The Simple Prompt I Use to Get Meaningful Responses From LLMs

I've been experimenting with LLMs since ChatGPT launched in late 2022. I've learned a lot about prompting them, i.e., interacting with them to get real value.

A couple of weeks ago, I posted about the AI prompt I use. Today, I'm going to elaborate more on this framework.

The below is based on my experience and personal research. I've come across many resources, and one of the most insightful was the AI Playbook for Research, CRO and Experimentation by Convert.com, brought to my attention by Morgan Legge during a conference talk she gave.

We'll take a real example and go over the different prompt steps along with the outputs. For this demo, we'll be using ChatGPT's Free Tier.

The Framework

The prompting framework I came up with consists of 4 components:

1. The persona: who do you want the LLM to be?
2. The context: why are you prompting the LLM?
3. The output: what is the information needed?
4. The format: which format do you want the answer to be in?

Step 1: Persona

I start off with describing the persona I want the LLM to adopt.

In other words, if the LLM was a real person, who would it ideally be?

(Don't underestimate how attributes you give to the LLM can influence its output.)

For our demo, we'll take the use case where I am a tech lead who wants to create a training program for my team members.

I want this program to be engaging and useful to the engineers and to the company.

Here's the Persona part of the prompt:

You are an experienced tech lead. You build great software that is scalable, reliable, and free of technical debt. You have a long record of scaling teams and successfully delivering SaaS products. You also have long experience creating training programs for your team members.

Step 2: Context

I describe the context: Who is my audience? Why am I doing it?

In this step, I provide all the information that could help my thinking partner deliver the best outcome.

The Context part of the prompt would be:

You want to create an environment for your team where continuous learning is adopted as a mindset. You want to push junior and senior engineers to learn and grow. This is in line with your company values.

Step 3: Task

Now to the detailed task I want ChatGPT to deliver. I describe the task in the finest detail.

In other words, I try to answer this question: How would I execute this task if I were in charge of it?

Here's the Task component of the prompt:

Come up with a detailed upskilling and training program that is suitable to every team member's unique character and skillset. The program should cover technical and soft skills. Make sure to include practical implementation steps.

Step 4: Output

This is where ChatGPT needs a lot of guidance. By default, it will resort to long sentences with a lot of redundant words.

I describe the output format as accurately as possible: What outcome am I expecting? In what format should it be?

The Output prompt:

Start off with the program outline in bullet points, short sentences, and in a professional and friendly tone. We will then work together to elaborate each bullet point.

I also make sure to include this instruction to get clear copy that does not include any clutter:

Sentences should be short. Use a friendly and professional tone. Do not include hashtags or emojis. Do not use complex wording and sentence structures. Do not use any word from the list between triple quotes: (list enriched since my earlier post about the word ban list)

'''

Hurdles

Bustling

Harnessing

Unveiling the power

Realm

Depict

Demystify

Insurmountable

New Era

Poised

Unravel

Entanglement

Unprecedented

Eerie

Connection

Unliving

Beacon

Unleash

Delve

Enrich

Multifaced

Elevate

Discover

Supercharge

Unlock

Unleash

Tailored

Elegant

Delve

Dive

Ever-evolving

Pride

Meticulously

Grappling

Weighing

Architect

Adventure

Embark

Navigate

Navigation

Dazzle

Foster

Testament

Landscape

Spearhead

'''

The combined prompt becomes as follows:

Iterating

The first result I get from ChatGPT is this:

The response is quite comprehensive, and this is one of the strengths of LLMs.

The suggested program starts with an overview of its vision and goals then moves on to practical implementation steps.

At this stage, I would prompt ChatGPT to elaborate more on each item based on the specifics of my team.

For example, for the tech conferences item under paragraph #3, I would ask ChatGPT to elaborate more with an actionable plan:

Elaborate more on how I can encourage participation in external tech conferences and include a practical implementation plan.

Here's ChatGPT's response:

This is a focused action plan that seems realistic and that could be a great way to encourage team members to attend tech conferences.

From here on, I work together with ChatGPT to focus on each of its suggestions.

Recap

To recap, the 4 steps of this simple prompt framework are:

1. Give the LLM a persona.
2. Describe your request context.
3. Explain the task in detail.
4. Make the expected outcome clear.

Once you receive the first answer, interact with the LLM back-and-forth to get to the ultimate outcome.

‍