AI Tools Deep Dive: ChatGPT, Claude & Gemini

Basic prompting — typing a question and reading the answer — gets you maybe 30% of what these tools are capable of. Advanced prompting unlocks the rest. This lesson covers the frameworks that power users rely on every day.

1. Role prompting

Tell the AI who to be before you tell it what to do. This shapes tone, vocabulary, depth, and framing in ways that dramatically improve output quality.

The more specific the role, the better. "You are an expert" is weaker than "You are a CPA who has reviewed hundreds of small business tax situations."

2. Chain-of-thought prompting

For complex reasoning tasks, ask AI to show its work before giving a conclusion. This dramatically improves accuracy on multi-step problems.

3. Few-shot prompting

Give AI 2-3 examples of the output format or style you want before asking for the real thing. This is the fastest way to get structured, consistent output.

4. Constraint prompting

Tell AI what NOT to do, in addition to what to do. Negative constraints often produce bigger improvements than additional positive instructions.

Examples: "Do not use jargon or acronyms." / "Do not include more than 3 bullet points per section." / "Do not make any claims I haven't given you evidence for." / "Do not use the word 'leverage.'"

5. Iterative refinement

Your first prompt is a brief. The second prompt is your feedback. The third is your approval or another refinement. Professional power users rarely settle for the first response — they iterate 2-4 times and the quality gap between iteration 1 and iteration 3 is enormous.

These three tools are genuinely different — in capabilities, strengths, and best use cases. Using the right tool for the right job consistently produces better output with less effort.

ChatGPT (OpenAI)

Best for: Broad general tasks, code generation and debugging, image generation (via DALL-E), custom GPTs for specialized workflows, web browsing, data analysis with Code Interpreter.

Strengths: Largest ecosystem of integrations, best code generation in most benchmarks, most-used tool (which matters for team adoption), excellent file analysis (PDFs, spreadsheets, images).

Weaknesses: Can be overly agreeable and reluctant to push back on flawed assumptions. Shorter effective context than Claude for very long documents.

Claude (Anthropic)

Best for: Long documents (100,000+ words), nuanced writing that sounds human, complex instruction-following, tasks where you need the AI to maintain careful qualifications, safety-sensitive content.

Strengths: Best at following complex multi-part instructions. Produces the most natural-sounding prose. Most likely to flag its own uncertainty. Best-in-class for legal, compliance, and nuanced professional writing.

Weaknesses: Can be over-cautious on borderline requests. Less strong on code than GPT-4. Fewer third-party integrations.

Gemini (Google)

Best for: Real-time web research with citations, Google Workspace integration (Docs, Sheets, Gmail), multimodal analysis (images + text), tasks where current information matters.

Strengths: Best native web access and real-time information. Deep Google ecosystem integration. Excellent for research tasks that require current data. Strong vision capabilities for image-heavy documents.

Weaknesses: Trailing in complex reasoning compared to GPT-4o and Claude 3.5. Google Workspace integration is powerful but requires being in the Google ecosystem.

Power users don't start from scratch every time. They set up their tools to know who they are, what they do, and how they work — so every conversation starts with that context already loaded.

ChatGPT Custom Instructions

In ChatGPT, go to Settings → Personalization → Custom Instructions. You'll see two fields:

• "What would you like ChatGPT to know about you?" — Fill this with your professional context: your role, industry, company size, key responsibilities, and how you typically use AI.
• "How would you like ChatGPT to respond?" — Fill this with your output preferences: preferred length, tone, format, things to avoid.

Claude's system prompt via Projects

In Claude, Projects let you create persistent workspaces with context and documents pre-loaded. Create a project for each major area of your work (e.g., "Marketing," "Client Communications," "Strategic Planning") and upload relevant documents, guidelines, and templates to each.

Building prompt templates

For any task you do more than once a week, build a reusable prompt template. A good template:

1. Sets the role and context
2. Describes the task with placeholders ([YOUR INPUT] or [PASTE HERE])
3. Specifies format and length
4. Includes any standing constraints

Store your templates in a dedicated note or doc. This becomes your AI workflow library — one of your most valuable professional assets as AI becomes standard.

Keyboard shortcuts and browser extensions

For ChatGPT and Claude power users: consider a text expander tool (like Espanso, free and open-source, or TextExpander). You assign short triggers (e.g., ";email") that expand to your full email-writing prompt template. Combine with browser hotkeys for instant access to your AI tool of choice.

This lesson brings together everything from this module into a complete workflow for the highest-value professional output: research reports, long-form content, and professional documents at scale.

The complete research-to-report workflow

Here's a workflow that produces a professional research report in 90 minutes that would traditionally take a full day:

Phase 1: Frame the research (10 min)

Phase 2: Gather and summarize sources (30 min)

Use Perplexity AI or ChatGPT with web access for current data. For documents you have, use Claude. Paste each source with the prompt: "Summarize this for a report on [topic]. Extract: key data points, main argument, limitations, and 3 direct quotes I could use."

Phase 3: Generate the outline (10 min)

Phase 4: Draft section by section (30 min)

Don't ask AI to write the whole report. Draft each section individually, providing the outline entry and relevant research as context for each. This produces dramatically better output than a single "write the report" prompt.

Phase 5: Edit and finalize (10 min)

Quality control for AI-assisted documents

• Every specific statistic gets verified in the original source
• Every quote gets confirmed as accurate
• Read the full document once as if you wrote none of it — catch things that don't sound like you
• Have a colleague review any document that goes to an external audience

Your AI power user toolkit — a summary

You've completed Module 4. Here's a reference card for everything you've learned:

• Role prompting: Tell AI who to be before what to do
• Chain-of-thought: "Walk me through your reasoning step by step"
• Few-shot: Give 2-3 examples before the real request
• Constraints: Tell AI what NOT to do
• Iteration: First response is a draft; refine 2-3 times
• Tool selection: Claude for writing/documents, ChatGPT for code/analysis, Gemini for research/Google Workspace
• Custom instructions: Set up persistent context in your tools
• Prompt library: Save every prompt that produces great output