Dan (6:20)

Show me for sure. I'll start with my background so I've been in startup life since getting out of college, mostly in operations roles. So started in marketing ops. Got a Rev Ops title was. I think I was doing RevOps while I had a marketing ops title for a bit there. Got about 10 years of like in house experience at earlier startups doing that. And then recently I actually started consulting on my own. And then I noticed I wasn't good at consulting on my own, mostly around sending invoices and stuff like that, all the operational stuff. So I'm like, oh, maybe I should check out working for an agency. So I actually found my current job at CGM through Exit 5, and now I'm doing GTM engineering there. And it was a good marriage between, like my rev Ops background and as well as I have a technical background, computer science. And when I was consulting, I ended up building a HubSpot app. So I was like, oh, I can kind of combine both of those worlds and get into GTM engineering and start figuring out what that. What that is. Which we discussed on our webinar last time, right?

Owen (7:16)

Yes.

Dave Gerhardt (7:16)

Do you feel what I said in the intro, like, do you feel like there's a gap between what people are talking about and then, like, the ways that we're actually using these tools? Or I find myself and this is, this is on me also too. But, like, man, am I the king of, like, getting things half baked and then never seeing them through with AI.

Dan (7:32)

Yeah, yeah. No, when you were saying that, I was thinking of like the LinkedIn post where it's like I replaced myself with AI and it has like a workflow with a hundred different nodes. And my examples I'm going to bring aren't that. But I was like, it would have been funny if I brought up one of those, like a screenshot of one of those and be like, this is what I'm going to walk through today.

Dave Gerhardt (7:49)

But okay, even just an example, like, everybody's going nuts because of, you know, Sora to, like, the new. My daughter's home sick. We got norovirus running through our home the last two weeks. She's. I'm sitting with her on the couch, she's watching a movie. I'm like, I want to make a Sora video to promote our New York City event. I couldn't even do that right? And so I'm like, the gap between what I'm seeing.

Dan (8:09)

People are like, check out this video. We're all so cooked.

Dave Gerhardt (8:11)

AI is going to take our jobs. And then I'm like, I like to think I'm A pretty good writer. And I think I know marketing, like, I was not able to effectively make a video without someone's, you know, hands messed up and the writing all messed up. So just, I think that's an important lens for sure.

Dan (8:24)

And I think some of these workflows and some of the topics I want to like shine a light on is like, there are certain tricks you can do to get better outputs. And if you're building something for the first time, you're going to have to work at it a few times, I guess, get some practice reps in and follow these guidelines and you will get better outputs that you can actually use. In my example, still a zero to one. Like, I still want a human to look over the output and like actually like refine it, right? I like to think of it as like, you sent an intern off to do it and then they brought you back all this research and you can like figure you can optimize it from there, except for it's instant or nearly instant, I should say. And they get really good if you have good inputs and output examples. The first example I'm going to walk through is going to be a clay example, everyone's favorite clay here. And just to give a little context of the scenario we're in here, I think I use exit 5 as the example 1 for this. But working at a company where webinars are driving a lot of pipeline, right? And we're doing one a month, we want to do more, right? So we want to do two or three webinars, but we're scared that that will eat up our small team's bandwidth and we won't be able to experiment, find new channels that work as we're supposed to do as marketers, right? So this is a good example of how you would build AI to help enable doing more webinars without eating up more bandwidth there. And actually I want to start kind of from the output. So here's kind of like what this whole workflow will bring out or will output is it's going to show us Google Docs that pitches the webinar title, right? You know, has a description, you know, has a potential agenda, has summaries for each of the guests that we have on it, has some host research where the host can go through and be like, okay, I want to read up on these three or four or five, I guess, things here. So I know what I'm talking about when I'm hosting this webinar. And then email sequence for this one. I only have one email in it. But you can imagine that there would be three or four before the webinar, three or four for after. If you're getting really fancy, you can have different sequences for different Personas that will attend the webinar. But that's kind of where the output of the workflow is. So let's go back into the workflow here. And as you can see, it only has a handful of nodes here, rather than the hundred that we were joking around about. But I like to make these nodes do, like, one thing at a time. It's really easy for me to understand it that way. Right. So the first one is going to be the ideation one. So here I'm just dropping in a topic that I want to do for a webinar. So GTM engineering is what I put in, and you could actually do a whole workflow around coming up with this topic. I think Owen and Air Ops has one where they actually put in the guest first, and it comes up with potential ideas for topics. But for this one, we'll just assume that we're starting off with an idea from there. The only thing this part of the table does is I wrote a script or a prompt here that says, hey, here's who we are. Exit is exit 5. Here's how you know our voice is. Here are audience pain points and themes that resonate with them. Right. And so for the task here, we want to have you look at the topic that I put in and pitch some different ideas for webinar titles. Right? And then instead of putting them in different columns, having five different columns hitting the spacebar there, we'll ignore that. I have it put it into different rows in another table. So here are the five that it came up with for me. And the good thing here is I can come in here and edit any of these. So if I'm like, hey, I like this one, but I want it to be marketing ops instead of revops, I can do that. Maybe this one gives me an idea and I just rewrite the whole thing. You can do that or you can go back and have it run again. But from there, now we have the title of the webinar and I have it. So you can select the one you want and it will send it to the next table. So you can ignore these four. I want to work with this one. The next point I have is I want to grab the guests or like, I'll actually Input in the LinkedIn URL of the guests that I want to have on. And then what it does is it does some research on the guest. So I use Owen and myself in this and then I have it write a guest summary. So you can see, again, this is where it gets really important to have the context. It has less context than the last one. But having good context and then having good example input outputs, this is really where you're going to actually have like, or where you should spend most of your time. Having more than one example. I think this one's only one because it's a demo. But having more than one example is really going to help you out. And just saying, hey, here's what we're going to give you for this prompt, which I gave the webinar name and the guest information, right? And here's the kind of output we want to work for. If you want it to actually be more bullet points, you should show that the more you tweak with this, the better the output and the closer it will be to, like, that final destination that you want for each of these pieces. Right? So I do that for both Owen and myself and then have it go to the next step. You can actually have it sent to the next step once this is filled in. But I like to have a little checkbox because I feel like I'm in more control here. The next one it has is topics again, and you can see it kind of repeating, right? So I pull in a lot of the context here. I pass in the information that we have from the past ones that we're passing through the tables here. I kind of collapse them here so it doesn't take up a lot of real estate, but they're still there. I'm passing it through. I want it to output for me both an agenda in here and also I have it output a summary again, a little bit more of the same. I think if there's one takeaway I kind of want people to focus on in work like this is you want to have good input output examples, and you want to use a knowledge base, right? So, like, hey, here's the tone of our company, here's how we like to write, here's who we are, here's information about our audience that we're going for and being able to reuse those is big. I really like how actually airopt has that set up. I can show you they are working on it with Clay, where they're going to have these beta document contexts. But I really like. And you'll see in Owen's examples how airops kind of handles that. But having reasonable pieces for that will actually get you up and running a lot quicker and you can use it over across multiple ones.

Dave Gerhardt (13:40)

Question in the chat from Beth is how is this actually a quicker way of developing webinar content and distribute. Sorry if I'm missing the point. Great objection. How would you answer that?

Dan (13:49)

Yeah, I think it allows you to from start to finish in this after I build it. So the investment time's upfront building it and it allows you to do all the repeatable stuff quicker each other set this up.

Dave Gerhardt (14:02)

You're going to build this once there is. What you're showing right now is a lot of work and it is effort.

Dan (14:07)

Yeah.

Dave Gerhardt (14:07)

If you build this machine, then you can then pass that off to anyone across your team to then plug and play and it from that point on will be more faster and repeatable for sure.

Dan (14:16)

And this is for again, going back to the example. It's for a company that already knows webinars are working and they only do one a month and they want to do more without eating up their bandwidth. Right. And so I wouldn't recommend you build this out the first. I would do webinars without this workflow and make sure they work right. Like, I don't want to spend all this time building this up because it is an investment. But then once you have this built up, you can start cranking out webinars and you're looking for reusable stuff. So this stuff in here, right? Like, I did some testing trying to generate an image, and I felt like you, Dave, when you were talking about sorrow, right? Like, I was trying really hard to get, like, pull in a picture of me and Owen and use a template that I grabbed from the website, right? And I'm like, okay, let me try to generate, you know, an image for it. And it didn't really work out. And I'm sure there's somebody who's better than me at image generation stuff that would actually get it to work.

Dave Gerhardt (14:58)

This is exactly like the goal of doing these things. So, like, even already that question is like, okay, oh, I get it. And so my brain is thinking like, man, we do the podcast every week, and every week we write podcast titles and descriptions and we do the newsletter, we do the webinar. There are all these things inside of the company that, okay, yeah, this could be doing this. Megan says in chat. How do you do this if you don't have access to clay or custom projects in GPT? My answer would be, well, get those things. I don't know why that's a blocker. Neither of them are super cost prohibitive.

Dan (15:27)

But yeah, Clay has a great on ramp too. Pricing wise too. Like you, you don't have to get up to the top one. My next example is going to be Zapier Agent, which has a really cool free tier. It's a different UI and you can't do exactly this in it. But there are other tools out there for sure too that are worth testing around. So the email sequence part, this one is where it gets a little different, where I'm passing all the context. I like to break up emails into subject and body. I've seen people break it up even further where it's like body paragraph 1 list in the body, closing body, right? Like the more you break it up, I think the more you can kind of control what's going on there. I like to separate the subject out because I like to pass it into the body content as context. But again, more of the same, right? I give a context, I give examples. At the end it's going to come up with a better one and you can kind of build it. So like, maybe it actually comes up on the subject line and you just keep hitting play until you find one you like or you can update it. So there's ways you can do it. I have it, just build it and then send it over to the body. And then the body one does the same thing. And then I'm going to skip over the host research, but it's more of the same where, you know, you pass in all the information and it goes online and finds the stuff there. But the final step is it takes all this information and it puts it in the Google Doc. I do have one step in between where I actually pass in all the information. And then I say, hey, put this, format it in markdown for this step. Because I was lazy over here and didn't ask it for markdown. So I realized that at the end and I added this step to like kind of clean it up and that's how we got this final output here.

Dave Gerhardt (16:49)

1, 1 more, 1 more thing on this. Somebody said, why use clay instead of this? Instead of making a custom GPT, can you explain the difference?

Dan (16:56)

Yeah, for sure. And I think this kind of ties into what we were talking about. Like last time I was on where I was like, oh, I'm really into these custom GPTs, right? But if I'm doing the same thing in six months, I feel like I've failed as a GTM engineer. And I think the benefit of this is you can control it step by step. And actually we Were just talking backstage. OpenAI just released a new tool that's more of like an agent GUI kind of similar to Clay. So that's something you could check out as well. I was going to look at it last night, but I was working on this and it allows you to like, jump in. Right. So like, I can say, hey, I want to come up with the ideas, but I want to be able to choose the idea that we end up going with. You can do that the same for the subject lines and you get to control each step of the process. And that, I think, comes out with a better output at the end. A lot of people I see when they try to do big things like, hey, design a whole webinar for me in custom GPT, like the output doesn't get to where they want to be and it's really hard to really refine it without having user input along the way. Cool.

Dave Gerhardt (17:53)

All right, let's bring up Owen. Dan, we'll see you in a minute.

Dan (17:55)

Awesome.

Dave Gerhardt (17:55)

So many questions about the tools. Like, how do you do this without the tools? I said, well, that's. You need to use the tools.

Owen (18:01)

Yeah. I actually had two points on that really quickly, which was like Dan's point there. Try something before you go and build a ton of automation around it. So like, like the webinar system, we do webinars every week. We did probably 15 before we did proper automation around it. And then to people's questions around tool selection. Sometimes you are just confined with what you have access to. Then you know where to start. Yes. Not different tools will give you like different elements. So, like what Dan mentioned as well about ChatGPT's agents, they're going to be really good if you want diversity and like talking to lots of different APIs. Probably not great for sales enrichment and other like GTM workflows like Dan was going through. Then you're kind of figuring out like, is this something I want to invest in long term? Do I want to go deep? Can I find a particular tool? So that there's my hot take to come in just on the last.

Dan (18:47)

No, it's great.

Dave Gerhardt (18:48)

Dude, that's. That's a great way to say this. And I do see some of the questions in the chat, like, okay, someone said clay is cost prohibitive. Maybe for you, maybe for your company. It is. But I think let's just take a step back and talk about marketing principles. And with all of these things, right? These are all guardrails. It's like hire a players. Okay, well, I'm not A venture backed company. I don't have budget. I can't just go and spend $200,000 on a content person and hire someone great. That's a guardrail for me. So maybe I don't have the budget. But also most things in marketing, if you can prove out a use case, then you can make the case for budget. So if I can do this all, like if I can manually send tons of cold email and I find cold email works, the next step would be then like now I can make the case to justify a tool to do this so I could do more and prove out better results. So I love that you came in with some principles, but I'm excited. I hear you got some good stuff. I'm excited to see what you got.

Owen (19:38)

Yeah, let's do it. So two examples I'll give today. We'll go through one at a time, but one is content creation. How should you do it today given everything we know about SEO and AI search? And the second one's going to be content refresh, which kind of like that conversation we just had. You've put loads of time into creating content in the past, now you can get extra value out of it. So just when it comes to like the principles, like you said Dave, the best and like most authentic content, again, before we get into any automation, is what you can create is going to come from original research and the opinions that you have internally. So not only is that what your audience wants, but nowadays it's what AI search and SEO is actually going to reward. So what they do reward, in short, is new context, new points of view that add to the conversation. So my example here, I'm using Carta as my dummy account. I am going to take both qualitative and quantitative insights and then turn that into a list of prioritized bottom of funnel content that I should go and create in advance of a new launch. So what I have here again might not be available to everyone, but you can find your proxy to it. I'm looking at given Carta care about performing in AI search, looking at what are the gaps they have on questions where they would want to get mentioned or cited, but also what are the gaps that their competitors are not like getting mentioned and cited in either. So this is like blue ocean relevant to their company, but they're not appearing in searches. And then secondly, kind of like what Dan was doing, like I'm relying on a knowledge base, I'm bringing in a ton of sample sales calls. It's purely the raw transcript and then mixed in there is some metadata which just gives richer context. So context is going to be key to all of this. And what I come out with at the end is a prioritized list of topic ideas with the target audience in mind. I kept it pretty straightforward. And then kind of running left to right, we're able then to perform other pieces of automation just like Dan had. I'm only going to go through one or two in detail, but we'll look at. Great. Now that I have an idea, is there traffic behind it? How competitive is it? And I can go left to right in terms of creating a brief that again, I human review. All of my automation that I go through today is focused on putting the human in charge of higher leverage tasks, not trying to just like go end to end. Hey, generate an article for me. Like that is.

Dave Gerhardt (22:02)

Yeah, I love this because there's so much with, you know, SEO aeo and it's like, look, I'm just going to make the content, you tell me. And it's like, boom. Here are three examples where our competitors don't have a point of view around this topic. Great. Let's start there. Love that.

Owen (22:14)

Yeah. So let's go under the hood. I want to go from having here's where the gaps exist. Here's where my sales team have literally been on conversations where gaps have existed. Maybe the rep wasn't comfortable answering something. So now let's go and figure out what are the best ones for us to go and prioritize. Because even with tools, we only have so many calories that we can spend. So end to end here, what I'm looking at is structurally, let's bring in the context. Then I need to do some preparation of it just so it's in the right format. And what I want to do here, and I'll click into each of these in a second. I want to analyze our sales conversations. I then want to overlay that versus kind of the quantitative data. So with what's coming up on sales calls, does that map where we have opportunities and then we'll generate an output in the end which relates back to. Okay, this is the summarized list that we should go and focus on for the hypothetical upcoming launch. So if I go a little bit under the hood, what we're doing, very straightforward, just pulling those examples in first, like where the gaps exist. Important just like again, always consider process. And what I do here is, it's Python code, but I'm pulling in. Let's just calculate what was the date 90 days ago and why is that? Relevant because when I go to pull the sales transcripts, a call that maybe Dave had with Carta two years ago is no longer really relevant. I want to pull in what's the freshest content that we have available. So I'm saying pull all of our sales transcripts, but from the last 90 days. So that's going to be my richest context. Then I need to prepare some of it. So just before we pass it into a model, here's everything I'm feeding it. I'm feeding it very clean questions. Again, this is where the opportunities exist. And then it is clean to an agent or it's clean to a computer. But here are all my sales conversations and it goes down for many, many rows. Then something I'm passionate. If we have time later, Dave, we should maybe dive into it of like system prompt versus user prompt and then user assistant pairs. Probably the most important part of prompt engineering that no one knows anything about. But here in my system prompt, I'm saying, okay, you as an LLM, you are amazing at analyzing sales conversations. And I need it then to dig in and figure out what are the explicit questions, but also the implicit information that's coming up in these conversations. Like what Dan mentioned. We'll say Carta plays in the financial space. An AE at NerdWallet might have a very similar conversation to one at Carta, but the context is completely separate. One is B2B, one is B2C. One is probably looking at a lower ACV than the other. So what is important within your user prompt is you give it the richest context possible. That doesn't mean everything, but it means what's like, pertinent for it to know at that moment in time. So it's kind of obfuscated behind each of these little purple pieces of text. But we're feeding in like Carta's brand case. So it's about the brand, who is their ideal customer profile. Because these sales conversations might be generic enough in their, like, actual transcription. But we want to make sure that we're feeding it with all of the background context people come into calls with. So in short, we're feeding it brand context, the sales transcripts, and then we're saying, here's the format. So what does that look like in practicality? Here's the output. And if you were to actually just stop here, we've only got three or four pieces into it. This is amazing. If I was to give this to my head of content, here is 10 questions that have recently come up on sales conversations for which we don't have content. Amazing. We can just go and prioritize it. I wouldn't need to do much else, but I am going to do more. So here, just to quickly go through it here is what the context behind it is. Who's asking, what are their Personas, what are variations and what are like maybe common objections. And then we go through what's the underlying topic that kind of flows through it all. After that we're going to prioritize the questions. So again, similar style prompt, but I'm going to say great. Here's what's coming up recurrently on sales calls. Now let's overlay it with that list of questions we saw earlier. And then what we get out at the backside is just a list of where they have common ground. So now we're saying what's a pain point for a customer and what's an opportunity for us to go and create content? Let's try and nail two birds with one stone. And then last but not least, we're just going to format that as we need so it has the query, it matches it to the audience and we can get as nuanced with this as we want. But then it pushes it not to that one, push that to my grid and now I can prioritize further if I want and I can just start to run research and brief creation and article creations left to right. There's many different things we can dig into, but maybe let's run through if there's any questions.

Dave Gerhardt (27:07)

This episode is brought to you by a team that I've personally hired twice, Compound Growth Marketing. And they're smart enough to sign up as a sponsor for us here at Exit 5. I work with John and the team at CGM, both at privy and Drift. And if you're trying to figure out demand gen, they're the team you should call. Especially in a world where so much is changing. With AI, they know what they're doing. They're grounded in first principles. But they're also fast and adapting to what's changing with technology today. They've managed over 50 million in ad spend for fast growing startups and public companies. But here's what really sets them apart. They don't just run campaigns, they build systems that scale. Compound Growth Marketing has leaders and consultants who've been in the trenches at companies like Hunt club, goto, workable, monster.com and IBM. So they show up like true operators, not vendors. They understand what it's like to have the pressure to hit pipeline targets and to be accountable to the sales team inside of your company and the biggest unlock they blend demand gen with something that they call GTM Engineering. It's a mix of large low code automation, AI workflows and systems thinking that helps drive more revenue. It's not just about leads, it's about building smarter, more efficient go to market machines. Most agencies are still stuck on cost per lead models but Compound focuses on full funnel roi, pipeline creation and long term growth. If you want a partner that understands your goals, moves fast and can actually help you win, go to compoundgrowthmarketing.com to that's compound growth marketing.com and make sure you tell them that I sent you there.

Owen (28:40)

I think Dan had a good comment as well. Around Air Ops have like a lot of recipes. Like everything I generally went through there is available like out of the box. So like in clay where you can pull in different elements and you can like customize. Within our own tool we have a lot of like individual what we call power agents you can just like pull onto the grid to help you get started.

Dave Gerhardt (29:01)

Here's a couple. One of the questions is how are you getting these questions from different online sources? How did you compile the AEO queries? Where did you figure those out for your company?

Owen (29:10)

Great question. Not self promote, but we have our own internal tool to go and do that. If you don't have a tool that's surfacing those insights for you, where you could go as like Google Search Console or you could go to Google Analytics. One thing I was working through with a friend earlier, not even a customer, is they have a ton of pages for which they have high impressions but no clicks. So that could be one really good place that you would start just for prioritization. So Google Search Console you can go.

Dave Gerhardt (29:40)

Yeah, we've done this. We get a ton of searches for like B2B marketing jobs and there's a ton of impressions that show up. And so it's like our content is showing up with that search query but no one is actually clicking on it. So then that creative exercise there is in like how do we match that gap you I'm not deep in the game like I used to be anywhere, but I feel like back in the day it would be like yes, there are a lot of tools. So obviously you could use air apps to do this. But isn't a lot of it also like just going in an incognito tab and like searching for things and seeing who's ranking for stuff and seeing who's showing up and seeing Google would also give you hints it would be like the suggested searches, you know, at the bottom of the page were a gold mine for figuring out what we could create from here.

Owen (30:20)

Yeah, 100%. Like even best cat management software. Let's just see what comes up. But you can even look at. There's usually like a people also ask tab. You go into Reddit for the queries. You go and care about. Like you were saying, Dave, you can do a double click. You can figure out, okay, is there interesting YouTube videos that exist that there isn't a matching piece of content for? There's people also ask. Just go to ChatGPT, put in the query and just figure out, okay, this is something I really care about. The sales team are constantly sending it to me in Slack. Do we have a piece of content that meets that need? There's ChatGPT, there's your sales team, there's. If you use a sales recording tool, there's many places like you honestly just need about one. And then you need the marketing confidence to be like, here's what I want to prioritize. I think that's an easy way to get started.

Dave Gerhardt (31:09)

And then one more and then we'll kick it back to Dan. And by the way, there's going to be a ton of questions following up post this. It would be. We could spend six hours and have to go deep on all this stuff. But I know Owen will be glad to have you send them a DM on LinkedIn and try to answer some stuff after this. But this is a good question from Christina. Okay, so you have the final content. Do you have to reformat it prior to publishing? So that's not evident. Like, are you just taking something that, that you get from this and like immediately throwing it up on the site? What's the process?

Owen (31:35)

Really good question. So please do not do that. I think like a year ago people were going to chat GPT and they're like, generate an article for me. It was like copy, paste, publish. Do not do that. You will end up getting more punished than anything. I'll show it in the next example after Dan in a bit more detail. But what we get out, the far side is for refresh content is like a vis a vis. So everything that's scratched. Imagine I have an amazing editor. Internally they're saying let's get rid of anything in gray and then anything that's proposed. It's also like written in purple so I can cleanly see it. But it's given a reason as to why to include it. So one thing I really liked earlier was this is where you should be working with the agent rather than just like taking everything for granted. But in this example, we had referenced on one of our pages a SparkToro report, and the agent was clever enough to be like, hey, you have a piece of content that also shares similar insights. You should go and mention that as well. So then the proposed improvement here related back to us is updated with their latest report and ours as well. So the TLDR is. No, we're not just taking this and like blindly publishing it. The value and what. And our team, we train people for and with our tool we like, invest in is like the collaboration features and like editing that is like, so, so important. That's like the last mile. Or we call it human in the loop. That is still, like, critical.

Dan (32:59)

Cool.

Dave Gerhardt (33:00)

Very good. All right, Go have a Gatorade backstage. We'll see in a couple minutes. Mr. Dan, let's bring you back up here. Good job, Owen.

Dan (33:05)

All right.

Dave Gerhardt (33:06)

On this one, before you dive into it, I want you to say, like, here's what I'm going to show. Here's why we do it and here's why it might be useful. And then let's go into this one for sure.

Dan (33:15)

So for this one coming up and the people that want a free tool, this one's free, so this is going to be good for them. The context is that you are a small team. In this example, I used HubSpot, which probably doesn't have small marketing team, but we can ignore that. You have a small team. You have a product marketer on your team. They're overwhelmed with work. They can't get to everything that is within their scope of. Of their job title. Right. So one of the things we wanted to set up is, hey, how do we listen to. We have all this rich data from sales transcripts. Sales call transcripts. Right. How can we take that and turn that into more quantitative data for the product marketer to look at and actually make decisions off of and actually be like, hey, this is something I should dive into. A new competitor popped up, stuff like that. So that's kind of the context of this is it kind of runs in the background. It gives them insights really quickly, and then they can call, prioritize their work based off that.

Owen (34:01)

Cool.

Dave Gerhardt (34:02)

Christine says, oh, this already sounds familiar. So you spoke to her soul on that one.

Dan (34:06)

Good, perfect. Yeah. We've been wanting to do a lot more actually with. And I'm trying to figure out how to share my screen here, but we've been wanting to do a lot more with transcripts. And there's, there's so much more you can do beyond this example.

Dave Gerhardt (34:17)

This keeps me up at night. I'm like, convinced we're not doing enough with the content that we're creating. And I, like, lay in bed. I'm like, we need to be doing more with transcripts. We have the transcript. If anybody on the team has heard me, I'm like, we have transcripts from 350 hours of content.

Dan (34:31)

Yep. No, and it's, you know, you just got to focus. I think it's a frozen lyric, but, you know, you got to focus on the next right step. Right? Like, if you look at the whole staircase, you're going to be overwhelmed. But yeah, we've been, you know, looking.

Dave Gerhardt (34:41)

Into, hey, can we just casually drop a frozen lyric?

Dan (34:44)

Sure, of course.

Dave Gerhardt (34:46)

Deadpan. He said, it's a frozen lyric.

Dan (34:49)

We've been looking at, hey, can we look at these transcripts of demo calls and be like, hey, how familiar are they with our brand when they are on the call? Can we get a score for that? Can we use that to kind of measure how our brand spend is working? Right? Like, if we see people are more familiar on that first call over the course of a year, then we're doing something right. So there's a lot you could be doing there for this. I shared my whole tab, a window actually. Here's kind of the output for this. And this is just based off one of the calls. But, you know, I have two tabs here. Which one is the competitor mentions. One of them is the objections. You could also have like a feature request tab. You could have a whole bunch of other stuff here, but it pulls in, hey, what is the date? What is the meeting id? I use this so you can search stuff later. Meeting title, the X is redacted. Competitor mentions or competitor name, how the context in which they were mentioned, the sentiment. And then for the objections, we have a few categories that we want to keep track of. And then like the, obviously the context and, you know, severity. Was it addressed? Stuff like that. If you had a month's worth of this data, you know, you can also make some dashboards, right, like of, hey, what's the brand share, I guess, of competitors mentioned on our sales call? Are there any new ones that are mentioned? What kind of objections are we seeing? Is there anything we can do from a marketing perspective to kind of get ahead of them? And now let's go back to how we actually got there, right? So again, this is one where if the last example I was focusing on, hey, you want to have input, output examples, you want to have context. I think this one what I want to highlight is you want to iterate on this process. The first prompt I made was awful and had a lot of errors and I just kept working piece by piece working out these issues and I'm sure there's more issues I can continue to work out. But you want to do stuff like that. So gave us some context of who we are. It runs on when this one's focused on my transcripts. But you can also.

Dave Gerhardt (36:29)

What is this tool, by the way? Did you say that?

Dan (36:31)

Yeah, this is zapier agent. So it's got a generous free tier. It's not unlimited, but it's enough that you can start testing stuff out.

Dave Gerhardt (36:37)

We can have a separate chat and I'll argue with anyone about budget. Yeah, Air ops is expensive at two grand a month. I'm like, you know, it's expensive. Hiring full time humans is expensive. Two grand a month is not a lot of money.

Dan (36:48)

For sure. Yeah, no. And I think there's different entry points, you know, and different tools you can test out. But yeah, for the competitors, I thought it'd be helpful if we mentioned what competitors we're primarily focused on and then give it some categories for the objections. Right now here comes the instructions here, right? So I tell it, hey, listen to the call. Right now it's only listening to my calls. I could make it listen to a team's call. So you'd probably do this for the sales team. Give some helpful things, right? Like if somebody mentions HubSpot, always spell it HubSpot and reference the competitor list to make sure whether or not I want it to include competitor mentions that aren't in my list is something you could dial in similar thing for the objection recognition. And then I just tell it, hey, put in the spreadsheet, right? Like put in that spreadsheet. Here's how I want you to format it and match it to the columns. And then here's where I did some tweaking. I had some quality control things at the bottom. So like it mentioned HubSpot as a competitor. Well, obviously we're HubSpot in this example. I don't want it to mention ourselves. So I made sure to go in there and mention that and then had some where the host mentioned a brand and it threw that in there and I'm like, okay, I want to make it so the host doesn't. So you know, there's stuff you want to tweak and test to get to a good spot. But you're not going to have like, if you build this and run it and then think you're going to be good to go with it, you're going to be disappointed with the first result and you're going to refine that until you're happy with it.

Dave Gerhardt (37:58)

Because I guess how much does the format, like all of these pound signs and asterisks and all that stuff is that like you make that up or that's your formatting? Does that matter?

Dan (38:08)

So this is markdown formatting, which a lot of LLMs are really familiar with it. It kind of structures the content. You're sending it to them so they can kind of understand it better. Honestly, my first draft of this, I probably put it in Claude and said, hey, here's what I want. Can you give me an example? And then I tweaked. That's what I did.

Dave Gerhardt (38:23)

I was like talking to ChatGPT. I'm like, tell me how to write a prompt for Sora, please.

Dan (38:27)

Yeah, yeah, and give it the context. Like there's hacks to get, you know, and, and you want to improve it from there and you want to make sure that you're like, hey, I want an input output section. I want that. So that's something that we actually don't have here that I could improve it. Like I could give it examples of what somebody might have said on a call and how I want it to actually track. Like that would improve the quality. There's even more you can do them what I've mentioned here, but those are kind of the tips and tricks, I guess.

Dave Gerhardt (38:48)

How do you handle data privacy in these platforms when uploading company owned sales calls. Transcript. I have an opinion on this. I don't know if I'm right or wrong. I'm going to let you answer, but this comes up in any time we talk about giving access to these tools for sure.

Dan (39:01)

And I think a lot of it comes down and I had a brief role in it and I was a person who was doing reviews of different platforms, so I might have a different perspective. But you want to be careful for what your company kind of recommends. And in a lot of cases it might be that you have to do a internal review process to get some of these platforms reviewed. A lot of times what I like to look for is SOC2 reports. Like usually if there's a SOC2 report and you can get access to that. Look at that for Zapier. It'll probably have Zapier agent within Scope and it gives you a level of comfort there. I'm hoping that's kind of the path they wanted me to go down there. But I think there was a lot of concern around, like, if you send off transcript data to an LLM, like, is it going to train on that data and stuff? And there's ways to kind of opt out of that training. And I think it was a problem earlier on. Like, you. You saw a lot of, like, news articles around it, but there hasn't been a lot recently because I think it's. It's getting too.

Dave Gerhardt (39:53)

That was Anthropic's whole pitch was like, we're like chatgpt, but we care about your data more anyway, that's a whole exit. That's a bigger discussion and you can.

Dan (40:01)

Opt out of stuff in Anthropic settings. One thing in Anthropic is you have to say, turn off feedback because you can opt out of it. But then if you have the thumbs up, thumbs down feedback, if somebody in your org clicks thumbs up, it will store that information and train on it. So it's like you have to turn off everything.

Dave Gerhardt (40:16)

A couple other things. When using sales calls as context, do you clean up any of the transcript first? We have a lot of transcripts in gone mode, but I've seen a good bit of poorly transcribed text.

Dan (40:24)

Yeah, for sure. I'm pretty happy with how fathom. So in this example, I use fathom. I'm pretty happy with the transcripts there. You could definitely have a step before this. Maybe not. I'm not sure exactly. I'd have to explore how you do it, but you could definitely have like a cleaning step before that. And it could do stuff like, hey, if we know competitors called HubSpot, like spell it this way, right, you could do some of that. But I handled some of that in the prompt itself and I'm pretty happy with how Fathom's transcripting goes.

Dave Gerhardt (40:49)

Cool. All right, we're going to bring Owen back up. I'm like, we need to come up with a format for this. That's like four hours. There's breaks, but there's just so much I'm trying to, like, give everybody the context of. Like, this is a free live session. We're going to cover as much as we can. The goal is to get your brain turning. But, man, there's. There's so much here that we can continue to educate. All right, so, Dan, I want to bring you back up after Owen and then we'll have the three of us on stage and we'll finish this out with Q and A. But Owen, what do you got? You got one more.

Owen (41:14)

So we did content creation. Now let's look at content refresh and there's one thing you can take away around what works in AI search. What's going to get you mentioned? What's going to get you cited? That again leans on where you've made previous investments. I would say it's content refresh from our own research and I know from what's been shared publicly from other folks as well. If you've gone to refresh your content in the last quarter, you are three times more likely to be cited by your chatgpt your perplexities than if you were to leave it six months plus. So there's like this steep drop off curve. So just keeping your content up to date is like super, super important and it's to get your rewards not just in AI search but in SEO as well.

Dave Gerhardt (41:55)

Like are you talking about there's value in like the stuff? Because obviously you're, you're going to put more new stuff out but then you also want to go back and update the existing correct.

Owen (42:04)

So like as soon as you put something new out, your counter starts. You're will say zero days at publish. But for everything you did previously, I was laughing earlier because I'm the same. We've done a ton of webinars, we don't get enough value from it. But even those pages that exist or conference recaps from like three, four years ago, you should come back, you should update it because it's kind of like what I'll go through here. Regarding there's like structure elements that work for AI search but then there's also like your brand and like Hair Ops is small enough company growing really quick, but are we have new products, we have a new positioning, we would have new content that we want to make sure is like internally linked or externally linked. So everything in your back catalog is valuable. What I will say is not everything should be refreshed. Again if you're layering on different solutions, some of what you wrote in the past might be no longer topically relevant and you'll actually get a bump in traffic and viewership if you were to get rid of it. So again a topic for another day. But content cannibalization is very much real. Like follow where you want your brand to be known today and that's where you should invest your time. But kind of getting back on track with what I'll demo here, I think you set it up well, for Dan, Dave, but this is important for the reasons I just mentioned. In terms of outputs, I have three key workflows here and if we were to do this process manually, I wouldn't do anything different. So the process I've built with automation very much follows what we would do manually if we didn't have access to a tool. So what I start with is, is a URL that's on our site that I'm interested in going to refresh. I'm going to do two forms of audit. One is how is it structurally presented to an LLM? Like, where is it weak? Where is it strong? We'll go into that brief in a second. I then want to do the brand alignment review, which is why I just spoke about your products. Change your new content. You should like make sure internal, external mentions of who you are, who you serve, your icp, all of that's going to change in time. You need to come back and refresh it. So those are my two audits. Again, I'm not just going to blindly say, great, trust the system, it's 100% right. Let's just go and like refresh it. No, I'm going to be able to like edit both of those, make tweaks around the edges where I see fit and then those both feed into my refresh agent then, and then out the far side. Like I showed for someone's question earlier, I guess kind of like the delta I get, like, what did the article look like? What is it being updated to? I can blindly accept all, Please don't do that. For lots of different reasons, but for anything I care about, I can go hover over it and I can see what's the reasoning behind it, which is really helpful. I can accept it or I can just, if this doesn't matter to me, I can delete it directly there and then. So a big question I get every single day of my life is what works in AI Search? We could spend a few minutes after this in the Q and A section looking at it, but here's like a brief I like to use and it basically pulls out some of the most important structures in a scorecard. Again, this is available for free. You just sign up for an account, attach some of your pages, see what it looks like, and then you can action on it manually if you want. But biggest things, I'll call out that matter today for you to improve your presence in AI Search. Freshness is one, content structure is two. What I covered in the previous example, authority. Like you have sales calls you have lots of context internally, you've subject matter experts put that out into the world. The more you're adding to the conversation, the more valuable your content is to an LLM. What is very helpful about this scorecard is I could probably go and do it myself, but it would take me a day to get to this level of detail. But for like freshness. Super interesting. You can see here there is a date there, but it is not relevant. And I only found this out today, it's not read by the agent, so it's actually not giving me like full value for this piece of content because it's saying no visible publication date. I'll show you how we found that. That's really important. So not just having fresh content, but communicating such. And then for content structure, super important. But with a CTA on that page, we need to make sure that the heading alignment is correct. So this in short, in like a two minute run gave me like three or four things I need to spend my evening updating so that we improve our own content. Then for brand alignment we can look at Brand Kit again. It's giving me like stylistic Dan mentioned earlier, like a knowledge base, your brand kit. There's some stuff here that's like inconsistent to our current tone. There's also, I think I mentioned it earlier, but there was a mention of a SparkToro report. Since we published that article, we have our own one, the system detected it, which is great. So even if you were to just run these and you were to do manual updates, I think this is basically elevating marketers to a higher value activity. And then we feed it into the refresh engine which is going to do what we saw earlier. The last thing I'll just click on really quick think of it open here is what are we doing under the hood? And this will kind of give you a feel for how AI search works. But in this workflow, which is behind the AEO scorecard, we're actually pulling the article. So we're not taking just the copy, we're pulling the raw HTML which is what ChatGPT perplexity, all these agents are actually reading and then we're just going to clean it. So can I write Python code kind of, but I just use like a copilot, you could use anthropic, whatever you want to just remove some of the extra stuff, style comments, extra tags your website team would have added in. And then really under the hood, like what we saw there, structure, authority, freshness, and again using. And this is all available like out of the box. You don't need to write Python code, but evidence. What matters is the number of links. Like, does it have references to other content on your website? Super, super important. And then I've shared a few metrics. Again, I might say this is the most important one of the day. But compared to ranking on page one versus actually getting cited by ChatGPT, not that anything of your core marketing principles is very different. The bar is just drastically raised. So you are three times more likely to get your page one article cited by an LLM if you just get your headings in order. Which sounds kind of dumb, but there's a huge correlation there. So that's what I'm doing. I'm computing all the scores and then I'm going to use a model to map it to a scale of 1 to 5. And then out the very backside I'm generating that improvement brief. So that is what we see over here. Again, you can manually interpret this, you can figure out. Great. For each of these elements, I have a header section, needs better meta description. I can go and update it, but kind of like end to end, like Dan's examples. Make sure that you have a process that you want to like, further invest in and then build automation where it makes sense. I like doing content, but I sure as hell don't want to have to produce this myself. That's where it's helpful. And then I'm better equipped to like evaluate the output and make the next step where we need to go.

Dave Gerhardt (49:17)

Okay, there's a million questions about this. I think we can. Owen's going to reach out. You'll get an email from them tomorrow or the next day. At some point I would suggest replying back and asking any specific questions there, just because we're not going to be able to cover them. I'll. I'll just cherry pick a couple of them. So one of them is like, where does this sit on the team? Who do you think in this world today should typically own this?

Owen (49:37)

Great question. What we typically see across our user base is a head of content, content marketer, like an SEO person, someone who has a really high bar for quality themselves, but can also kind of get in under the hood. And what we call it ourselves and what we like train people up on is around content engineering. It's kind of like the mixture of those three aspects. You can build, you understand strategy, you understand content. But I think someone who comes from the background of content plus the ability to build could easily get in under the hood and own systems like this.

Dave Gerhardt (50:11)

What do you guys think the biggest gap is? If you're watching this and you're like, oh my gosh, this stuff was great, but at the same time I'm feeling so overwhelmed about. I feel like I just saw like some PhD level examples of AI and like I'm still just trying to use Canva in the right way. How would you. Do you have to carve time out at night to learn this stuff? Like, how do you bridge the gap more as just like entrepreneurial type of marketer? How would you approach this?

Dan (50:36)

Oh, I'm going to promote for Owen, but they just launched today in Academy. I think it is Owen for learning. So really great for your learning resource. So there's a lot of tools like that. Like, you know, Clay has their own too. Right. Like there's resources out there, I think in the like to learn by doing so. Like just making a zapier agent account and getting in there and seeing what you can do. So it depends on like what type of learner you are. I want to know if you had kind of thoughts on that.

Owen (50:59)

Yeah, no, a hundred percent. You want like training that helps you like progressively build. Like, we've all done the webinars or courses where like. And it was the antithesis of what we wanted to set this example up for. But you want to do like a course where it's kind of like helping you build something progressively. But I think start super small, pick a workflow that's core to your business and not even try and do like what we show there probably could be PhD. Some of it is just like available out of the box. But start doing like one or two simple things. Like in Dan's example, we have a webinar flow internally, but it started with, hey, can we do research on the guest so that we turn up to the first call. Basically more knowledgeable, trying to figure out where they overlap with what we provide. So I would say start really small and then try and learn in addition to like some online resource, if that works with you, text or video, whatever.

Dan (51:52)

It is, and your first result might be underwhelming to you. And that's fine. Just keep iterating on it. Like even my first goes for a lot of stuff. I've been doing it for a bit. I don't like the output and I just keep working until I do. And it's not even that hard of a process. The first build's always the hardest and the more you learn there, that can be taken to other, other steps. I think there is Also a gap in like identifying where you can use AI. And that's what I was hoping to bring with some of these examples of like, I was hoping somebody would see the, you know, the webinar one and be like, hey, podcast is another spot I can build something like that for. Or, you know, just showing examples. And I like to categorize them into like three categories. Right? Like speed to value, which is the webinar one I showed. How do we get to zero to one quicker and then we can refine it there? I think a lot of Owen's stuff can fall into that, enhance the experience. This is something I didn't have an example for, but, like, say you're doing in person events and they're really killing it, but you want to enhance the experience. And you can literally just have a workflow where you drop in the address of the venue and it can go out and find restaurants, you know, coffee shops, bars with high Google my business ratings and build like a, a little list that you can give to all your attendees. Right. Like, and the last one, increasing scope without increasing bandwidth. I think some of Owens fall into that as well, but I think that's where also my zapier one falls into. But like, if you're looking at, in those three areas, I think you'll. You'll start finding stuff.

Owen (53:05)

Cool.

Dave Gerhardt (53:05)

All right, man. We're just starting to chip away at this, but we have so much more here. I saw a bunch of plugs for what we're doing inside of exit 5. We have a free AI course. We do a lot of discussions inside the community. We're going to continue to do more conversations like this. Dan on my team has been blowing me up in the, in the, in the slack this whole time being like, you need to do more than this. You need to do more of this. We need to do more live sessions. We need to show more AI stuff. Like, totally agree.

Owen (53:29)

We will.

Dave Gerhardt (53:30)

For now, thank you for all the comments today. If I were you on this, honestly, I would reach out to Dan, reach out to Owen, ask for questions. After that, go to them on LinkedIn. I want them to come back to me and say, man, I got so many messages on LinkedIn and just stay curious. I think my takeaway on that is I think everyone that attended this just by nature of like, if you're going to take an hour out of your day and go to a webinar about marketing, you're already probably in the top 1% of people who care about their job and want to treat this as a craft. And so it's easy to try to take all the courses and study all the materials, but I think truly the best thing is to, like, yeah, do what Dan said, which is log into Zapier, or try to find a workflow and try to learn something like, hey, what's a task that I could try to replace or replicate with AI? Or go check out the free version of Air Ops. There's so many ways. And everything I've learned in my marketing career almost always starts by like, yeah, cool, someone showed me it. But maybe it's just how my brain works. It doesn't work for me unless I actually go and try to, like, move the stuff around together and then I can figure out how to scale it. So, Dan, Owen, virtual round of applause. Thank you guys for coming here today. Thank you. We had an awesome, awesome turnout. So much good questions in the chat. So much room for feedback and ideas for the future. A lot to learn about AI. Stay curious. There's a lot we can keep doing with AI and we'll see you all around the XE5 universe on another day. Owen, Dan, great job. See you guys soon.

Owen (54:43)

Thanks for having us, Dave. We'll see you later.

Dave Gerhardt (54:49)

Hey, thanks for listening to this podcast. If you like this episode, you know what? I'm not even going to ask you to subscribe and leave a review because I don't really care about that. I have something better for you. So we've built the number one private community for B2B marketers at Exit 5. And you can go and check that out. Instead of leaving a rating or review, go check it out right now on our website, exit5.com. Our mission at Exit 5 is to help you grow your career in B2B marketing. And there's no better place to do that than with us at exit 5. There's nearly 5,000 members now in our community. People are in there posting every day, asking questions about things like marketing, planning, ideas, inspiration, asking questions and getting feedback from your peers. Building your own network of marketers who.

Dave Gerhardt (55:32)

Are doing the same thing you are.

Dave Gerhardt (55:33)

So you can have a peer group or maybe just venting about your boss when you need to get in there and get something off your chest. It's 100% free to join for seven days, so you can go and check it out risk free and then there's a small annual fee to pay if you want to become a member for the year. Go check it out. Learn more exit5.com and I will see you over there in the community. Hey, this episode is brought to you by our friends at Customer IO. You do you remember? I'm old enough to remember this. You remember when a personalized message meant slapping someone's first name into an email? Hello David or hello Gerhardt?

Dave Gerhardt (56:15)

Yeah.

Dave Gerhardt (56:15)

Well, those days are long gone in marketing. AI has raised the bar for lifecycle marketing because now you can deliver smarter context aware communication that actually feels personal. And you can do it at scale without hiring five more content people. Personalization today doesn't just mean using my name. It actually means having context about any previous interactions. But the the problem here happens because even though this sounds great in theory, most teams aren't actually doing it. They're stuck with broken reporting, siloed data and outdated stacks. It's often easier just to keep doing things the way you've always done them right. Isn't that kind of the norm? Default to the status quo. So customer IO, they did a survey on this. They surveyed 600 marketers just like you and me to figure out what's actually working and what's broken in. This is what we call lifecycle marketing and they detailed how the best teams are actually solving these problems. The report breaks down 2025 priorities, where budgets are moving and how to tame the measurement mess. Real world examples from brands like Notion and Monarch Money that use AI personalization experiments and understanding the next chapter of AI, what's on marketers Wishlist right now and how customer journeys can get smarter, not just faster. It's packed with examples, data and strategies you can put to work right now. If you want to get smarter about lifecycle marketing, this is a great free resource. So go check it out. You can get it@customerio exit 5 and you'll learn how to build lifecycle marketing that keeps up with today's expectations. That's customer I.O. exit 5.