⫹⫺ Good Writing, Great Prompts - AI Field Guide 🌾
CMMN WLTH 126 - Cheat Codes for Creators, Makers & Doers
Hello my friends,
Ignore the headlines.
If you’re in a creative role, you’re not going to lose your job to an ai agent anytime soon.
But the game is changing fast.
The ai wave is here and those who know how to leverage it will thrive.
In CMMN WLTH 124 I shared my thoughts on why I believe millennials have a head start on Gen Z in the ai economy.
It struck a chord with many of you.
“Any ai masterclass you would recommend?”
My inbox was inundated with questions and requests for resources on the topic.
So I thought I would create one.
CMMN WLTH ai Field Guide
It’s easy to get caught up in the hype, but the real life applications are impossible to ignore.
You don’t need a degree in computer science, expensive software or a powerful computer to access some absolutely mind blowing technology.
Learning to leverage ai is one of, if not the, highest agency skills you can acquire right now.
From creating unique imagery for mood boards, building a personal workout plan to reviewing consumer insight data, used effectively ai applications will not only save you time but expand your thinking.
The techniques and tactics I describe below are based on extensive reading, research and many hours of trial and error.
I am by no means an expert, but I wanted to share what I’ve learned in the hope that it will help your process and workflow.
Contents
Definitions
Recommended ai Applications
Think Smarter Not Harder
Prompt Type Definitions
How To Write Effective Prompts
Prompt Examples
Conclusion
🪢 Definitions
Artificial Intelligence (ai) - The broad field of computer systems that can perform tasks typically requiring human intelligence. This includes recognizing images, creating images from text, understanding speech, making decisions and solving problems. Examples: Chat GPT, Midjourney, Claude, Kling.
Large Language Model (LLM) - A specific type of ai that focuses on understanding and generating human language. They are trained on vast amounts of text data and can write essays, answer questions, translate and have conversations. Examples: Chat GPT, Claude and Gemini.
This guide will focus on LLMs and the example prompts below will work with any LLM model. You will however see variances in output depending on which model you choose (e.g. Claude vs. Chat GPT).
📲 Recommended ai Applications
Claude (Free, $17/month)
My current preference.
Best for: Writing, editing and project work.
Chat GPT (Free, $20/month)
The best known and most popular LLM.
Best for: Writing, editing and project work
Notebook LM (Free)
Google AI research assistant that functions as a notebook.
Best for: Analyzing and summarizing large text files (essays, research papers, etc).
Gemini (Free, $19/month)
Google LLM that can integrate with Gmail, Workspace and Search.
Best for: Scientific and mathematical reasoning.
Midjourney ($10/month upwards)
ai image generator that generates high quality images from text.
Best for: Generating high quality and creative images using text prompts.
💭 Think Smarter Not Harder
Let’s start with a mental model that completely changed how I interact with ai:
What separates LLMs from search engines is the ability to use them as thought partners.
Once you start treating the LLM as a smart co-founder or colleague output value runs 100x.
How do you leverage this ability?
By learning how to structure the right questions and setting the right context ↓
🧂 Prompt Types
Prompts are the text inputs that you enter into dialogue box of the LLM.
Creating effective prompts is like baking a cake.
Develop a recipe by combining the right ingredients and modifiers (inputs) to get the result (output) you are looking for.
There are roughly 12 methods of prompting.
Below are the four that I use most frequently:
Zero Shot - Asking the ai to do something without showing it how first. Like asking someone to bake a cake without giving them a recipe.
One Shot / Few Shot - Give the ai examples before asking it to do a similar task. Like showing someone how you solved two math problems before asking them to solve a third one.
Role Based - Asking the ai to pretend it's someone specific (like a product manager or manufacturing VP) to get more specialized responses. Similar to asking a friend to "put on their lawyer hat" when giving advice.
Chain-of-Thought - Asking the ai to think out loud and show its work. Like asking someone to explain each step while they're fixing your computer.
These can be used independently of each other or interchangeably throughout a dialogue with the LLM.
I provide some prompt examples below, but first lets learn how to write them in the most coherent and efficient way ↓
✍️ How to Write Effective Prompts
1. K.I.S.S. - “Keep It Simple Stupid”
Great prompting requires concise, clear and easy to understand writing. Focus on simple language, remove jargon and unnecessary information. Explain the task like you would to a new hire on your team.
This:
"Write a 300-word summary of climate change solutions that are being implemented today. Include examples of renewable energy, policy changes, and community initiatives. Use simple language that a high school student would understand."
Not:
"Analyze the current state of global warming and provide an extensive review of all possible solutions with exhaustive citations, references, and technical specifications, while ensuring comprehensive coverage of international policy frameworks dating back to 1992, and simultaneously addressing economic impacts across developed and developing nations without omitting any relevant statistical data points or scientific perspectives."
2. Be Specific
Ask for the exact output you are looking for and be as descriptive as possible. Vague requests will likely not produce the result you want.
Use instruction or constraints as the blueprint to your prompt. Clear blueprints produce better results.
Instructions: Guides the model on what it should do. Use explicit details on the desired format, style, or content of the response.
Constraints: Guides the model on what it should not do. A set of boundaries that make the output sharper.
Generally, instructions work better than constraints so focus on providing clear instructions vs. telling the LLM what not to do.
With that said, a constraint I use often is specifying a max. length. I use an exact word count or tailor it to tweet length or instagram caption.
This:
"Create a 5-day workout plan for building upper body strength. Format it as a table with columns for Day, Exercises (4 per day), Sets, and Reps. Include a mix of push and pull exercises, and add a brief note about proper form for each exercise."
Not that:
"I want to get stronger. Can you help me with some workout ideas? I'm thinking maybe focusing on my arms and chest area but I'm not really sure what I should be doing or how often."
3. Switch Roles
See yourself as the director and editor rather than the doer.
Don’t write an executive summary and ask ai to review it, ask it to draft one based on a strategy document, and use your time to tighten it up.
4. Don’t tell it what to do, ask how to do it
When you work with a co-founder or colleague the conversation flows between asking and answering questions. There is a creative friction where the best ideas emerge. See the LLM as a live partner. If you find yourself stuck on a problem spar and reason with it to get to a solution.
🎮 Example prompts
Below are some examples of prompts that I use in my daily workflow.
To use these yourself copy and paste the text into your LLM of choice.
Replace the text in < > and attach the documents you would like analyzed using the applications attach button or by copying and pasting the data manually.
Zero Shot:
Analyze the following consumer feedback data for <insert product or service>. Identify the top 3 concerns, categorize overall sentiment (positive, negative, neutral), and suggest two actionable improvements we could implement in the next iteration. Format your response as a concise executive summary followed by detailed findings that I can share with the design and manufacturing teams.
<Insert your actual customer feedback data here>
One Shot / Few Shot:
Analyze the following consumer feedback data for <insert product or service>. Identify the top 3 concerns, categorize overall sentiment (positive, negative, neutral), and suggest two actionable improvements we could implement in the next iteration. Format your response as a concise executive summary followed by detailed findings that I can share with the design and manufacturing teams.
Example analysis:
[Customer feedback data for another product e.g. water bottle]
"Love the concept but the lid leaks constantly."
"Too expensive for what you get."
"Beautiful design and I feel good using it."
Executive Summary:
Analysis of consumer feedback for our water bottle reveals mixed reception. While the sustainable design concept is appreciated, functional issues (leakage) and pricing concerns are hampering customer satisfaction. Two key improvement areas: redesign the lid seal mechanism and evaluate cost reduction opportunities.
Detailed Findings:
Top 3 Concerns:
1. Product functionality - Leak-prone lid (Negative)
2. Price point - Perceived as too expensive (Negative)
3. Design aesthetics - Appealing visual design (Positive)
Sentiment Analysis:
- Positive: 33% (appreciation for design and sustainability concept)
- Negative: 67% (functionality and price concerns)
- Neutral: 0%
Recommended Improvements:
1. Technical: Redesign lid seal mechanism with improved gasket material and conduct additional leak testing before release.
2. Business: Evaluate material or production costs to identify potential savings that could be passed to consumers, or better communicate value proposition to justify price point.
Now analyze the following feedback for <insert product or service>:
<Insert your actual customer feedback data here>
Role Based:
You are an experienced product development specialist with 15+ years in the consumer goods industry, specializing in <insert product or service>. Your background includes <specific expertise> and <specific expertise>.
Task: Review our consumer feedback data for <insert product or service>. Based on your expertise, identify the top 3 concerns, categorize sentiment (positive, negative, neutral), and suggest 2 actionable improvements we could implement in the next product iteration. Format your response as a concise executive summary followed by detailed findings that I can share with the design and manufacturing teams.
<Insert consumer feedback data for product or service>
Chain-of-Thought:
Analyze the following consumer feedback data for <insert product or service>.
Think through your analysis step by step:
1. First, read through all the feedback carefully and list each distinct concern or praise mentioned
2. Next, categorize each piece of feedback as positive, negative, or neutral
3. Then, identify patterns to determine the top 3 most frequently mentioned concerns
4. Calculate the overall sentiment distribution based on your categorization
5. Considering the concerns and sentiment, brainstorm potential improvements
6. Finally, select the 2 most impactful and feasible improvements to recommend
After working through these steps, identify the top 3 concerns, categorize sentiment (positive, negative, neutral), and suggest 2 actionable improvements we could implement in the next product iteration. Format your response as a concise executive summary followed by detailed findings that I can share with the design and manufacturing teams.
<Insert consumer feedback data here>
🔮 Conclusion
Search engines have us conditioned to fire questions at them and haphazardly unravel the results to reach a solution.
LLMs, on the other hand, have contextual understanding and conversational ability enabling you to work together to find solutions.
If you have a half baked idea → Discuss it
Stuck between options → Simulate it
Need another perspective → Challenge it
A few final thoughts:
Validate: Outputs are not always accurate and LLMs are notably bad at math. Mistakes, called “hallucinations”, are currently one of the major challenges for all models. Always double check results.
Experiment: Using the same prompts with different models (e.g. Claude, ChatGPT) will result in drastically different outputs. Test and experiment with different models and note the results.
Share: Send your best prompts to your friends and colleagues (and me) and ask them for theirs.
If you learned something I’d appreciate if you’d share CMMN WLTH with your closest collaborators and hit the like button at the top.
My inbox is always open for feedback, just reply directly to this email.
Thanks, as always, for your time and attention,
— Andy