The world will never be the same again now that ChatGPT and other LLM (large language model) text-based AIs are available to the average consumer. Now you too can become a cyborg with superhuman capabilities, by simply subscribing to OpenAI’s ChatGPT 4.0 paid membership which unlocks a significantly more intelligent version of the AI with the ability to integrate with other systems through plugins and surf the Internet through Bing, among other features.

But like all emerging technologies, ChatGPT has its good days and its bad days -and its good results and its poor, misleading, and worse –boring– results as well. Consistently, the biggest difference between crappy results and really good results is spending a little more effort on the prompt; by using a prompt framework, providing detailed instructions and constraints, examples, and answering follow-up questions.

The point being, investing more time and detail into the front-end (the prompt) yields exponentially better returns and turns ChatGPT from a novelty chat bot into an articulate oracle that you can rely on for wisdom and power.

It all comes down to the magic spell you use –your prompt

How to create a very effective ChatGPT prompt every time

When prompting ChatGPT for anything besides a basic search question (and even then I wonder), it’s worth investing just a little more time to follow a simple protocol when building your prompt. The one I was first introduced to, that works very well for me, is called the RICE framework.

R – Role (that you want ChatGPT to play)
I – Instructions (what you want ChatGPT to do)
C – Context (who the audience is or why you’re doing this)
C – Constraints (what you don’t want ChatGPT to do)
E – Examples (what you want ChatGPT to copy)

To use the framework you format your prompt to incorporate all of these details, like so:

You start by telling ChatGPT to “act as” whatever role you want it to play. In the example above, I ask it to act as a market researcher (role) and ask it to provide a brief on the state of the DFW startup scene (instructions), tell it to be concise, within 1,000 words and to write for an audience of investors (context and constraints), and then provide the DBJ and Crunchbase as examples. This addresses all the requirements of the RICE framework and should provide a useful response.

Not bad…I got a lot of very specific information -with sources- with just one prompt!

When you work with ChatGPT on complex requests using a prompt like this guarantees the best possible results because it is easy for ChatGPT to understand exactly what you want it to do.

But it can take some time to write out all of that for every question that you ask ChatGPT, so instead it’s often faster to ask ChatGPT to “output only the text for a detailed prompt, without executing the prompt, using the RICE (role, instructions, context, constraints, examples) prompt framework to produce the best possible response for the following task:” and then entering a simple version of the prompt off the top of your head.

Yes, that’s right, you can ask ChatGPT to write a prompt for itself . Mind-blowing, right?

That looks like this:

Asking for the prompt, rather than the output, first allows you to tweak the instructions (once you’ve seen them in detail) to fine-tune the output you’ll generate. When you’re happy with the prompt you’ve engineered, you can simply tell ChatGPT to “execute your last response as a prompt” and it will generate high-quality output according to the prompt it generated.

For deeper matters that involve multiple queries and nuanced responses, there is another approach that you can use, where you engage the platform in a lively discussion and then refer back to your chat thread for a specific request.

Having a little chat before getting to work

I’ll often perform market or subject matter research through a “conversation” with ChatGPT and when learned something and come up with a unique take from this research, I’ll vet the idea with ChatGPT by articulating my point to it and asking it to correct or disprove what I’m saying. When I get to the point where ChatGPT “agrees” with my take, then I’ll ask it to provide a very detailed outline for an article making that same point. It will draw on our entire conversation to provide the best response and produce a detailed outline, typically in bullet points, for me to follow while writing; saving me a ton of time.

This method works for any topic that you need to research. I’ve personally used ChatGPT to perform market research, looking for markets of opportunity in the specific US cities where I want to do business from abroad, fine-tuning my ICP (ideal customer profile), and doing competitive analysis on my offerings. Plus, any time I want to write about a subject that I only know conversationally, I start with a “conversation” with ChatGPT to fill in the gaps in my understanding and to discover threads to follow for deeper research outside the chat session.

Here’s how you do this, step-by-step:

  • Explore a topic with ChatGPT in one chat session until you come up with a compelling perspective: your opinion, a prediction, a “hot take”, etc.
  • Articulate your point to ChatGPT and ask it to confirm or correct your understanding.
  • Adapt your argument until ChatGPT agrees that what you’re saying is essentially correct or has the potential to be, if you’re making a prediction.
  • Finally, instruct ChatGPT to summarize your argument into an outline, brief, etc. so you can easily take the next step with it.

It sounds pretty simple but, actually, it can be very challenging to get ChatGPT to engage in a real, in-depth conversation with you in the first place. It all depends on the subject matter and your own level of understanding of it.

In my case, for example, I don’t benefit from having surface level chats about marketing and technology -because I already have a lot of expertise and have formed strong opinions on the subject. So sometimes when I ask a broad question, I’ll get an eyeroll-inducing response from ChatGPT telling me a bunch of things I already know this was very frustrating for me until I figured out what to do.

Tell ChatGPT to treat you like an expert

Context is everything when working with ChatGPT, and particularly the context within a given chat session (which is saved automatically and you can return to it over and over again). In the RICE framework, the first step is to tell ChatGPT the role you want it to play. This would be written in a prompt like this:

“Act as an expert in content marketing and generate an outline for an article about…”

But if you do that and don’t tell ChatGPT anything about yourself, it will assume that you no nothing about content marketing (in this example) and provide a very basic, high-level response that is perfect for a newbie and not at all what you want.

Instead, write your prompt like this to provide ChatGPT context about you as well:

“Act as an expert on content marketing, and treat me as one as well, and let’s collaborate on an outline for an article about…”

This will produce results that are in-depth and useful to someone who is already versed in the subject matter. ChatGPT also tends to make more helpful intuitive leaps and make more interesting recommendations when primed with this style of prompt as well.

Here’s a sample of the output from the prompt: “act as an expert content marketer and provide a brief paragraph explaining marketing trends in 2024.”

And here’s what it looks like with the prompt: “act as an expert content marketer and treat me as the same and let’s collaborate on a brief explaining the marketing trends of 2024 in detail.”

You can see the tone and level of detail are quite different and the latter version is clearly speaking to an audience of marketers rather than generic readers. And, for brevity’s sake I kept this example very short, so this isn’t even a particularly impressive sample. This is especially helpful when outlining an article or other piece of long-form content or a presentation.

The power of conversation

Everyone is saying it so it’s already becoming cliche, and yet, I don’t think it’s generally understood yet that absolutely every insane thing that AI can do today is the worst version that will ever exist. AI technology is getting so much better so fast and affecting so many aspects of life that the way essentially everything will work very differently than before in only a few short years.

It’s impossible to anticipate exactly how this is going to shake out but it is fascinating to watch the AI transformation take place. And even though we can only glimpse a tiny sliver of the radical changes to come, that sliver is still tantalizingly compelling.

One of the early changes we can already see taking shape is how we interact with computers and data. Thanks to LLMs like ChatGPT we’re now increasingly able to write and talk with computers as if they were people -people with millions of encyclopedias worth of information stored in their heads.

So you don’t have to worry so much about how you search to find what you’re looking for, you can just have a conversation with the search bot: more of this, less of that, you know what I mean…

People don’t need other people to present and interpret documents to them. You can load a document into ChatGPT and then have a conversation with the document. You can ask for a summary, in bullet-points, and then engage it in a conversation to get answers to any questions you have. You can ask it to find and display any parts of a document which relate to a point you want to learn more about. And I don’t mean searching for phrases; I mean you describe the concept to ChatGPT and it finds anything related to that topic regardless of the specific words used in the text. That’s what an LLM is great at doing: understanding human language and implied meaning.

There’s no need to type in search phrases or worry about keywords in website copy if the AI reading them (and writing them) understands the underlying context. And there’s no reason to type at all if the AI can transcribe spoken language to text reliably, so the majority of searches are going to be voice searches very soon.

The bottom line is that we all should get used to the idea of talking to computers to answer our questions, process data, and produce whatever work we do. There’s very little reason to read or write anything beyond the sheer love of reading or writing now that AI can handle the drudgery for us. So it won’t be long until we’re all talking to documents and training AI to research things for us -but for now, all this stuff sounds like science fiction and most people haven’t started moving in this direction yet.

We’re all going to go through this transformation over the next 5-10 years (at the latest) but if you start the process now it won’t be as harsh of a transition and you will be able to benefit (and profit) from being an early adopter of these soon-to-be required habits. As sure as the mouse and keyboard quickly replaced the typewriter as an essential skill, learning how to talk to computers -and get the responses you need- is the essential skill of the near future.