Tips & Tricks

Time: ~15 minutes What you'll learn: Advanced techniques to maximize AI effectiveness

---

This Page Covers

• Multi-LLM Workflows - Using multiple AI tools together for better results
• Context Optimization - Reducing token usage while keeping quality
• Prompt Iteration - Building on AI outputs effectively, including using AI to generate better prompts
• Quality Control - Techniques for more reliable results

---

Multi-LLM Workflows

Different AI models have different strengths. Using them together can produce better results than any single model alone.

Cross-Model Review

Use one AI to review another's output:

1. Generate with your primary AI (e.g., Claude)
2. Review with a different AI (e.g., Gemini, ChatGPT)
3. Refine based on the review feedback

Example workflow:

Step 1 (Claude): "Write a product description for [X]"
Step 2 (Gemini): "Review this product description. What's unclear,
                  missing, or could be improved? [paste output]"
Step 3 (Claude): "Update based on this feedback: [paste review]"

Why This Works

Each model has blind spots. A second model often catches:

• Logical inconsistencies
• Missing information
• Unclear phrasing
• Factual errors
• Alternative perspectives

TIP

This is especially useful for important documents - proposals, contracts, reports, or anything that will be read by many people.

Specialized Strengths

Different models excel at different tasks:

Model	Particularly Good At
Claude	Nuanced writing, long documents, following complex instructions
ChatGPT	Broad knowledge, creative tasks, conversational flow
Gemini	Google ecosystem integration, current information, multimodal
Perplexity	Research with citations, fact-checking

Consider using the right tool for each step of a complex task.

---

Context Optimization

AI models charge by tokens (roughly 4 characters = 1 token). Large documents eat through your context window fast. Here's how to optimize.

Prefer Flat Text Formats

Important

Always convert documents to plain text before uploading. This dramatically reduces token usage.

Format	Problem	Solution
PDF	Contains formatting metadata, sometimes OCR artifacts	Convert to .txt or .md first
Word (.docx)	XML structure adds overhead	Copy/paste as plain text, or export to .txt
PowerPoint	Slide metadata, images described verbosely	Extract text content only
Spreadsheets	Cell formatting, formulas	Export relevant data as CSV or markdown table

Conversion Tools

Simple options that work well:

• Copy/paste - Often the easiest; paste into a text editor, then into AI
• "Save as" .txt - Most apps support plain text export
• Online converters - Search for "[format] to text converter"
• pandoc - Command-line tool for developers: pandoc doc.pdf -o doc.txt

Extract Before Uploading

Instead of uploading a 50-page document:

1. Identify what you actually need - Usually a few key sections
2. Copy just those sections as plain text
3. Provide context - "This is the methodology section from a research paper about X"

Before (wasteful):

[Uploads entire 50-page PDF]
What does this say about methodology?

After (efficient):

Here's the methodology section from a paper on customer churn:
[paste 2 pages of relevant text]
Summarize the key approaches used.

Summarize Large Contexts

For truly large documents, use AI to create a working summary:

1. First pass: "Read this and create a detailed summary I can use for follow-up questions"
2. Save the summary in a text file
3. Use the summary for subsequent conversations instead of the original

---

Prompt Iteration Techniques

Getting great results often requires iteration. Here are techniques that help.

Generate Better Prompts with AI

Not sure how to phrase your request? Ask AI to create the prompt for you:

Generate a detailed prompt for the following question:
"How do I improve my team's meetings?"

The AI will produce a well-structured prompt with context, constraints, and specific asks. Then copy that generated prompt into a new conversation for a fresh start with better framing.

This works because:

• AI knows what information helps it give better answers
• You get a prompt with structure you might not have thought of
• Starting fresh avoids any context confusion from the prompt-generation conversation

Ask AI to Ask Questions

Before diving into a task, have AI identify what it needs:

I want you to help me write a business proposal. Before you start,
ask me 5-10 questions that will help you write a better proposal.

This surfaces requirements you might have forgotten to mention.

Request Multiple Options

Instead of one answer, ask for variations:

Give me 3 different approaches to [X]:
1. A conservative/safe option
2. A creative/bold option
3. A balanced middle-ground option

This gives you options to choose from or combine.

Iterate in Layers

Build complex outputs step by step:

1. Structure first: "Create an outline for [X]"
2. Review and adjust: "Move section 3 before section 2, and add a section about Y"
3. Expand sections: "Now write section 1 in full"
4. Refine: "Make the tone more professional"

This gives you control at each stage rather than hoping one prompt produces perfection.

Save Your Good Prompts

When you create a prompt that works well:

• Save it in a notes app or document
• Include an example of good output it produced
• Note what task it's useful for

Over time, you build a personal library of reliable prompts.

---

Quality Control Techniques

The "Explain Your Reasoning" Check

After getting an answer, ask:

Explain your reasoning step by step. What assumptions did you make?

This often reveals:

• Incorrect assumptions
• Gaps in logic
• Places where the AI "hallucinated"

Request Confidence Levels

For factual questions:

Answer this question and rate your confidence (high/medium/low)
for each part of your answer.

Low-confidence answers warrant verification.

Verify with Sources

For important facts:

• Ask AI to provide sources
• Check those sources yourself
• Use a search-enabled AI (like Perplexity) for current information

WARNING

AI can and does make up plausible-sounding citations. Always verify important facts independently.

The Adversarial Check

Ask AI to critique its own output:

Now act as a skeptical reviewer. What are the weaknesses
or potential problems with what you just wrote?

---

Quick Reference: Power Tips

Tip	When to Use
Cross-model review	Important documents, proposals, anything public-facing
Convert to plain text	Any time you're uploading documents
Extract key sections	Large documents where you only need parts
Generate prompts with AI	When you're not sure how to phrase a complex question
Ask for questions first	Complex or ambiguous tasks
Request multiple options	Creative decisions, strategy choices
Build in layers	Long or complex outputs
Save good prompts	Any time a prompt works well
Ask for reasoning	When accuracy matters
Request confidence levels	Factual or technical questions

---

Key Takeaways

• Use multiple AIs together - Have one model review another's work for better quality
• Optimize your context - Convert to plain text, extract what you need, summarize large docs
• Iterate strategically - Ask for questions, request options, build in layers
• Verify important outputs - Ask for reasoning, check confidence, verify facts
• Save what works - Build a library of effective prompts for reuse

---

Course Complete!

You've finished the AI Power User course. You now know how to:

• Get consistently better responses from AI
• Set up Claude Projects for your specific work
• Analyze documents and data effectively
• Create quick prototypes to visualize ideas
• Know when AI is and isn't the right tool
• Use AI responsibly at work
• Apply advanced techniques for even better results

What's Next?

Ready to build real things? The Builder path teaches you to:

• Deploy real applications
• Connect AI to your systems
• Maintain what you build