Prompt Engineering, Part 2: Techniques That Matter

In part one, we covered CTFE, the most common mistakes, and four templates you can copy straight away. That gets you far.

But I ended with "that's next level." This is next level.

The techniques you'll learn here make a difference when CTFE isn't enough. When AI gives you okay answers but you want genuinely good ones. When you want to stop repeating yourself and start working smarter.

Think Out Loud: Chain-of-Thought

The simplest technique that makes the biggest difference on complex questions. Instead of asking AI to jump straight to the answer, you ask it to reason step by step.

Why does it work? AI that "thinks out loud" makes fewer logical errors. It doesn't miss important factors, and you can see if the reasoning goes off track.

Without chain-of-thought:

Should we raise our prices by 15%?

"A 15% price increase may be justified if your costs have risen. Consider communicating the value to customers."

Generic and useless.

With chain-of-thought:

I'm considering raising prices by 15%. Think through this step by step:
- What risks are there?
- What are the benefits?
- How does it affect different customer groups?
- Then give me your recommendation with reasoning.

Now you get a thoughtful answer that weighs pros and cons, identifies which customers are most affected, and gives a concrete recommendation you can actually act on.

Use this for: decisions, analysis, problem-solving, strategy questions. Not for simple writing tasks.

Show What You Want: Few-Shot

Few-shot means giving AI 2-3 examples of what you want before you ask your question. Instead of explaining the style, you show it.

Think of it this way: if you ask a designer to make a logo and say "make it cool," that can be interpreted a thousand ways. But if you show three logos you like, the designer gets it immediately.

Example:

Write product descriptions in the same style as these:

Example 1:
"Cut by hand. Dried for weeks. The taste of real sourdough
can't be faked."

Example 2:
"Fifteen minutes. That's all it takes. Toss it in, stir, serve.
No stress, just good food."

Now I want a product description for: Organic olive oil from Sicily

AI matches the style impressively well. Short sentences. Sensory details. No exaggeration.

Few-shot works best for: brand voice, consistent formatting, specific writing styles, and anything where "tone" matters.

Say What You Don't Want: Negative Instructions

We touched on this in part 1. Here we go deeper.

Negative instructions are surprisingly powerful. AI tends to fall into patterns: corporate-speak, overblown adjectives, repetition. By explicitly stating what you don't want, you close those doors.

Thank-you note to a customer:

Write a thank-you note to a customer.

Avoid:
- Formal language ("hereby", "we wish to inform")
- Excessive superlatives
- Repetition
- Passive voice

The tone should be warm but professional.

LinkedIn post:

Write a LinkedIn post about us hiring three new team members.

Avoid:
- "We're proud to announce"
- "Exciting journey"
- Words like "passion" and "driven"
- More than two emojis

Write like a human who actually talks to people.

The pattern: always say what you want instead. "Avoid formal language" + "the tone should be warm" works better together than either one alone.

Give Direction, Not "Redo It": Iteration with Feedback

AI is like a designer. "I don't like it, redo it" doesn't help. "I like the structure but the tone is too formal and the second paragraph needs a concrete example" helps a lot.

Anti-pattern:

This wasn't good. Redo it.

Specific feedback:

Good starting point! Adjust like this:
- Shorten the first paragraph to max 2 sentences
- Replace "innovative" with a concrete example of what we do
- Add a clear call to action at the end

Multi-round iteration:

Round 1: Ask for a first draft using CTFE. Round 2: Give specific feedback on tone, length, and content. Round 3: Fine-tune details. "Change the heading," "add a sentence about X at the end."

Three rounds is almost always enough. If you're not happy after three rounds, it's usually the prompt that needs rewriting, not the answer that needs more tweaking.

System Prompts: Your AI Persona

In part 1, I mentioned you could "create a system prompt with all context about you and your business." Now we actually do it.

A system prompt is background information that AI carries with it in every conversation. Instead of explaining who you are and what you do every time, you create one prompt and reuse it.

Template for a business persona:

You help [Name] at [Company].

About the company:
- [What you do, in one sentence]
- [Target audience]
- [Tone: describe how you communicate]

Rules:
- Always write in [language]
- Keep answers [short/medium/detailed]
- Avoid [things you hate in writing]
- Use [words/phrases typical of your brand]

Common tasks:
- [Type 1: e.g. customer emails]
- [Type 2: e.g. social media]
- [Type 3: e.g. proposals]

Where to save it:

• Claude Projects: Create a project and paste the system prompt as "Project instructions." Everything you write in that project automatically gets the right context.
• ChatGPT Custom Instructions: Go to Settings > Customize ChatGPT. Paste it in the "Custom instructions" box.
• Simplest approach: Keep the system prompt in a document and paste it at the start of new conversations.

This saves time every day. No more "I run a small business..." in every prompt.

When to Use What

A quick guide so you don't have to guess:

Writing tasks (emails, posts, copy) → Few-shot + negative instructions

Decisions and analysis (strategy questions, comparisons) → Chain-of-thought

Refining answers (when the first draft isn't enough) → Iteration with direction

Recurring tasks (same type of question every week) → System prompt + saved templates

And yes, you can combine them. A system prompt with few-shot examples and negative instructions for chain-of-thought analysis? Absolutely. That's how you go from good to genuinely good.

6 More Templates You Can Steal

In part 1, you got 4 templates. Here are 6 more.

Improve Text

Improve this text. Keep the message but:
- Make it more [concise/engaging/professional]
- Remove repetition
- Clarify unclear sentences

The goal is [what the text should achieve]. The audience is [who's reading].

[Paste text]

Explain a Concept

Explain [concept] as if I were [audience/knowledge level].
Avoid jargon. Use concrete examples.
Max 200 words.

Review a Document

Read through this document and identify:
1. Ambiguities or contradictions
2. Things that are missing
3. Phrasing that could be misunderstood
4. Suggestions for improvement

Be specific with quotes from the text.

[Paste or upload document]

Create a Template

Create a template for [type of document] that I can reuse.
Mark places where I should fill in my own information with [PLACEHOLDER].
Include brief instructions for each section.

Rephrase for a Different Audience

Rephrase this text for [new audience].

The original was written for [original audience].
Keep the core message but adapt:
- Word choice and tone
- Examples and references
- Level of detail

[Paste text]

Analyze a Problem

I have this problem: [describe the problem]

Help me analyze it:
1. What's the root cause (not just the symptoms)?
2. What options do I have?
3. What are the pros and cons of each option?
4. What would you recommend and why?

Save What Works

You'll find prompts and templates that work well for you specifically. Save them, or you'll rewrite them from scratch every time.

Simple folder structure:

Prompts/
├── Customer communication/
├── Social media/
├── Internal/
└── Analysis & strategy/

Each file: the prompt + an example result + one line about when it works best. It takes 30 extra seconds to save and saves minutes every time you need it.

TLDR

1. Chain-of-thought for better reasoning: "think through this step by step."
2. Few-shot for the right style: show 2-3 examples instead of explaining.
3. System prompt to stop repeating yourself: create once, reuse always.
4. Iterate with direction, not "redo it": give specific feedback in 2-3 rounds.

CTFE from part 1 + these techniques. That's all you need.

Frequently Asked Questions

Do I have to use all techniques every time?

No. CTFE is enough for most tasks. Chain-of-thought is great for analysis, few-shot for writing styles, and system prompts for recurring tasks. Pick what fits the situation.

Which technique makes the biggest difference?

Chain-of-thought, if I had to pick one. Asking AI to reason step by step improves the quality of almost any answer that requires some form of analysis or judgment.

Do system prompts work on free versions?

ChatGPT Free has Custom Instructions. Claude Free doesn't have Projects, but you can paste your system prompt at the start of each conversation. Works just as well, it's just an extra step.

Can I combine the techniques?

Absolutely, and that's the point. A system prompt with few-shot examples that asks for chain-of-thought reasoning? That's how you get genuinely good results. Experiment and see what works for your tasks.

---

Want us to put together system prompts and templates for your specific business? Check out our AI strategy service or book a call and we'll sort it out.