Platform Mastery: ChatGPT, Claude, Gemini

🔄 In Lesson 4, you learned advanced techniques: XML tags, conditional logic, few-shot examples. Those techniques are universal — but every platform implements custom instructions differently. Let’s go platform by platform.

ChatGPT: Three Layers of Instructions

ChatGPT gives you three levels of customization, each more powerful:

Level 1: Custom Instructions (Always-On)

Found in Settings → Personalization. Two fields, 1,500 characters each. These apply to every new conversation.

Best for: General preferences that never change — your role, communication style, output format defaults.

Limitation: 3,000 total characters means you can’t fit everything. Use this for your “80% case” — the preferences that apply to most conversations.

Level 2: Custom GPTs (Task-Specific)

8,000 characters for instructions. Plus: upload knowledge files, add conversation starters, and share with others.

Best for: Specific recurring tasks. A “Code Reviewer” GPT, a “Meeting Summarizer” GPT, a “Marketing Copywriter” GPT.

How to create: GPT Editor → Configure → Instructions. You can also use the GPT Builder conversation to create one interactively.

Pro tip: Your Custom Instructions still apply when using a Custom GPT. They stack — the GPT’s instructions layer on top of your global instructions.

Level 3: Projects (Context-Specific)

Per-project instructions plus file attachments. Everything in a Project shares the same context.

Best for: Ongoing work with specific context — a product launch, a research paper, a client engagement.

✅ Quick Check: You want the AI to always respond in bullet points with American English. You also want a specialized code reviewer setup. Where does each instruction go? (Bullet points and American English go in Custom Instructions — they apply everywhere. The code reviewer goes in a Custom GPT with specific coding rules, checklists, and examples.)

Claude: Projects + XML Power

Claude’s approach is different from ChatGPT’s — less about global settings, more about project-based workflows.

Projects

Claude Projects give you:

• Custom instructions — as long as you need (no 1,500-char limit)
• Attached documents — PDFs, code files, style guides that the AI can reference
• Conversation history — within the project context

Best for: Any workflow that needs reference material. Writing with a style guide attached. Coding with your codebase context. Analysis with your dataset.

XML Tag Optimization

Anthropic’s official recommendation: use XML to structure instructions. Here’s the Claude-optimized format:

<assistant_role>
You are a technical editor specializing in developer documentation.
</assistant_role>

<style_guide>
- Active voice: "The function returns" not "A value is returned"
- Present tense for current behavior
- Second person: "you" not "the user"
- No marketing language in technical docs
</style_guide>

<review_process>
1. Check technical accuracy of code examples
2. Flag jargon that needs definition
3. Identify missing prerequisites
4. Suggest structural improvements
</review_process>

The tags are parsed cleanly — Claude treats each section as a distinct set of instructions rather than one blended paragraph.

Claude’s Unique Strength: Long Context

Claude’s context window handles up to 200K tokens. This means you can attach entire codebases, lengthy style guides, or multiple reference documents without worrying about hitting limits. Your instructions plus attached context can be far more detailed than any other platform.

Gemini: Gems + Google Integration

Gemini’s custom instruction system centers around Gems — custom versions of Gemini for specific tasks.

Gems

Think of Gems as Gemini’s version of Custom GPTs. Each Gem has:

• A custom name and description
• Specific instructions for behavior
• Integration with Google’s ecosystem (search, docs, etc.)

Creating a Gem: Gem Manager → Create → Define instructions.

Best for: Tasks that benefit from Google integration. Research with live search, document analysis with Google Docs, scheduling with Calendar.

Google Ecosystem Advantage

Gemini’s edge is integration. If your workflow involves Google Workspace, Gemini can access your Docs, Sheets, and Drive files natively. Custom instructions combined with Google data access can be powerful for business users.

✅ Quick Check: You’re choosing between a ChatGPT Custom GPT and a Gemini Gem for a research assistant. What would tip the decision toward Gemini? (If your research involves searching the web for current information and working with Google Workspace documents, Gemini’s native Google integration gives it an edge. If you need uploaded knowledge files and shareable tools, ChatGPT’s Custom GPT is stronger.)

Copilot: Instructions + Memory

Microsoft Copilot has a dual system:

Custom Instructions: Explicit rules you set in Preferences.

Memory: Copilot learns from your interactions and remembers preferences over time. Unlike the other platforms, this is partly automatic — it adapts to you.

Best for: Users deep in the Microsoft ecosystem (Teams, Outlook, Word). Copilot can apply instructions across Microsoft 365 apps.

Cross-Platform Comparison

The Multi-Platform Strategy

If you use multiple platforms (many people do), here’s a practical approach:

1. ChatGPT Custom Instructions → General preferences (tone, format, role)
2. ChatGPT Custom GPTs → Specialized tasks (code review, writing, analysis)
3. Claude Projects → Deep work with reference documents (long writing, complex analysis)
4. Gemini Gems → Google-integrated tasks (research, workspace)

Don’t try to replicate the same instructions everywhere. Play to each platform’s strengths.

Key Takeaways

• ChatGPT has three layers: Custom Instructions (global), Custom GPTs (task-specific, shareable), Projects (context-specific)
• Claude excels at long, XML-structured instructions with attached reference documents
• Gemini Gems integrate with Google’s ecosystem — ideal for search-heavy and Workspace tasks
• Copilot adds adaptive memory on top of explicit instructions within the Microsoft ecosystem
• A multi-platform strategy plays to each platform’s strengths rather than duplicating instructions

Up Next

You know the techniques and the platforms. In Lesson 6, we’ll build real-world templates — ready-to-use instruction sets for writing, coding, analysis, business, and creative work. Copy, customize, and deploy.