• AI Growth Insights
  • Posts
  • Studying The Prompts Of The Top GPT Assistants - Here's What I learned

Studying The Prompts Of The Top GPT Assistants - Here's What I learned

What Can We Learn From The Most Succesful GPT Assistants?

Hello!

In today's article, I want to share what I've learned from reverse-engineering and studying the prompts of the four top-performing GPT assistants...

My goal was to uncover the top prompt engineering techniques used by the top GPT product builders.

By looking under the hood, I wanted to understand what makes these prompts exceptional and these products successful.

And I'm very glad I did, because I gained valuable insights that go way beyond prompt engineering itself.

You see, creating prompts for GPT Assistants is a different beast.

Prompt engineering skills are highly important - of course - but there is something else that makes these products tick.

And that's exactly what I want to share with you in this article.

And no matter if you're a beginner or a pro in prompt engineering, I think you will beefit from incorporating the lessons we are about to discuss.

Ultimate Prompting Guide

Want to learn how to create world-class prompts? I have a guide where I teach you exactly that.

  • Learn the cutting-edge prompting techniques;

  • Real world prompting examples and breakdowns;

  • Full module on AI Automation - From Start To Finish;

Secret Prompting Technique?

I started reading through these prompts looking for something that I didn't know already or had never heard before.

And to my surprise, there's no secret little-known prompt engineering technique that none of us has ever come across.

Not one of these GPTs are using complex, out-of-the-box prompting techniques that you can't replicate yourself.

And this is great news. Because once you grasp the fundamentals of prompt engineering, you can apply all of these techniques and build similar GPT assistants.

It's not as if they are using something beyond your reach.

With a solid understanding of the basics and the core principles of prompt engineering, your own creativity is the only limit.

What Makes Them Different and Better From Others

But even though I'm saying they are not using any unheard of prompt engineering technique, I don't want to mislead you into thinking it's simple to replicate what they did.

There's a lot of skill involved, especially in creating a great product experience and utilizing data effectively. 

Prompt engineering is simply a method of merging these two elements into a single GPT iinterface for others to use.

So, if you want to create something that is perceived to be better and more useful than all of the other GPT assistants out there, you also have to master both of these skills.

So allow me to go through each one of them...

Great Product Experience

You see, creating a GPT for others is different from making one for personal use.

When it's for personal use, you know exactly:
• How the prompt is structured;
• What inputs to expect
• What its capable of doing and what it's not capable of doing;
• How to optimize your queries to get the most from the prompt;

But when building for others, you can't assume they have the same knowledge about the prompt.

People are unpredictable, and they will approach the GPT assistant with all sorts of random interactions.

And to manage this randomness, you must adhere to UX-first prompting principles.

Take time to clearly explain in the prompt:

• How users should interact;
• How GPT should interact back in order to get the necessary information from the user;
• How to handle exceptions to user behavior;
• What features are available and EXACTLY how they work;

Again, you can't assume that your users are as experienced as you are in making the most out of your prompts.

And if you don't clearly outline how each one of the pieces of your assistant works, everything is going to fall apart.

Let's look at an example so you understand what I mean...

The following example is from AI PDF, currently the third most used GPT Assistant.

AI PDF works the following way: Users can upload their PDFs and start asking questions, and the chatbot will summarize and respond based on the PDF's content.

One of the standout features of AI PDF is the 'folder search'.

Now pay attention to the prompt snippet below...

Notice that it clearly explains how the 'folder search' feature operates.

It details the process, what inputs to expect, how to export folder links, and provides rules as well as a step-by-step to follow.

If, on the other hand, you don't clearly specify this in the prompt. GPT will be left to figure everything out on its own.

And we all know how that goes: The results would be so random and unpredictable that the feature would be nearly useless.

Using External Data

Besides using the cutting-edge prompt engineering best practices to enhance user experience, these products are also tapping into a resource that often goes unnoticed: external data.

You see, your GPT assistant can connect to the internet to access real-time data and interact with online resources through something called "Actions". Actions works similarly to the now defunct "Plugins".

They enable your GPT to connect to the internet and access websites and services via an API.

Thus allowing your assistant to break free from the confines of the ChatGPT interface.

This unlocks so many possibilities its not even funny.

Take a look at the following example... It's from Canva GPT.

Prompt Snippet of the Canva GPT outlining how its plugin works

Canva is a user-friendly design tool aimed at people who may not be skilled in graphic design and seek an easier way to create designs.

But, unlike Canva, ChatGPT doesn't have design capabilities and isn't well-versed in design beyond Dall-E 3.

So how can they both be merged in the same interface?

Here's what CanvaGPT did:

1. It used ChatGPT merely as a front-end to interact with users, identifying their design needs through conversation;
2. Once these needs are identified, the information is sent to Canva through an API.
3. The complex design work is done on Canva's platform, not within the GPT interface.
4. After Canva completes the design, it sends the finished product back to GPT, which then presents it to the user.

As you can see, GPT is used only to discover the user's needs. And then this information is processed by a specialized third-party service.

This means that, if you can find (or build) an API to process data, you can greatly expand the capabilities of your GPT assistant.

It could be used to pull in news articles, Youtube videos, or any other data you can think of.

And no to mention that APIs are not the only way to use external data inside of ChatGPT.

You can also upload your own files with detailed information about topics your GPT covers.

Alternatively, you can combine both methods to come up with something truly unique.

I've been emphasizing this a lot lately: don't focus just on prompt engineering.

Also focus on how to use data to enhance your inputs and how to create a great user experience for those on the other side of the screen.

If you focus on two key principles, you'll significantly increase your chances of creating something that people will not only use but also recommend to others.

And that's all for today's article. Hope you've learned a thing or two.

See ya in the next one.

PS: Need help implementing AI in your business?

  • AI Automation;

  • Prompt Engineering;

  • AI for Content Creation & Social Media Growth;

  • And much more.

Reply

or to participate.