The AI-Powered Marketing Playbook

We have got the most ambitious show that we are ever going to do. This is a culmination of years of, when I say years, two years of working with AI. We are going to try to take everyone From a novice AI user to an expert AI user using a five step framework. In 30 minutes we're going to take everyone from a novice to an expert with a five step framework. That is the goal of this session.

No pressure on us. Okay with it. We only have 30 minutes. We got to get straight into the show. You and I have done a massive amount of prep and we're here to talk prompting. Why don't you kick us off, explain to everybody there's probably people here have no idea what the heck prompting is. Give us the rundown of what prompting is.

Okay. We're going to go from most easy to hardest thing to do. So this is like the one session where we're going to actually say you can take out your phones and start doing this as we go through it to try, try it out if you have Claude on your phone, if you have OpenAI on your phone. And so one of the interesting things that you and I talked about off mic was we are two years into the AI evolution, the revolution, and this is a really large conference. How many sessions are about AI? How many, how much AI are you going to hear across this week? And the thing you want to figure out is how do I differentiate on AI, how to become a master of AI? Well, there's a small secret that maybe none of no one else has told you. And when you think about how to get started, it's really prompting, right? And I've talked to a lot of companies to try to figure out how they are going to differentiate their AI platforms, how they are going to integrate AI into their go to market, just how are they going to be better at AI? And it always comes back to prompting, prompting, prompt and prompting. And maybe we can't be as good as some of the prompt engineers that the large, large companies have employed. It's their full time job. But there's a secret to getting really good at prompting really, really fast. And this is what some of the AI influencers are using on X and these other platforms to wow you with their prompts. And the secret is Claude is a much better prompt engineer than most of us. And if you ask Claude to create a prompt for you, it will create a pretty great first attempt. So, so let me give you an example. One of the things I am trying to do right now is create a little Presentation style guide agent. We're going to talk about agents later.

We are.

And I start everything via a prompt to begin with. Can I actually do this with a prompt? And so I just go to what I would used to do is I would spend hours crafting the best prompt, going to Claude and asking Claude to do something for me using this prompt. Now I just say to Claude, yo, I want a presentation style guide. I want to be able to input PDFs and then I want you to templatize that style guide so I can recreate that style. Give me a prompt and it gives me a pretty great prompt. Then I screenshot that I go onto LinkedIn and I say, look at my great prompt. I constructed this myself and I watched my follower count go up.

Well, what's interesting about Claude, and we're talking about anthropics, we say Claude, that's Anthropics AI model. Talking about Sonnet 3.5, which is what we're using the most. And when it comes to prompting, what we're telling you is don't overthink the prompt, first of all. Second of all, be very clear on what you want and any supporting documents that's going to aid the prompt. Tomorrow I'm talking to the awesome Shannon Duffy, who's CMO of Asana on the main stage. We're kind of doing like a play on actors on actors for CMO on cmo. And I was like, all I literally did was made a PDF of my LinkedIn profile, Shannon's LinkedIn profile, and of one of the samples of the actors on actors. And I uploaded to Claude and I was like, hey, can you just make me a list of questions for this conversation? Because I think you're doing a better job than me. Can you give me like the prompt and iterated on it with me? And it did an amazing job. The other thing about prompting though, that I remember through that experience is that you, you have to go back and forth. We are used to everything being like a single shot. Like, I have to do this email perfectly this time or I have to write this copy perfectly this time. With AI and with prompting, you're iterating A lot of times, I think you and I tend to iterate five to 10 times to get a prompt really, really good before it can actually get us the output we're looking for.

Yeah, it's a speed of iteration and we'll end on this. So OpenAI came up with their latest model. It's a great brand name 01 preview and actually one of the things they just released is a new prompt guide to how to prompt that model because it's changed in how it actually interprets your asks, right? There's four core things that you have to do to prompt that model really well. And when you actually, when I actually tell you about them, they're basically how you talk to a human. They how you talk to a colleague. The first is be simple and direct in your asks. Be very simple in your directing your ass. The other thing is for anyone who's done any amount of prompting, they want you to avoid chain of reasoning. Basically we used to ask OpenAI to do something step by step by step. Now it does all of that reasoning internally so you don't have to ask AI to do that. The second, the third one is use delimiters. And so deliminators are how do I segment my section? Anyone who has done coding in the past, it's like creating an XML file. I have done coding very poorly in the past, but I can actually struggle to do that. And the fourth is basically only give it the context it needs. And that brings us on to the second thing that we want to share with you. How you be successful with AI is you have to teach it what good looks like. You have to show it what is a good example of this task being complete. I cannot emphasize that enough. AI is like a junior employee. OpenAI's latest model is like a expert level PhD. So they're getting much, much better. But you still have to onboard that employee to like what is good. How do I complete this task? And let me give you a really good example of this. So Sebastian, the CEO of Klarna Klarna have had a ton of impact using AI within their business. And he talked about the fact that their customer Support, they've integrated AI. It's equivalent to about 700 support agents today. But they did not start there. Their initial attempts to integrate AI into that function were not successful. And it was not the models that were not good enough. It was not the execution that was not good enough. It was something really boring. It was the onboarding docs for a customer support rep. They did not have great onboarding docs, so they could not teach the AI assistant what good looked like. Once they rewrote the onboarding docs for the human, the AI agents all got better because now they understood how to do that task and what good look like. I believe in the future, what we are going to see in B2B companies internally is technical writing teams who can document and process what good looks like for every singular role. Because that is actually how we're going to teach the AI agents to actually get better at the thing we're asking them to do.

So, Kieran, you and I, we built a massive inbound marketing engine together over the last 15 years. And the secrets in doing that kind of came down to not doing the big, flashy, sexy stuff. It was always doing the really important, shitty, boring stuff. Like that is the secret. It. If you want a secret, that is the secret. And what we're telling you when it comes to AI, yes, the models are cool, yes, the technology is cool. The number one thing that's going to have you fail is that you're not clear on the work you're trying to do. And when you force yourself to document it in real clarity, one, you get clarity on the work you're doing. And then your employees, your humans and your AI can both go with that clarity. At HubSpot, you and I were just talking last night at dinner. We're going to build a substantial technical writing team over the next few months to just do documentation, marketing, sales and customer success. Use cases like that is what we're going to do. The more we do that, the more we're going to understand the opportunities where AI can really provide value and scale and automate for us. We're going to force some clarity that we probably had some ambiguity on, which is a big problem we want to. We want to figure out. And three is like, we will get very clear on what humans should be doing and what AI should be doing. Right?

Yeah. I had a. I remember having dinner with a marketing leader and they were like, well, they were, they were advocating that AI is not that great because they wanted to feel safe in all of the work that was really valuable that they did. And they talked about the fact that AI, AI could never do strategy. AI could never do marketing strategy. It couldn't do what I do. I've tried it, I've prompted it. And I ask, I asked them to show me, like, how did you teach the AI what a good marketing strategy was? Well, I typed into the box, do a good marketing strategy for this product. Usually your results with AI are a reflection of you as a manager, of how you are managing that AI assistant. And so if you do not set context, if you do not teach the AI how to do a role, then it is not going to be a successful thing. Most of the upside that I have got from AI in my writing, in how I create content, and a lot of the things that I do each and every day has been because I have spent weeks, and I mean like weeks and weeks and weeks on teaching AI my domain expertise. And I think what I love about AI is it's going to be a boom for people who really understand their craft, have deep domain expertise, and then can teach that to the assistant to replicate that. And that comes on to my most favorite tip that we are going to get.

This is number three you have. This is your pet rock. This is your thing that like, if you could have everybody do one thing, the third thing would be it. Mine would be documentation, I think candidly, because I think it is so important to drive clarity and actually help you scale once you've figured out something. But what is the third step in our framework that you're like, you gotta go and do mine.

So look at where we've come from. We've said, okay, we can do some basic prompting now. Now we can actually onboard the AI system. We've taught it what good looks like, we have onboarding docs. What I am fascinated about AI is its ability to templatize. Why something is so good, right? Why something works. Someone laughed at me in my LinkedIn comments and trolled me. I get trolled a lot.

This is like, you are so sensitive to trolls.

No, I actually love trolls.

Let's be real.

If I were sensitive to trolls, I would leave the Internet.

But you hate trolls.

I would leave the Internet. Am I actually. But someone said, hey, I made the point. One of the things I believe Claude can do is take something like Olgavy's ads and templatize that ad and basically be able to replicate that ad. And someone said, you're crazy. AI could never get to the human nuance of why that thing is really good. It can. Spoiler alert, I've done it. One of the things I have been doing a lot of is trying to templatize different things. And so one of the places I started was I'm fascinated by why certain writers are successful, why certain creators are successful. Let us give an example. We had an incredible creator come on the show, Jenny Hoyas. She talked about her process to do YouTube shorts within a four minute window. She gave the best, and I mean.

The best, the best by a mile.

Articulation of someone who truly understood her craft. She broke down a 30 second YouTube video. She did a second by second of why every part of that video was in there, from the aesthetics to the.

Talk track, everything, the hook, the visual she's using, why she is Leading folks to the conclusion how to get the human psychology, the human psychology so much.

And so when she's telling me this Mr. Beast actually just released a similar thing for you, YouTube where he did a minute by minute, when I'm listening to this, I'm thinking, I'm going to templatize.

This, I'm going to steal your ideas.

So what I think about AI and I'm going to give you some examples here, it is not a copy and paste tool. It is an accelerated learning tool. And I really want that to be the takeaway from this third templates. Because when I've talked about templates publicly, everyone comes at me that it's a copy and paste tactic, then I am stealing someone else's work. Think about how humans learn. I would watch that video from Jenny and then I would try to replicate what she has told me. But now I can just put her YouTube video into a transcript, get the transcript, go to Claude, tell her to pull out that four minute section and to create a template. And the way I ask it to create the template is I say create this template so an AI assistant could replicate this. And what it will put in the template is pro prompts for itself to replicate that exact style. And anytime I want to recreate Jenny's style, I can say create a YouTube short on this thing and upload that template and it will recreate it perfectly. Now what I would do again accelerated learning is edit the template to your taste, edit the template to your style. It is just accelerated learning. I have just accelerated my learning in terms of how I can replicate what someone has taught me and then make it my own by adding my own unique style.

You're talking about building templates to more quickly reverse engineer best practices, right?

Exactly.

You want to find the best experts, use AI to help you reverse engineer what they're doing very quickly. And then you are asking AI basically build me a prompt so that I can do this at scale instead of doing this as a one off provider project. I can do this every day if it's a social post or multiple times a week if it's a LinkedIn video. And that's where the magic is, because I've talked to you a lot and you have a whole file of these templates now. And depending on the the work you're trying to do, you pull in different templates, create new prompts and that you have a new workflow. But that's not possible without the prompting knowledge. It's not possible without the documentation and understanding of what you're trying to do, and then it's not possible without taking the true expertise and reverse engineering it.

Right. And so I think the one place I would look at across your business is where I started was I wanted to get better at different writing styles. Like, I wanted to be wittier. I wanted to be not more direct. You're witty. I definitely shouldn't be more direct. I should tone down my directness.

That's true.

And so what I basically started to look at was I catalog. I basically started to take creators. I categorized them into different topical styles. I said, these creators represent this style. These creators represent this style. These creators represent this style. I took a ton of their content. One of the cool things you can do is YouTube is a treasure trove of content and they actually have the transcripts in the content. Garmesh talked this morning about agent AI. There's an agent that I built on there that actually takes the transcripts from the video and turn them, in short, from video. But basically you can take a. You can get a creator's content bank, whether that's YouTube, whether that's SlideShare, which is another treasure trove of content that people are not using, whether that is LinkedIn posts, whether that is X, Twitter threads, whether that is whatever it may be. You can categorize all of those creators into different styles and then teach Claude what that style is. And Claude will actually produce a template of the key ways to actually start to replicate that. So write in is like one of the ways you could do this. I'll give you another one that's similar to this that's really interesting for people who are trying to work to get more meetings booked for their team. One of the interesting things you can do is you can start to look at how do people open and interact with emails. How do people book meetings with your sales team if you were doing outreach, not by the typical data analytics, but. But by writing styles. I can have a range of different writing styles that Genai can use to create outreach emails. Whether they're witty, whether they're direct, whether they're informational, whether they're controversial. Spicy take. Spicy take might not be the best one to use in a sales outreach email. Or maybe it is because every other email I get from outreach is like, pretty bland. And then you can start to look at the data to see how it looks in terms of what style is performing best to get meetings booked with your reps. So templates for me are a way that you can accelerate your learning across all parts of Your customer journey.

Well, what I loved about what you just said before we move on is I don't know about you all, but I have had endless debates with people who are like, I know the customer, I know the customer likes this tone, or they only like this language. And it's like, that's like a very subjective thing. Now you can just test, right? Like, you can say, hey, we have these four different types of tone, voice, everything that we think our customers like. And we're just going to test different iterations of them and see what is actually true. And it's actually pretty hard to scale that with normal kind of like one or two people doing email marketing, for example, or a team of sales reps who might be doing a bunch of different stuff. But when you can templatize all of that, you can actually run very structured tasks, tests on unstructured information and stuff, which is, I think, pretty massive. And one of the reasons, I think templates are going to be a huge part of how you scale with AI.

Yeah, I think unstructured to structured data. Yamini had it this morning. Dharmesh had it this morning. It is one of the best use cases of AI. I want to check in to see how we're doing. We are really excited to be in the creator stage. One of the things that happened last year we were actually on the main stage again this year is we were on the main stage. It was my first time being on the main stage page. And for anyone who was there getting to watch you might not. The video is online. I spent the entire time fascinated by myself on the big screen, yourself on this. So I feel like this, I'm starting to this. I just want to do them better because there's no screens. But so we've got our. We've got the prompting. Anyone tried out a prompt yet?

Prompt in the audience? We got some folks raising some hands.

Got some prompts going. Please tweet us, send us your prompts, let us know what Claude has done for you. We've got our contact set in. We've got my favorite accelerated learning not copy and paste tool. Don't come at me on LinkedIn when I show you my templates. Okay, we should get into. Okay, again, very linear process here. Right? Going from hard or easy to hard prompting, onboarding and templates to agents.

Agents. We heard a lot of talk about agents on the stage this morning.

We've heard. So agent is basically when I've actually figured out how to do something via a prompt. The logical next step is Actually to try to integrate that into an agent. And so recently I built an agent on. I was very fortunate. I got access to Dharmesh's platform early agent builder. I took one of my prompts, which was the ability to take a YouTube video, take the transcript, categorize it into the short form content that does really well. Let me give you those categories right now because this took me a long time to actually learn. There are three types of short form content that do really well on the Internet.

Hold on. Tell everybody. Kieran is trying to get LinkedIn famous. So follow him on LinkedIn. He'd make him, make him, make him really happy. And so you are trying, you basically tried to build an agent to help you do better on LinkedIn. That's what you're trying to do.

I have built an agent. My impressions on LinkedIn have come from the hundreds of thousands, thousands a month to multi million a month. Because I have figured out how to use agents as part of my process. And so I taught this agent. There's three types of short form content that do really well. There is a clear and concise educational lesson. You have to have one lesson. You cannot have multiple lessons in short form content. People get distracted. They will not reshare it. There is this spicy take which I do enjoy but does get you trolled, which is you want to have counterintuitive arguments. You want to say things that are the opposite from what most people believe. And then there is the head nod. The head nod is really one of the best ones that I've articulated and taught the assistant. The head nod is everyone kind of believes this, but no one has been able to articulate it. A really good example of that is Paul Graham's founder mode. Most people actually believe that, but he articulated it in a way that resonated with people. And so they're the three types of content. And so I basically, the agent would categorize all of the YouTube transcript into one of these categories. And then it would take another thing I taught it, which is there are five types of posts that do really well on LinkedIn. There are five styles. Then you can pick the post style. You apply the short form category and it will create a first draft that is an example of an agent. I believe in the future every single team is going to be made up of humans and virtual agents, agents. And those agents will have email addresses, those agents will have Slack handles, those agents will be in Google Docs commenting. I think that's the future that we are going in. And I really Believe that for our audience who want to be ahead of the curve, starting to play around with how you can integrate agents into your own team and work is a really good use of your time.

I think the LinkedIn agent that Kieran built and you released it, and I think like tens of thousands of people.

And then I took it back, and.

Then, then you selfishly took it back. So there's a little bit of backlash.

Took it back.

You're trying to tweak it, make it better. I think it's a good example, though, that AI is much better than we think. And sometimes when it comes to artificial intelligence, I think, yes, it has a lot of limitations. I think sometimes those limitations are us. Like, I even find myself being like, oh, I can't ask it this. And then I'm like, why can't I ask it that? I should, like, I should try to do this. I should at least see if it's possible. And what you have shown is that, hey, you can be very aspirational and you can say, hey, I want to go from hundreds of thousands of LinkedIn impressions to millions of impressions. Can I use AI to do that? And then you basically told us how you. You figured out prompting, you figured out. You. You basically went and researched who was doing the best job, then you templatized it, and then you build an agent so that you could do it really quickly and really repeatedly to get the result.

Exactly. And there's multiple. So that there's two others that I'm working on. To give you an example of one of the things I. The reason I talk about AI a lot, and just so you know, you might see me a lot at this conference talking about AI, because I signed up to a lot of sessions and they're all on AI. But one of the things I believe about AI for, For. For marketers is, and just people in general is it is the best unlock for creativity, which is counterintuitive. Spicy take to what most people are think of AI or worry about AI, that it kills creativity because it's doing create creative roles. I actually think it unlocks the creativity within us because you now can do much more with an incredible assistant that you can teach how to do the things that you do. Two other agents that I'm working on, I kind of mentioned it earlier. One of the things I love to think about and I would encourage everyone to think about is where is there a treasure trove of data and learning that people may not be using or getting less leverage from? SlideShare is a great example. It is. I don't know if people remember. It used to be a really cool place to go and get a ton of data and a ton of content. It's still there. And so one of the things I want to do is have an agent that can take PDFs from SlideShare and turn them into incredible content and add some modern twists. The other one that I'm trying to work on right now is a YouTube agent. Because there's two sources of incredible learning, accelerated learning, learning that I can use. There's a Jenny Hoyas template, and Mr. Beast just released a template for his YouTube videos which gives a minute by minute breakdown. Again, this is incredible. When you see that, your eyes should light up because now you have something that you can teach the agent and replicate at scale. And that is the way to think about these things. It's like, where can I get accelerated? Where can I get upside from accelerated learning? And how do we add these to the team and to get. Get more productive, more efficient, more upside.

You love agents. I love build and use more agents than anybody.

I was building a chief of staff agent last night for fun. That's what I.

You need. You need to get a life.

I do need to get everyone. Was that. So I got invited to go to like a pro Jam concert. Did anyone go to the pro jam concert last night?

No.

Yeah. I should have went. I should have went. I didn't realize it started. It ended at 10pm I'm like, I can't be outweighed. And so I was like, should I go to Pearl Jam or should I create another agent?

And the answer was obvious.

And that's why my personal life is a mess.

It's true. Okay, the last thing we want to leave everybody with, the fifth thing is AI coding. It's apps. And this to me is probably where I have felt the most personal transformation with AI. I think two weeks ago, there's a company called Replit. Great company. Assad is the founder there. He worked at Facebook for a long time and he basically wanted to build a product to make coding accessible to everyone. And what was wild is two years ago he posted on Twitter, he was like, soon you will be. Within years, you will be able to just type in what you want to build in your. Your native language, and AI will write and build that for you. And about two years to the day, they released their coding agent. For $130 for the year, you can have access to an agent that builds apps for you, which is for me, somebody who didn't study computer science. I have a degree in journalism. It's great. I can write. It's awesome. Claude's going to. Going to make me obsolete in a lot of my writing stuff. I want to be able to take my ideas to actual fruition. That's where apps and AI coding is pretty transformational. So immediately when it came out, we got on, we were doing a show. We built a workout app for us to do, like hit workouts in replit in literally like three minutes during the.

Show, on the intro. Just going to. It's also great. It's also great for people who did computer science and couldn't code.

True. But what I learned from that is that agents are really good at automating very specific processes. You kind of graduate to needing an app when you need lots of different data sources. What you're trying to do gets very complex. And AI coding demystifies a lot of that. What I'm working on right now is building an app that takes the S&P 500, takes all the publicly available marketing data that you can get from APIs, predominantly the similar web APIs. I think they have some of the best available marketing data. I want to build an algorithm of that marketing data and rank the S&P 500 based on how good they are at marketing. So I can basically build a marketing index fund and I can then see what this look and see what the growth of those companies that have better marketing scores looks like over time to, like, understand how does marketing really impact, like, our most successful companies in the world. And what's crazy is that I could have never done that before. I could. I wouldn't even thought about doing that before. Now it's building all of this. It's asking me to just put in API keys, go get API keys, things I would have never thought I was doing. And it was so incredibly straightforward and simple. I had one other revelation in doing that, Kieran. I was doing this and the AI was picking all the technology, it was picking the languages, the app was built in, it was going to pick out how it was going to be hosted and deployed on the Internet, all of these things. And it kind of hit me that we're moving from a word of mouth to a word of machine world where AI is going to direct a lot of the adoption and buying decisions in the future. And that really came through in using the replit agent.

Right. And there's two quick things because we're going to run out of time. The other cool thing that you did was you can give replit the API and ask it to suggest apps I should build, which is like, really cool. I believe marketers and all of us are going to be much more point solution driven. So we're going to find areas in our, in our customer journey and be able to build point solutions to solve those problems. I could give an example of Churn, but we're going to run out of time. I do want to end with this, with this framing for AI, because I think it's my favorite framing, which is from Kevin Scott, the CTO of Microsoft. He talked about the fact that the best way to think about AI is it makes the impossible problems just really hard. And every time there's a model update, there's another batch of things that used to be impossible, that now we're just really hard. And I think for everyone, doing a couple of the hard, hard things is actually remote because most people will not do the hard thing. And so the takeaway on that fifth in terms of apps is the way I would look at that across my business is what are the things that used to be impossible now with AI are just really hard? And what's the one, what's the one that we should take on that's truly impactful? And that's the fifth step. We went from prompting context setting, which is onboarding. We went to templates, we went to agents, we went to apps, and we just did it in 4, 3, 2, 1.

30 minutes exactly. We covered a lot of ground really quickly. Thank you all for coming out watching today's show. Thanks. We'll see you again on marketing against the Grain.