AI can certainly do a lot.
But without a solid direction, it may not take you where you need it to go.
Which is why Christopher S. Penn developed a framework to limit and direct tools like ChatGPT so they repeatedly generate a usable outcome.
And, incredibly, Chris shared this valuable process with his audience at B2B Forum 2024.
Now you can discover this simple, six-step framework for yourself in the video or transcript below.
And because AI moves so fast, you can expect even more advanced, mind-blowing AI techniques at B2B Forum 2025—plus dozens of inspirational, educational sessions on all things B2B from our accomplished presenters.
Don’t let the future of marketing pass you by. Join over 800 of your B2B marketing peers at B2B Forum, live in Boston this coming November. Discounted tickets are available now—claim yours before prices increase!
The prompt framework that we’re going to walk through today, with multiple examples, is what we call the RAPPEL framework.
And it’s six pieces.
And if you brought a laptop or you brought a tablet or something that can do this stuff, you’ll get a chance to follow along.
We built this framework based on about a dozen different academic papers that have researched them, how to prompt things well.
There’s probably close to a hundred different papers.
We picked the practices of probably the easiest to learn.
Number one: the role.
Tell the model who it is, eith jargon, with specification. “You are a Google Analytics certified professional skilled in GA4, Google Tag Manager, Google Marketing Platform, Google BigQuery. You specialize in attribution modeling from UTM tracking codes and tagging to complex uplift modeling…”
So the role is filled with jargon. It sets specific terms that automatically start helping the model go, “oh, this is the domain of knowledge I’m working with.”
Second: the action, the overview of the task.
You’re going to say, “we’ll be performing some attribution analysis today on Google Analytics 4 data to better understand what’s working in our marketing. Later, I’ll be providing you with some screenshots from Google Analytics 4, along with information on our goals and key conversions for you to analyze…”
So we’re giving it an overview of the task.
Third is: prime.
This part is really important.
“Before we begin, what do you know about this topic? What are common mistakes less experienced people make when it comes to Google Analytics for attribution analysis? What are your expert tips and tricks to make GA four attribution work well?”
And then you wait.
And you let whatever tool it is you’re using foam at the mouth for paragraphs and paragraphs…
And you validate it. You say, “is what it’s saying, correct? Does it have current knowledge about, in this case, Google Analytics 4?”
I’ve seen some horrendous prompts that say, “you’re an SEO expert, you’re going to SEO optimize this post.”
And then when you ask it about SEO, it’s giving you advice from 2012 because it has no idea what is relevant or not.
It just knows by volume of information.
So the priming step helps us to better understand, “what does the model know?” Validating its knowledge, and—remember what I said—every part of this prompt, the previous conversation becomes a part of the next prompt.
So if the knowledge that spits out is correct, then you’ve automatically now added several thousand words of relevant information for the next prompt.
Fourth step: you actually write the prompt.
“Let’s begin the attribution analysis. I’m going to attach two screenshots from GA4 of our conversion paths…
“The first screenshot uses default channel groupings. The second screenshot uses session source. The conversions of key events are newsletter signups…
“Create an attribution analysis of what’s working, what’s not working, and give concrete recommendations for next steps to improve newsletter signups.”
So now we’re saying, “do the thing.”
We’ve now established all the background. We’re providing our data and saying, “do the thing.”
After the model responds, you have a conversation with it.
Say, “well, that wasn’t what I was looking for,” or “can you make it look like this?” Or, “can you explain it in these terms?”
And then at the last step, we say, “take our process, and turn it into a prompt.”
What we want to do to make AI work well for us, over and over again, is to not do this process every single time.
It is to say, “help me build instructions for the system of your choice so that all I have to do is paste in the instructions the next time and the relevant information.”
And I know the model has builtin instructions for itself that work best for it.
By the way, if you have prompts that work well in ChatGPT, they probably do not work as well in Gemini. They probably do not work as well in Claude.
This step is what you would use to convert any prompt to a prompt for that specific model.
Because think of these things, as I said, like libraries. If you were to pick up a librarian from the New York Public Library and drop them off in the London Public Library, yes, the librarian would have a lot of transferable skills, but the librarian has no idea where anything is in that library, so it’s going to underperform, compared to keeping that employee in the library where they work.
The same thing is true of these things.
A prompt that works well in one system will give you some results in another system, but it’s not optimum for that system.
So this is the six step process for building better prompts.
Published Feb 13, 2025
MarketingProfs B2B Forum 2024 was our favorite yet. There was a buzz in the air that’s hard to explain. It’s like we all needed to be in those rooms together more than ever to combat the chaotic state of the world around us and how challenging B2B marketing has been since the last few years.
If you missed it, you can’t go back to get that feeling. But there is still a chance to catch the keynotes and all other recorded sessions to get the amazing insights our speakers shared. Attendees who were there and our PRO members get access to all recordings! So if you’re PRO, simply log in and get to watching. Or, upgrade to a PRO membership today for access. Recordings are available through March 31, 2025.
And don’t miss out on MarketingProfs B2B Forum 2025. Registration is now open at our lowest rate!
