress, store it in wp-config.php or an environment variable.
System vs User
system = Claude's briefing (your agent prompt). user = what the human typed. Keep these separate and clear.
max_tokens
The maximum length of Claude's response. 1024 ≈ 750 words. Set it based on how long you expect responses to be.
Conversation History
To give Claude memory of the chat, pass all previous messages in the messages array — both user and assistant turns.
Go to console.anthropic.com
Log in with your Anthropic account (same email as claude.ai).
Click "API Keys" in the left menu
Then click "Create Key". Give it a name (e.g. "JoeBot Production"). Copy the key immediately — you won't see it again.
Store it securely in WordPress
In wp-config.php add: define('ANTHROPIC_API_KEY', 'sk-ant-...'); Then in PHP reference it as ANTHROPIC_API_KEY.
Set a usage limit (important!)
In the Console under "Billing → Usage Limits", set a monthly cap so unexpected traffic doesn't create unexpected costs.
Your Power Setup
Put it all together. This is how to configure Claude for maximum output across all your EdtrepreneurIgnite and JSG work.
📁 Project 1: EdtrepreneurIgnite Core
System prompt: EdtrepreneurIgnite brand, visual identity specs, curriculum philosophy, Groundhogg CRM rules. Upload: your brand guide, curriculum framework, quest system structure. Use for: all web content, quest blocks, email sequences, student materials.
📁 Project 2: JSG Business & Funding
System prompt: JSG brand identity, Zeph as mentor/influencer, proposal writing style, funder tone. Upload: existing proposals, business plan, pitch deck content. Use for: EDHE proposals, funder outreach, business development writing.
📁 Project 3: Tech & Systems
System prompt: WordPress + Elementor stack, Groundhogg CRM, JoeBot plugin context, PHP/JS coding style. Upload: JoeBot plugin code, Groundhogg form IDs, site structure notes. Use for: plugin development, API integrations, CRM automation setup.
Open the right Project first
Don't use the general chat for specialist work. Open the project that matches what you're building — Claude is already briefed.
Lead with context, not just a request
Even inside a Project, start with: "I'm working on [specific thing]. Here's where we are: [3 lines]. I need: [specific output]."
Iterate, don't restart
When a response isn't right, correct it in the same thread. "Good structure, but make the CTA more urgent and cut the last paragraph." Claude improves in-context.
Save your best prompts
Keep a personal prompt library (a simple Google Doc works). When a prompt produces great output, save it. This becomes your productivity asset.
Chain Prompting
Break complex tasks into steps. "First, give me an outline. Then I'll ask you to write each section." Better output, less hallucination.
Ask Claude to Critique Claude
"Review what you just wrote. What are the 3 weaknesses? Now rewrite it fixing those." Self-review loops dramatically improve quality.
Persona Lock
Start each session: "For this conversation, you are [X]. Maintain this perspective throughout, even if I ask you general questions."
Numbered Instructions
For complex requests, number your requirements: "1. Tone: warm mentor. 2. Length: 200 words. 3. Include: one metaphor. 4. End with: a question." Claude follows lists reliably.
Design your 3-Project structure
Based on your actual work, write the system prompt for your most important Project. Include: your identity, what Claude is helping with, brand voice, 3 rules Claude must follow, and 1 thing it must never do.
