• Brain Bytes
  • Posts
  • 💭 The "Chain-of-Thought" trick that fixes bad AI results.

💭 The "Chain-of-Thought" trick that fixes bad AI results.

PLUS: How to get professional-grade output from free AI tools.

October 30, 2025

I've been testing AI tools for 18 months now, and I've noticed something interesting: Two people can use the exact same tool and get wildly different results.

One person asks ChatGPT a question and gets generic fluff. Another person asks a similar question and gets insights that actually solve their problem.

The difference isn't the tool. It's not even the question. It's the technique behind the question.

Most AI tutorials teach you what buttons to press. Today, I'm sharing three advanced techniques that change the quality of output you get—regardless of which AI tool you're using.

💡 You can still build your own personalized AI Learning System for a one-time $29 fee — designed to help you master these techniques inside a structured system.

🎯 Technique #1: Chain-of-Thought Prompting (Make AI Show Its Work)

What It Is: Instead of asking AI for an answer, force it to break down its reasoning step-by-step. This dramatically improves accuracy for complex problems.

The Framework: Add this phrase to any complex question: "Think through this step-by-step, showing your reasoning at each stage before giving your final answer."

Real Application: Instead of: "Should I take Social Security at 62 or wait until 70?"

Try this:

I'm trying to decide whether to take Social Security at 62 or wait until 70.
My current savings are $400K, I'm in good health, and my estimated benefit is
$2,200/month at 62 vs $3,100/month at 70.

Think through this step-by-step, showing your reasoning at each stage:
1. Calculate the breakeven point
2. Factor in opportunity cost of early benefits
3. Consider longevity risk
4. Evaluate tax implications
Then give me your analysis with specific numbers.

Why This Works: When AI explains its thinking process, it catches its own errors and gives you something you can actually verify. You're not just getting an answer—you're seeing the logic so you can make an informed decision.

Best Tool for This:ChatGPT (o1 model designed for reasoning) or Claude (excels at structured analysis).

💡 Click the underlined tools above to explore their official homepages.

🎯 Technique #2: Multi-Modal Research Using NotebookLM

What It Is: Most people know ChatGPT can answer questions. Few people know NotebookLM can analyze multiple sources simultaneously and generate podcast-style audio summaries.

The Power Move: Upload 5-10 sources on the same topic (PDFs, articles, YouTube transcripts) and ask comparative questions across all of them.

Real Application: Let's say you're researching the best treatment options for a health condition. Instead of reading 8 different medical articles:

  1. Go to notebooklm.google.com
  2. Upload the articles, your doctor’s notes (as PDF), and relevant YouTube video URLs
  3. Ask: "What are the points of agreement and disagreement across these sources about treatment effectiveness?"
  4. Then: "Generate an audio overview of the key decision factors I should discuss with my doctor"

You get a 10-minute podcast where two AI hosts discuss your research in conversational language—perfect for listening during a walk while you process the information.

The Advanced Move: After reviewing the audio, ask specific follow-up questions like: "Source 3 and Source 7 contradict each other on side effects. Which one is more recent and what’s their methodology?"

Why This Beats Traditional Research: You're not just reading—you’re having AI synthesize multiple expert perspectives and identify gaps in consensus. It’s like having a research assistant who read everything overnight.

💡 Click the underlined tool above to try NotebookLM directly from Google.

🎯 Technique #3: The "Act As" Framework for Expert-Level Output

What It Is: Most people ask AI generic questions. Smart users tell AI to embody a specific expert persona with relevant constraints.

The Framework:

Act as a [SPECIFIC EXPERT] with [RELEVANT EXPERIENCE]. I need [SPECIFIC OUTPUT] for [CONTEXT]. Consider [CONSTRAINTS/PRIORITIES]. Format your response as [STRUCTURE].

Real Applications:

For Writing:

Act as an estate planning attorney with 20 years of experience. I need to write a letter to my children explaining my end-of-life care wishes and where to find important documents. This is sensitive, so the tone should be clear and loving, not morbid. Format as a letter they can keep with my will.

For Analysis:

Act as a financial advisor who specializes in retirement portfolio rebalancing. I'm 58 with $600K in stocks, $200K in bonds, and $150K in cash. I want to retire at 65. Analyze whether my allocation makes sense given current market volatility and give me three specific rebalancing scenarios with pros/cons of each.

For Problem-Solving:

Act as an occupational therapist specializing in home modifications for aging in place. I want to make my two-story home safer as I age without major renovations. Walk through each room and suggest practical, budget-conscious modifications prioritized by safety impact.

Why This Works: Generic questions get generic answers. When you specify the expert role and constraints, AI taps into deeper training patterns specific to that domain. You get nuanced, professional-grade responses instead of surface-level summaries.

Best Tools:Claude excels at taking on expert personas with nuanced thinking. ChatGPT works well too, but Claude tends to maintain the persona more consistently across follow-up questions.

💡 Click the underlined tools above to explore their official homepages.

Just before you go

Hope you enjoyed this edition of Brain Bytes. Got feedback, suggestions, or cool AI tools to add?

You can always reply directly to this email — I read everything you send.

See you later this week, — Oliver

Oliver