ChatGPT is writing dumb & it’s because you don’t know how to prompt right…

5 tips only 3% of AI users know for correct prompting

Welcome again everyone,

Summer is getting warmer.

It almost reached 30 degrees here last week.

Are you also having a warm summer? I’d love to hear.

By the way… there’s one thing everyone talks about but not everyone implements:

“Clarity matters everywhere.”

It does when you talk, when you write, and even when you prompt.

If you’re familiar with copywriting, you know what I mean.

One of the best ways to write clear copy is by using frameworks. Some frameworks are:

PAS, AIDA, FAB, and Storytelling.

They help in writing with a clear structure.

Using them, you can tailor a copy for a specific purpose.

You don’t need to spend time structuring your copy every time.

Quite a time saver. Isn’t it?

In the same way, we now have frameworks for prompting.

Imagine writing a prompt for ChatGPT without even thinking about the frameworks.

Imagine just writing something like:

Write a short sales email for a middle-aged man who is interested in supercars.

Ok. Call this a prompt. ChatGPT doesn’t say no to anything.

But look at what the response was to this:

One thing that’s now become a fact: a lot of what the LLMs generate depends on how you prompt.

The output above is way less specific because the prompt wasn’t clear.

It seems like a desperate sales email, and that’s it.

Let me rewrite it below, but this time more specific.

(I will use one of the frameworks for this one, and you can later decide which one)

Suppose you have 10 years of experience in sales. I have listed below all the details about the prospect we will be targeting.

- A man aged between 30 to 40 years
- Has a lot of money
- Is interested in supercars

Write a sales email in a friendly but professional tone which talks to the prospect’s heart. Look for what interests him and try to build a rapport in the email. But try to avoid being desperate. Be as more friendly.

The purpose of the email is to remind him that it is time for an upgrade to a supercar. I also want the prospect to know that our new high-speed supercars are now available for sale after a long wait.

Here’s how this prompt went:

Now, make a quick comparison.

Which output is more specific and personalized?

I like the second one because it has a more genuine subject line.

Also, the email starts with the reader’s passion and smoothly transitions to the pitch.

If you were a cars passionate and received both emails, I bet you’d reply to the second one.

Yeah. All this impact comes from how the prompt is phrased.

Frameworks help you prompt in many ways.

They fix the order. They help the LLMs understand better and generate better.

If you want an well-engineered output, engineer your prompt.

And you can do this easily with frameworks.

So here are the prompting frameworks:

1: The RTF Framework

RTF stands for:

R = Role
T = Task
F = Format

Well. You might’ve seen the “Act as a…” phrase a lot in many prompts.

They are based on this framework, and…

Each part of the framework has a job to do.

Role specifies whose perspective the AI should use. A manager? A doctor? An experienced or inexperienced person?

Task tells the AI what to do. If the Role is that of a doctor, the Task can be “coming up with diet solutions.”

Format is how you want to order the output.

Sentence length, spacing, and punctuation usage are all formats.

Now, I’m going to give you an example of this prompt. Each part with its own task.

Role: Act as a Language expert,
Task: Research various types of literature,
Format: And show me what is the most common thing among all the languages in the world.”

Pretty neat. Isn’t it?

Okay. The next one…

2: The TAG framework

TAG stands for:

T = Task
A = Action
G = Goal

Well again, each of these forms a part of a prompt.

A Task defines what the AI should do. And this time, we start by putting it first.

“I want you to evaluate this monthly report…” is how it can be. The rest comes next.

Action is about invoking it on how to do it.

Now remember. Action is different from Task. It tells AI how to do something.

Writing a book is a Task and using a pen for it is Action.

In other words, it’s the approach in which we act.

Then comes the Goal.

Goal tells AI what to get from the evaluation.

Once we’re done with both these Task and Action, then what?

Then we should get a result. That’s the goal.

An example prompt will be like this:

Task: I want you to evaluate this monthly report,
Action: Use qualitative evaluation,
Goal: Tell me what can be done to improve the efficiency of each factor for next month.

Clear. Yeah?

The prompt can be bigger at times but we should keep 3 things in mind.

Task, Action and Goal.

Here’s another sample prompt:

Task: Write a professional sales email for car lovers.
Action: Research all the available information and find out what most car lovers like, what their preferences are and what they don’t like about cars.
Goal: Provide some convincing facts on how buying a new car can help them live happier and how it can improve the fun in their life.

Again, look at how I framed each part.

On to the next…

3: The BAB framework

BAB stands for:

B = Before
A = After
B = Bridge

Before specifies what the current problem or situation is.

It talks about the starting point from which we will go to the desired output.

If you have an issue, Before will define its context.

State in the prompt why it happened and what is causing it.

After acts as a specific desired solution that you may have in mind.

Let’s say; what will it look like once ChatGPT comes up with the output.

Like a cure for a problem.

Every time you’re specifying the After part, try to be as specific as you can.

It’s like a goal we set. So make it clear.

If you provide a point A and a point B, then whatever is connecting these points is the Bridge.

Or even clearer, it’s “How we achieve that goal?”.

Let’s go through an example:

Before: My time management is really getting bad. I barely finish work on time.
After: I want to be able to work more in less time and achieve more real results.
Bridge: Come up with some suggestions on how I can boost my productivity, following simple but actionable tips.

Ok. Now the 4th one…

4: The CARE Framework

CARE Stands for:

C = Context
A = Action
R = Result
E = Example

Context… it’s a picture of the situation.

If we are talking about a problem, it should refer to the solution.

Have a certain output in mind? Context is an explanation for that.

Action defines how to do something.

It’s a process through which we can achieve the desired output.

Then comes the Result. It is the output.

Each time you prompt, you expect a Result.

It depends on how you can explain that result in your own words.

Finally, Example.

Example helps a lot in setting an ideal picture.

If you know an output you like, provide it for inspiration.

Yeah. Examples inspire AI well.

Examples also help you have more control over what and how AI should generate something.

Look at this sample prompt:

Challenge: I want to lose 5kg of weight in the next month.
Action: Give me an actionable plan that is real and doable.
Result: The plan should help me lose 5kg of weight and do that in one month without burning out and having a slimmer body.
Example: I would love to have a suggestion for the workout routine of some of the well-known athletes who got some early results.

What about the next?

5: The RISEN Framework

R = Role
I = Input
S = Steps
E = Expectation
N = Narrowing

Role talks about the involvement of someone.

In prompting, the role is played by AI. But it’s better to specify whose role it should play.

A marketer’s role? Or an accountant’s?

Input is always what data you feed the AI in.

It can be a story, a problem, a statistic, or any information.

Steps is the plan.

Yes. A plan works better to generate an output.

You can also say it’s specifying how to achieve something. Start with a and end with b.

Expectation is more like the Result.

But what’s the difference?

Well. Result is more specific about what to achieve.

It already mentions what the output should look like.

But in Expectations, we still don’t know what the output will look like.

We hope it looks something similar to the example we provide. And we depend on other training data for how the output will be.

Finally, Norrowing. It’s the limitations we set for AI in a prompt.

You may specify a particular area to focus on or avoid a certain things in the output.

Take an example for this framework:

Role: Act as an experienced fitness coach.
Input: Recommend a plan on how I can build lean muscle.
Steps: Provide the plan in steps and then break down each step.
Expectation: I want to lose some weight but also have a nice body.
Narrowing: Keep the output in 50-80 words. (Not more than that)

Understood? Perfect!

Frameworks are a lot useful. But remember…

They alone won’t make you a prompting genius, too.

They help you as structures for curating your prompts. But you should know how to add details in those frameworks.

I suggest keeping these frameworks handy.

Use them. They will help a lot.

Thanks for reading this email,

And see you in the next one.

Sami Sharaf