Prompt Engineering Fundamentals: How to Get Better Results With AI

Happy 2026. Hey, how's this for a New Year's resolution? How about you Master AI marketing? The AI Business Society gives you live training with experts who are getting real results. No hype, no fluff, just proven workflows that work. And you've got access to an incredible library of training from many of the incredible guests that you've heard on this show. So secure your AI transformation by visiting socialmediaexaminer.com AI welcome to the AI Explored.

Podcast, helping you put AI to work.

And now, here's your host, Michael Stelzner. Hello, hello, hello. Thank you so much for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzner, and this is the podcast for marketers, creators and business owners who want to know how to use AI. Today, I'm really excited to be joined by Chris Penn. We're going to explore something that I think you're going to find absolutely fascinating. We're going to talk about prompt engineering and how important it is to get great output out of your AI. Now, I can assure you, no matter what level of experience you have with using generative AI tools, you're going to find today's interview absolutely fascinating as we go through a bunch of really, really useful frameworks that you can use whenever you're using artificial intelligence tools like Claude Gemini, ChatGPT, and so on and so forth. By the way, I'm at Stelzner on Facebook, Ike Stelzner on Twitter, slash X. And if you are new to this podcast, be sure to click the Follow button on your favorite podcast player so you do not miss any of our future content. We've got some incredible talent lined up. Let's transition over to this week's interview with Chris Penn, helping you simplify your AI journey. Here is this week's expert guide. Today, I am very excited to be joined by Christopher Penn. If you don't know who Chris is, he's a data scientist and author of the book AI for Marketers. He is the co founder of Trust Insights, a consultancy that helps brands with analytics and AI. And his course is Generative AI for Marketers. Chris, welcome to the show. How you doing today?

I am doing great. Thank you for having me. It's a pleasure to always chat with you.

Well, you have been a regular on so many of the other things that we've done, but not everybody here maybe knows who you are because this is a new show. So I'd love to hear a little bit of your backstory. How'd you get into artificial intelligence? Start wherever you want to start.

So I started my career in really the early 2000s as an IT person. That's my background, that's my training. And I started this financial services startup and a funny thing happened along the way. It became marketing, right? So update the web server became update the website, you know, operate Linux, send mail became send the email newsletter, and I started doing that. And then around 2005, Google bought this piece of software called Urchin and then they then gave away for free called Google Analytics. And, and that was sort of my foray into analytics and I spent a bunch of time doing that. And then in late 2012, I left the company I was at and started working in an agency. And it was a PR agency. And one of the big challenges that they faced was we've got all this data, but how do we measure the value of pr? There's so many intangibles. And that was when I started working in machine learning and what we now call classical AI because it was the only set of tools that could do the complex calculations needed to, to say, okay, well when this happens and you get an article in Social Media examiner, for example, what is the impact of that intangibly in addition to the click to traffic? So that was really when I started working in AI, started working in the R programming language, started looking at things like Markov chain modeling, uplift modeling, etc. And then in late 2017, the PR firm went one direction and my partner and my, who's my account director at the time now my CEO Katie Robert and I, we're like, there's a there there with all this AI stuff. This was before generative AI and like, we want to do something with this because this is going to go places. And so we started Trust Insights, launched in 2018, then of course pandemic chat, GPT. All this other stuff happened since then. So it's been a kind of a wild ride. But I've been working with AI in some capacity since 2013 and I started working with generative AI in 2021 when the first functional GPT models came out, GPT3 and that GPG J6B from your Lothar AI, those are the first models that could coherently write. And that was when in 2021 we were like, this is going to go places because regular people can use this.

Were you using Jasper back then or were you using direct accounts on OpenAI their sandbox or whatever they call it?

It was the OpenAI playground. Yeah, I was using that and I was using Ulithra AI's infrastructure as well.

So you've got a background obviously on the data side of things and you've also got a background in marketing. Tell us a little bit. You decided to kind of focus a lot like obviously Generative AI you probably saw as a big opportunity. Right? You ended up writing a book. Tell us a little bit about the book. Tell us a little bit about the course.

Well, the book's out of date. We're not going to talk about the book because It's. It's from 2022.

First. It's crazy that something in 2022 is out of date already, right?

It's just a little bit. It is wildly out of date because so much has happened. I mean, that's one of the things that makes this so interesting to me is that this field changes by day. I was working on stuff last week with the Llama 3 model. I'm giving a talk tomorrow and there's a whole bunch of new stuff about a certain type of mechanism within a model that. That explains why these things misbehave sometimes. So I got to go read after. Dude, here. I'm going to get on the road. I need to actually read that paper to understand what's going on. But we made this course called Generative AI for Marketers, which you can find at Academy trusteeinsights AI to go through what marketers need to know about using AI, how this stuff works, what are the major use case categories? How does prompt engineering work at a very basic level? What are the things that can go wrong? And then what's likely, you know, coming around the corner? And that's. That's a crapshoot.

First of all, Chris has been speaking obviously at a lot of events. You're going to be speaking at Paul Raitzer's upcoming May Con conference, which I will be attending. You spoke obviously at my conference, Social Media Marketing World on AI this year.

I want to point something out real quick here, Mike. Yes, I talked about AI in marketing at Social Media Marketing World in 2016.

Really? Okay.

Yes. That's how far ahead that conference was that we were talking about AI in 2016.

Well, yeah, and we had Paul Racer there, I think, in 2018, doing a workshop on it, and nobody really understood what it was. So you are somebody who is deep in the trench on the technical side of things. I think there's going to be a fair amount of people listening today that are not nearly as technical as you are, but we're going to do a Good job. Trying to bridge some of the insights that are in your mind in a way that I think my audience can understand. I want to start with what is prompt engineering? Define it so everybody understands it. And then let's talk about why is it so important.

Got it? When it comes to AI models, this is like any other software. You program this software and except instead of writing in C or Java or Python, you're writing in English or Danish or Japanese. Prompt engineering is the process of writing code for these models to use. It just so happens that we're writing it in natural language as opposed to computer language. That's what's so important to understand about this stuff, is, yes, these things kind of can seem like they're sort of like people, but we're really, if you think about them as, hey, you're just going to sit down and write some code today in English, you will typically get slightly better results. Or the way I like to talk about it with people who are very non technical is think of these tools as the world's smartest interns. Right. New intern joins your staff. This person has 255 PhDs. They have a PhD in everything. But they don't know your company, they don't know you. They know nothing about how you do things. So when you give directions to that intern, you have to be super specific. You've got to give clear, detailed instructions. And then that super smart engine is like, cool, God, I can do that. If you were to go to an intern and say, hey, write me a blog post about social media marketing, you're going to get, okay, ambiguous, boring stuff back because you didn't give the Internet enough details.

So what I'm hearing you say is that prompt engineering, the reason this is so important for marketers who are using artificial intelligence is because the way that we essentially program the AI to do what we want it to do is with the way that we communicate with it. And instead of using programming language, which is how it used to be, now you can use everyday language. And if you get that down right, you can result in much better output. Is that what I'm hearing you say?

Exactly. And as you get better at it, your prompts get more sophisticated and you start to discover the crazy things these tools can do. A lot of people assume, oh, it's just, it's a chat, but I'll have it write me a blog post. Right? That's like buying a harrier and taking it to the grocery store. Like, yes, you can do that, but it's all it's more capable than that. So we like to tell people these tools are capable of six major categories. Generation might be a blog post summarization. Turn big data into little data extraction. Get data out of this data and do something with it. Rewriting. Turn this data into other data classification. Help me organize this pile of data. And then question answering, hey, I've got some data. I want you to answer some questions about it. So a real simple example. I was just doing some work. Just before our call, I loaded my entire inbox from this one client into a protected AI system because I don't want to be sharing confidential data. And I said, summarize our interactions so far, you know, with this client. And it gave me a nice summary of here's everything that's happened with you. And this client said, great, what things have I forgotten? So I go into from summarization to question answering, what have I forgotten to do? It's like, oh, these three things are still outstanding. And this is over a two year period of stuff. So now I can provide great service to a client just by having a tool read through all that information and pull out the stuff that maybe I forgot to do. So those six use cases are kind of like the building blocks of generative AI. You learn how to use the tools for each use case category and they start snapping them together like Legos.

I love the fact that you've gone down the path of explaining these different categories. Most people are using AI right now to generate information by asking it a very simple question. I love the fact that you're helping people understand. It can summarize existing information that you feed it. It can extract insights from data that you essential. The when we say extraction, was that the example that you just gave, was that an extraction example?

So that was a summarization example. I'll give you an extraction example. I told one of these tools recently, go to the Netflix website, go to their jobs page, extract all the open positions and put it in a bullet point format list. And it did that. And it did a great job. And it said, okay, great, now I know what Netflix is hiring for. And then I flipped the question answering. I said, we know that when a company is hiring, hiring is a big deal. It's expensive, it's time consuming, right? It's, it's very, it's not a fun process. We know a company only hires for things that they care about. So now tell me what Netflix's corporate strategy is for the next 24 months based on the positions they are hiring for and it gave me this nice example. I'll tell you a secret about these tools, Mike. They are better at processing data you give them than they are creating data on their own. So a lot of people use it. Write me a blog post, write me an ebook, whatever, that's an okay use. But they're really so good. When you give them data and say work with this data, then they don't hallucinate, they don't make things up because they have all the data in front of them.

Well, and I'll give a short example. I had done a survey of about 2,000 people in our audience and I had open ended questions in the survey and anybody who's done open ended analysis, it's extremely time consuming because you've got unlimited amount of variation. So I asked Claude was what I was using. I asked it to do an open ended analysis and look for trends and it found summaries and then I asked to dig deeper on the summaries. It would have taken me days to do that work and it did it in a matter of minutes. And this is just the example. And I know we've gone down a little rabbit trail, but it's a worthy rabbit trail because, you know, I just want people's eyes to be open to the possibilities here. I want to get into prompt engineering basics because that's really what we're here to talk about today. You have come up with a framework that is an acronym and this is going to help people do what exactly if they follow this framework.

So the prompt framework is called the race framework and it's, it's a starting point. It's a starting point for writing good prompts. I actually shared my screen here, but if you want to pull that up, RACE stands for role, action, context and execute. So let's talk through some of these pieces. One of the things that these models do really well is they do they understand who you tell them to be. So that first part is tell the model who it is. You are an expert social media marketer. You have expertise in TikTok marketing, YouTube marketing, content marketing, Twitter marketing, blah blah, blah. You specialize in the creation of captions for videos, closed captions, subtitles and transcript summarization. So these models essentially are word prediction engines. And this is the most you remember nothing else this entire episode. Remember this. The more relevant specific words you use in your prompt, the better will perform.

Real quick on the role. I love that. Go ahead, finish what you're going to say. I got to ask a question. Go ahead.

The more relevant specific words you use, the better prompt will perform. Because what's happening behind the scenes is these models are almost creating like a word cloud for every single word that you type in a prompt. And as you add more relevant words, YouTube marketing, TikTok marketing, Twitter marketing, more of these word clouds form and what it returns is basically the intersection of all those word clouds. So the more specific you are in any part of a product, the better it will perform. The role is checking the box on a whole bunch of things like this. These are the knowledge bases to load, these are the the topics to load into your short term memory. Remember, we're talking about closed captions, we're talking about subtitles, we're talking about this. And that model behind the scenes is making these associations so that it knows what to write.

Okay, so ROLE is the first R in this acronym called race. And ROLE is hey AI. I want you to take on this particular human kind of role, right? Like you are a data analyst, for example, or you are a professional copywriter. Right? So when we're describing the role, do we just tell it words that are that simple, like you are a copywriter or do we need to give it a little bit more context?

Give it some more context, tell it not only who it is, but what it's good at.

Okay.

And you might even use it. This is going to be sounding really weird. You might even use terms like you are an award winning operator. You are, you are a social media marketing world keynote speaker on this topic. Why? If you think about a post, some of these longer articles that are out there on the web, what do they all have at the bottom? They have a byline says, you know, Michael Stelzer is the founder of Social Media Examiner. If you want higher quality output, tell it to what higher quality input looks like. And that typically means things like you are an award winning so on and so forth. Because now that those word associations are happening in the background and it's going to say, okay, I remember reading associations about this and about this and about this and it pulls it all together. So that's what that role statement's really good at. It loads all these concepts and helps narrow it down a little bit to higher quality content.

Okay, cool, so we've got the R is role. What's the next part of the process?

Next was action, where you tell them all, here's what we're going to do today. In this case, I said your first task will be to write an Instagram reel caption based on the following background information and Transcripts. So we're saying this is the task, this is what we want you to do. And again, you want to be specific. You're going to be writing an Instagram reel caption as opposed to saying you're going to write something. You want to be very specific about that action statement. We want to shape it with verbs. In this section, here's what you're going to be doing.

Okay, so a couple of questions on this. What if you don't know exactly what you want it to do in the beginning? Can we say something along the lines of, your role is to guide me down a path to solving a particular problem, or does it need to be a very specific thing that you're asking it to do?

In fact, that's a great use case for these tools. You can have them say, hey, I want to have a discussion about this topic. I want you to ask me questions. We'll talk about in a second framework. But you really can have a conversation as though, again, it's the world's smartest intern. Like, hey, you know stuff about social media. You have a PhD in social media. What do you think about this idea? You know, how would you improve this idea? Or how would you critique this idea? You would absolutely do that.

Okay, so now we've got the R, which is role. We've got the A, which is action. And what's the next part of the process?

C is context. This is the background information you provide. And this can be things like, here's a whole bunch of examples of what good Instagram, in this case, good Instagram reels looks like. Or this is who this is for. But you know, for Christopher pettitrust Insights, here are the hashtags you might want to use. You should follow these rules and you might have a whole bunch of data in here. So in the example I was giving you previously about summarizing an inbox, my context was the whole inbox right here is probably close to a hundred thousand words of text. That is the context that we're giving this in the background information. Today's models, Claude 3 Opus, which is the, the paid version. Chat GPT, GPT4 Turbo, which is the new paid version. And then Google Gemini 1.5, the paid version of Google's. All of these have very large memories. And by memory, I mean how much you can put into a prompt. Claude and Gemini have a working memory of about 700,000 words. For context, this book is about 70,000 words, right? So you could put 10 of these into a prompt. That's how much they can hold.

So what does that really mean? Like, help people understand, what does that make possible?

I put two years of an inbox from my email in as part of this, right? It was a tremendous amount of data. I could put an entire piece of code, complex code in and say, debug this whole application. I can take an entire book. You know, you and I, we have the very popular newsletters. We get book pitches all the time from authors like, hey, would you review my new book? Like, sure, send it to me in a PDF format and I will have a machine read it. And you know, here are the five criteria that I used to review books. Write the book, review from these five criteria. And there's the whole text, right? It gives you the power to process hundreds of thousands of words at a time and do very complicated things. Reason, inspect, ask questions, Distillery. I'll give you another example. So in the state of Tennessee, the state of Tennessee just passed what's called the Elvis act, which is a political piece of legislation preventing basically making deep fakes illegal. And I said, okay, here's the legislation. Drop that in the model. I said, I want you to read through this and analyze it, summarize it. And did that. And they said, tell me how an unethical politician could use this legislation to further marginalize already disadvantaged peoples. And it said, hey, you know what? This act is so poorly written that an unethical politician could use it to suppress dissent. Right? You could use it to suppress dissent. I'm like, oh, I wonder if the people who wrote this know that. So that's the kind of thing is with these big, big memories. They're capable of very advanced reasoning tasks.

Okay, so this is kind of a bit of an eye opener for a lot of people because I think everyone understands giving a role to the AI and saying, this is what I want you to behave like, like a professional speaker or copywriter. And this is what I want you to do. Draft me an outline for my next presentation that I'm going to be giving at Social Media Marketing World. The context side of it, though, most people, I think, don't provide the context. They just assume that it understands the context. You're talking about providing context at a level that blows most people's mind. Is there like a generic beginner level of context? What if I don't have a book to drop into it? I mean, can I just give it like a couple of paragraphs or what's your thoughts on this?

You can, you can give it as little, as much data as you want. But the more relevant data you can provide, the better. So, for example, this is a book from years ago, Chip and Dan Heath, about how to make ideas sticky. It's called Made to Stick. Right. It's a good book. It's a fun book. A lot of the times we have this problem, and this is not just true of marketing, but in general where we read a good idea, Marcus Sheridan's they ask, you answer. You know, my book, whatever, Tamsen Webster's Red Thread and like, great idea. And then we forget about it. We never use it. One of the things you can do is you can take that context, or maybe the cliffsnodes version, or maybe the points that you remember most and say, okay, here I want to use this as part of the context. So for example, Tamsen Webster has this great book called Find the Red Thread. She did a great presentation about it. I have taken that presentation, boiled it down to essentially like a 12 or 13 page prompt, and I said, now here's the blog post, something about writing. I want you to edit my blog post and tell me how to line it up with Tamsin's framework so that I make a more compelling blog post. Because I don't remember all the pieces in that book. Think about every management book, finance book, marketing book that you've read. You're like, oh, there were such good ideas, but I forget to use them. Now, with the power of these tools and their huge memories and their ability to process complex language, any idea that you've ever had or read about, you can ask it as part of the context in this framework. Apply these ideas to my input.

That's really cool. Let's talk about filling the gap between AI potential and AI results. I'm sure you've tried ChatGPT. Are you seeing real ROI? Most marketers are barely scratching the surface of what's possible with AI. And it's because they're missing the training and the frameworks that will allow them to be very successful in their careers and in their future. AI Business World allows you to get that training. When you attend, you get two full days and 20 expert practitioners. Many of the people that you've heard on this very show that will teach you exactly what you need to know to build the workflows and to succeed as an AI enhanced marketer. It takes place April 29th and 30th in Anaheim, right alongside Social Media Marketing World. And here's the good news. You can save $300 when you register by January 16th. Head to AIbusinessWorld live right now, this discount ends soon. All right, so we've got role, which is the R. We've got A action C context. What's the E?

E stands for execute, which is do the thing. And in the execute statement, you typically specify formatting. I want you to produce a table, I want you to produce eight paragraphs, I want you to produce in markdown, or I want you to write Python code. Whatever. The thing is, this is where you specify the formatting and also reinforce specific directions. There's a direction that's really important. Like, hey, when you're doing this transcript, do not summarize it. Do not, you know, I want you to edit this transcript to remove grammar mistakes and punctuation, but do not summarize my words, which the models will tend to do. So in the execute statement, you would reiterate the most important instructions.

Now, real quick, what do you mean by do not summarize my words? Help people understand. What does that mean exactly? Because why wouldn't you want them to summarize your words?

So, for example, when I'm doing transcript cleanup, I'll take a podcast or like a show like this, and I will take the transcript from it. And there's a lot of things that I say in spoken word that are not necessarily grammatically correct. Sometimes I'll double up on a word. Well, on stop words I'll say or, you know, or. Right. And you could have a language model remove that stuff, but not change what I am saying.

Ah, okay, I see what you're saying.

If you don't specify it and give examples like, this is the right way to do it, this is the wrong way to do it. Do not do these things. Then what it will do is sort of boil it down into that, you know, that. That model specific way of speaking. So you might take a 40 minute episode of the AI Explored podcast and Chat GPT will make it all sound like ChatGPT wrote. Like, no, that's. I want you to preserve the way the speaker spoke, but I want you to just take out the. Do the technical editing.

I love it. Okay, I'm going to pick a topic and let's just work through the race system real quick on the fly. Let's say that I am trying to persuade my boss to Invest in a ChatGPT paid license. Okay, so how would we work this through the race system so that I get something that's persuasive for my boss?

So you would start saying you are an expert business. Well, let me ask you, what do you think? What are the major points that you think would persuade your boss?

Well, let's say my boss is skeptical of AI in general. And my boss basically is a tight wad and he doesn't like spending money even though it's only 20amonth. And my job is I'm just, I'm just a regular marketer working for the company. So what would we do with the role and the action in this particular case?

Okay, so in this case you're going to be an expert business consultant who it specializes in financial optimization or financial engineering or something along those lines. And you might have some additional words, you know, accounting, you know, this, that and the other thing, whatever those terms would be. Your first task is going to help write a business case to use a generative AI tool such as Chat GPT in the paid version to help our company save money. And so that's role in action context would be the background. The use of generative AI can reduce tasks that took hours down into minutes. And so you would save money by helping a person get to more. There's a study from BCG last fall, last summer saying that top end consultants, when they used AI got 12% more tasks done 25% more faster with 40% higher quality. So that would go in the context.

Okay, hold on, let's say I don't know that. Can I start by asking ChatGPT to give me the context? Do you understand? If I don't have that insight already in my brain, you could.

However, what I would suggest doing is using an AI based search engine that can find you content.

Perplexity.

AI Perplexity would be a good example. Microsoft Bing would be a good example.

So do a little research and give that research to the AI in the context window. Is that what I'm hearing you say?

Exactly, exactly. And so you say here's some context about why generative AI can help us save money. And then in the execute statement you would say now write the business case outline for the use of generative AI as a money saving tool for our company. And then you might say, you know, break down this outline into operational costs, personnel costs, etc. And it'll write the outline. There's your, you know, your one page outline. And then you would say, and this is where you'll start to repeat the execute statement. You'll say okay, maybe the outline's got five sections, you'll see now fully write out section one and it'll do it. Fully write out section two and it'll do it, and so on and so forth. So that's how I would approach that particular task.

It's really intriguing that you took a different perspective than I would have imagined. This is Chris GPT at work, if you pardon the pun. But you know, like it's a marketer talking to another marketer. But you know that marketer's predisposition is financial. So you said, hey, ChatGPT are a financial expert. Right? And you did that because this person is not a financial expert, the marketer. And by telling the role to be something that they're not, it can provide you insights you could have never imagined. Is that kind of the idea here? You don't say you're me, you say you're someone who's other than me that has expertise that I do not have. Is that kind of the idea here?

When you said the boss is a tightwad, I immediately went thinking, okay, there's three things everybody wants, right? Save money, save time, make money.

Right?

Everybody wants one or more of those. When we talk about saving money, that's really not a marketing thing. That is a finance thing. And so the topic is finances. Because these tools are word prediction engines, we want to front load it with money talk and not marketing.

I love that. And I would imagine if you wanted to, you could say, when the final output is done now translate this into language that a marketing director would understand potentially if it went too financial. Correct.

You could do that for sure. If you wanted to get super, super Powerful, export the PDF version of your boss's LinkedIn profile and then load that in to the engine and say, now rewrite this in a way that would appeal to my boss and you give it all the specifics and it will come up with stuff that is highly specific to the personality, the type of person your boss is, because all the stuff that's on your LinkedIn profile is who you are.

Fascinating. Now, isn't true that some of these tools can just go out and scrub the web, but is LinkedIn closed network? Is that part of the reason why you can just give the LinkedIn profile?

Yes, exactly. Right. LinkedIn is locked down super, super tight, but you can still copy and paste the page. If you can see it, you can copy and paste it.

Awesome. Thank you for going through this experiment with us. This is really, really helpful.

I wanted to say one more thing on that topic because I think for folks who are listening, who are trying to get a sense of this technology, that technique of providing information about who the audience is is going to make you a gazillion dollars. Think about this you are about to. Maybe you're going to. You're running the social media marketing world, planning for next year. You're like, oh, I want to have this company as a sponsor. Go on LinkedIn. Find the person who is in charge that you say, help me. Here's my normal sales pitch. Rewrite the sales pitch for this person. Right. So it's not for a Persona, it's not for an architect.

I was thinking about that. Salespeople could go crazy with this, right?

Salespeople will make a gazillion dollars if they learn to use these tools well, because you will be customizing your sales pitches for the person. Right? Not a category. That person. Michael Stelzer gets a sales pitch from Christopher Penn that is made for Michael Selzner.

That's really cool. All right, so we've introduced the fundamentals of prompt engineering with race, which is role, action, context, and execute. Now we've got this next level. Like, sometimes the framework needs a little bit more help. Right? So you've come up with another framework. Explain this other framework and exactly what it is, this para framework, what does it mean? And if there's something you want me to share on the screen for the YouTube audience, I can do that. For the audio side, Sorry, we'll just try to describe it in words.

So pair stands for prime, augment, refresh, and evaluate. And what this means is there's a set of questions that you can ask a model as part of either the initial prompt if you're just not getting good results, or. But in the. In the conversation. So prime is the first one asking a model, what do you know about this topic? As a fast way to get a lot of information into the model very, very quickly, what it does is it also covers gaps in your knowledge. Right. Because you may not know a lot about a specific field, you may know enough, but you may not know a ton about it. So let me show you an example. We're going to do this live. Says you are an events marketing expert.

And for people that can't see what Chris is doing, he's got gemini.google.com up on his screen, which is Google's large language model. Keep going, Chris.

Yep. You'll be helping me build a sales pitch for executives who justify their trip to social media Marketing World 2025 justifying the benefits of attending a professional conference. So that would be sort of that start. So a little bit of role in action. But now I'm going to say to prime the model, what do you know about this topic?

Okay.

That question is going to have Gemini reach into its understanding of what are the benefits? Why would someone go to a conference? Networking, cutting edge knowledge, skill sets, boosting company visibility, motivation and inspiration. What do you know about social media marketing? Well, it's a premier event in the social media marketing space. Scale building your sales pitch. Here are the, you know, the keyboard. So this has now primed this conversation. We've loaded all this from its long term memory into its short term memory so that we can now interact with this. So instead of me having to write, this goes back to what you were asking earlier, Mike. Instead of having to write a whole bunch of context, the model just did it for me. Now I have to go through and vet this and you know, are these things that it's saying true? Yes, they are true. Right. Social Media Market is the premier social media marketing conference. People do go to events for networking, something that is all true. But now I'm up to speed that much faster.

Okay, I want to ask a couple of questions on this. First of all, so prime, which is part of Para, Para, this acronym, which is basically allowing you to go deeper than what we've just done, right? It's like the next level.

Exactly.

When you prime it, you ask it what it knows about a specific concept. Right. And by it sharing what it knows about the concept, it is actually bringing it into its, its memory almost right from long term to short term. And as a result of it, what does that make possible? Is that make it better output? I mean, what are we talking about here?

It makes it better output because it introduces all the concepts that you're going to put and maybe some you wouldn't. So like I have gotten. And we have all got. Hey, justify your attendance at this event. Right. I don't necessarily see people talk about boosting company visibility. That is not something I see a lot of sales pitches. I see networking all the time. I see get cutting edge knowledge, the best speakers and stuff like that. I see that all the time. I don't see people talking about boosting company visibility just by attending an event. So automatically this prime concept has brought in a concept I would not have thought of. And now if I want to use this session to write that sales pitch, I'm like, huh, that might appeal to a specific kind of audience.

Got it. So then what I really hear you say is that by using this concept, we're talking about, about asking it what it knows, it's not just for the AI, it's also for the prompter.

Yes.

Right. Because you as the prompter might discover something here that might alter what you're about to do next. Is that really what I'm hearing you say?

Exactly. And it will tell you what you've forgotten. We're human. Our memories operate very differently than machine memories. And so by going through this saying, what do you know about this topic? It will remind us. Oh, yeah, that's a thing. I, you know, I should have thought of doing that.

Okay, cool. So the first part of para is prime. What's the next part?

The second question is augment. What questions do you have? So we'll continue on. This is. Okay, great. We're going to start writing this sales pitch. What questions do you have? Right. So what the Gemini model is going to do is it's going to look at parent role action, the context and the execute statement. We're writing the thing and it's gonna say, hey, I got some questions for you to help you write this pitch. What field is this company? What company size? What are the social media goals? Who's the target audience? Who are your executives? What do they like? What's the budget? All of this information is stuff that I didn't provide in the prompt, and if I had, it would have done a better job.

So just so we're clear, are you specifically asking it what questions they would like to ask you? Is that what it's.

Yes.

So what was the exact way you worded that question?

All I said was, what questions do you have? We're going to start writing sales pitch. What questions do you have? Okay, so we've given it that execute statement, but then we said, what questions you have? Sometimes we'll say, I've got no questions ready to go. Sometimes we'll say, yeah, there's a whole bunch of things I don't know that I need to know to make this happen.

What do we do with this information? Do we go ahead and answer all these questions or.

You answer all these questions.

Okay, so literally, you just start basically typing in information right into the language model so that it is better informed for whatever you're going to do next. Is that what I'm hearing you say?

Exactly. We want it to just to fill in those blanks.

Okay, so that's really intriguing. So I know you've messed with this para concept quite a bit. How big of a difference do you think it makes? Because to me, as someone who interviews people and ask questions for a professional living, it makes sense that I want to ask questions of you, Chris, because you're very smart and you know A lot of things. And the more I can narrow my question, the more you can narrow your answer. But I don't think most people think of the AIs as like a give and take, you know, back and forth, back and forth and back and forth. Is the AI smart enough to understand that you're giving it information that's somehow going to be used for some future action that it's going to take?

It is because we're loading into its short term memory, reloading to this conversation. A conversation is basically the short term memory of this AI. And so by loading up this information, it's going to know it and keep it in memory for. Mostly for the duration of the conversation.

Okay, cool. So we've learned about how to prime it by asking what it knows, and we learned how to augment it by asking it, hey, what questions do you have for me on this task? You're going to go ahead and answer those questions, presumably, right, or drop in some documents or something. Any of those are sufficient, Is that correct?

That's right. So in this example, let's say, you know what, here's my boss, right? And I'm gonna put in the text of Mike Stelzner's LinkedIn profile, okay? So that's gonna answer a lot of those questions instead of me having to type out those questions. That's very valuable information. So here's some key takeaways about Mike, right? So crafting the pitch, highlight Mike's own mission, peer level, networking, and things like that. So now we're at a point where it has some additional questions I think we can give a pass on to say, great, we're ready to start writing the sales pitch.

Just for people that aren't watching this. It did ask a couple extra questions, like what were some specific goals and also some of the past experiences that Mike or the team have had at the conference. So that's kind of fascinating that it is kind of interacting with you. So now we're getting ready to do what exactly are we executing something here?

So the software is going to write that sales pitch for to justify the visit right now I'll look at this and go, huh, maybe that this wasn't satisfactory. I'll say, I might say this didn't match my expectations. We're now going to go into the R of pair refresh. What did I forget to ask? What did I overlook that would make this pitch better? And so I know that it didn't do a great job in this example. So I would say, hey, there's some Reasons for. For this mismatch. The draft is formal. The value proposition is aligned with Mike's mission. You need some more specific examples and stuff. So by doing this, it basically creates another crop of questions that are diagnosing why its output didn't meet my expectations. Again, like, if we think of this like a human Internet, we're basically saying, hey, intern, that first draft is okay, but it missed the mark. And then instead of saying, here's how it missed the mark, I'm saying, intern, help me figure out why it missed the mark.

Okay, so just to summarize here, we've asked it what questions it has for us, we've answered that information, and then we've asked it to go ahead and do the action in this particular case. And maybe we weren't completely pleased with the output. This is where at this point we do the R in para, which is refresh. And that's where we say, hey, what.

Did I forget to ask?

What did I forget to ask of you, AI or yes, okay, tell me more about that. Help me understand. What does that mean? What did I forget to ask? I'm not sure I'm tracking.

Essentially, it means that there was something in the topic or something in the context or something that wasn't there that we both forgot about me and the AI. So I said, hey, AI, what did I forget to ask? And it went back and said, well, hey, you know, I gave you a list of questions. You gave me a really long, lengthy answer, but there are still some things missing, and these are the things that are missing.

Got it.

When you were doing stuff like writing code, this question is super valuable because, you know, when you write like a Python code, it will say, hey, you forgot to ask about error checking.

Okay?

You forgot to ask about documentation.

Got it. Well, and I would imagine, Chris, you can use this. What did I forget to ask? At any point, whenever you're prompting the AI, right, it's one thing to say, what questions do you have for me, but when you say, what did I forget to ask? It's also kind of like, what did they forget to ask? Right? I mean, as I. I'm wrapping my head around that, like, what did we forget? You can almost say, what did we forget about? In this context here, that's really what you're saying. And that's when it's going to take another analysis on what it just generated, and it's going to come back to you and say, well, we might have wanted to consider this, that or the other thing. And what do you do with that. You provide the context, and then you ask it to regenerate it again. Is that the idea?

Exactly. You say, you know what? The tone is too formal. Let's try a different, you know, a more collegial tone. And then it would go back and do that. So you're basically helping it triangulate on the things that you forgot about it forgot about, or that you just know from personal experience. This person would like. Right. If I wanted to appeal to Mike, might say, like, hey, I know that family is so important to Mike, so you got to work something in there about, like, how the. The time attending this conference is worth trading that time with his family. That's what you would cover in these refinements.

Okay, so we're talking about this para framework, which is a refinement, really process here. We started with prime, and then we went to augment, then we went to refresh. What's the last part of it?

Evaluate. Okay, so when we ask the model, hey, did you check your work? Did you do as you were told? Did you follow the instructions that you were given? And sometimes models will be like, yes, I did. And a lot of times it will say, if the output doesn't look right, oh, I forgot about that. Or I didn't do that thing. If you go back to the original race framework and the evaluate stage, you say, hey, produce a table, and the model spits out a paragraph. You're like, check your work. Did you fulfill the conditions of the prop? And we'll go, oh, yeah, you asked for a table, didn't you? Sorry, here's the table.

So I was calling it para, but it's really para, isn't it? It's P A R, E. Okay, so we're asking it. What's the prompt that we're asking it when we're in the evaluation stage? Like, what are we specifically saying here in this kind of a prompt?

We say, check your work. Did you fulfill the conditions of the prompt Fully?

Okay. And generally speaking, does it find problems?

It. Yes, especially when you have. You've been very clear in your directions. In that initial prompt, we say, here's the format I want, but here's the tone I want, or here's the language to use. And what happens is, sometimes we can give conflicting instructions. And so the model will just not know what to do. And so at that point, the more advanced models say, well, I tried, but you told me to make a table, and then you told me to make a paragraph. Which one is it?

Right. Okay. All right, well, this is fascinating because this kind of reminds me of like when I was in college, right, where one of my professors would send back the paper and say, you need to take another stab at it. Clearly you didn't read the instructions properly. You know what I mean?

Exactly.

Is there something about this that also applies to human behavior as well when we're talking to humans?

Yeah, it's funny you mentioned that. My, my CEO Katie said the pair framework in particular is also a human management framework where you say to a human, hey, what do you know about this thing? Or how skilled are you at this thing? What questions do you have? Like, that's a question we should be asking each other. Like, if you come to me, Mike, and you say, hey, I want you to speak in social Media Marketing World 2025, talk about this topic. One of the things you say is, hey, Chris, what questions do you have? Like, I don't know anything about that topic.

Well, and that's kind of how we prepared for this interview, right? I said, chris, what should we talk about that you feel like you're a domain expert in and what are some of the questions that I should be asking you? And we just kept going until we really nailed this down. Chris, this has been really instructive and I know that we're just scratching the surface of a lot of the stuff that you have available. If people want to connect with you on the socials, what's your preferred platform and if they want to learn more about your course, Generative AI for marketers and or your business, where do you want to send them?

The best, easiest place to find me is Trust Insights AI. You can get to all the socials, you get the course, you can get everything from the website. TrustInsights AI if you want, you want to get to the course directly, it's academy. Trust Insights AI will get you to the course. But yeah, I'm on most social networks at CSPenn but I, you know, just doing silly stuff in all the different places and things. I. My preferred network these days is a toss up between LinkedIn and Threads.

Very cool.

But no, if you want to get to actually useful information, go to Trust Insights AI.

Chris, thank you so much for coming on and sharing your insights with us today.

Thank you for having me.

Hey, if you missed anything, we took all the notes for you over@social mediaexaminer.com A3 by the way, be sure to follow this show so you don't miss any of our future content. And would you do me a favor, let your friends know about the podcast. You can tag me if you want to. I'm Elsner on Facebook and ikestelsner on X Twitter. And do check out all of our other shows. We've got the Social Media Marketing Podcast hosted by me, the Marketing Agency show, the Social Media Marketing Talk show, and we've just got a lot of great content to keep you occupied throughout your week. This brings us to the end of another episode of the AI Explored Podcast. I am your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day and may AI help you become more successful. The AI Explored podcast is a production.

Of Social Media Examiner.

AI Business World, early bird pricing ends on January 16th. You can save 300 on your tickets. Register at AIBusinessWorld live today.