B2B Marketers Confront Generative AI

 

Drew Neisser: Hey, it’s Drew. And I’m guessing that as a podcast listener, you will also enjoy audiobooks. Well in that case, did you know the audio version of Renegade Marketing: 12 Steps to Building Unbeatable B2B Brands, was recently ranked the number one new B2B audio book by Book Authority. Kind of cool, right? Anyway, you can find my book on Audible or your favorite audio book platform.

And speaking of audio before we get into today’s show, I do want to do a shout out to the professionals that Share Your Genius. We started working with them several months ago to make this show even better, and have been blown away by their strategic and executional prowess. If you’re thinking about starting a podcast or want to turbocharge your current show, be sure to talk to Rachel Downey at shareyourgenius.com and tell her Drew sent you.

Okay, let’s get on with today’s episode.

Narrator: Welcome to Renegade Marketers Unite, possibly the best weekly podcast for CMOs and everyone else looking for innovative ways to transform their brand, drive demand, and just plain cut through. Proving that B2B does not mean boring to business. Here’s your host and Chief Marketing Renegade Drew Neisser.

Drew Neisser: Hello, Renegade Marketers! Welcome to Renegade Marketers Unite the top rated podcast for B2B CMOs and other marketing-obsessed individuals.

You’re about to listen to a recording of CMO Huddle Studio, our live show featuring the CMOS of CMO Huddles, a community that sharing caring and daring each other to greatness every day of the week.

The expert at this particular huddle was Noah Brier, founder of brXnd.ai. He joined us to discuss how AI is transforming B2B content. Let’s get to it.

I am really thrilled to introduce you to my friend Noah Brier. Noah’s notable achievements include: founding Percolate (now part of Seismic) and two other companies, including brXnd.ai, which we’ll be talking about in a little bit. As it happens, Noah started his career as a writer at this agency called Renegade—some of you may have heard of—in 2004. And what was amazing is that he had already distinguished himself as a journalist and an early blogger while he was at NYU, which blew us away. Four years ago we did a couple of episodes on his entrepreneurial journey. We’ll share one of those in chat. So anyway, we caught up a few months ago, and I wasn’t at all surprised that Noah was starting something new in the AI space. And it had already written some code that integrated—I don’t know if it was with Chat GPT or something, but we’ll get into that. So naturally, I thought, okay, if we want to talk about generative AI and marketing, we really need to get Noah on the show. So, Noah, hello, welcome!

Noah Brier: Thank you for having me.

Drew Neisser: It’s great. It’s really amazing. I love talking to you, as you know. I mentioned you are an early blogger. I mean, this is early 2000s. And it was about the Chicago Bears as I recall. And so I have to say, you know, you’ve been in New York a long time. Are you still a Bears fan? Do you still have that blog? Have you just been worn down?

Noah Brier: Well, I’ve been pretty worn down by the Bears. It’s not been a fun 15 years. So at this point, I’m technical support for the blog. A good friend of mine is the full time writer. And I basically have to fix WordPress every 2 months or something when something breaks. But that’s the extent of my involvement thankfully.

Drew Neisser: Yeah. Well, and you know, you also helped me get started with my blog and how to do that back in 2008. We were early in this world of adopting.

Let’s just start with brXnd.ai. And for those who are not looking at it, it’s brXnd.ai. What’s your vision for that?

Noah Brier: That is a good question. I’m still trying to figure it out. I think a little bit broadly, you know, we’ve known each other for a long time. I really like brands. Fundamentally, I think they’re sort of fascinating things. And I got swept up in AI stuff earlier last year. Naturally, I combined those two interests, right? I built a little fun weird collab experiment where you could crash two brands together and make a brand collab out of it. And it was all generated by AI. The marketing copy was generated by AI. But honestly, you know, as fun as that was, and it made really cool visuals. The thing that really struck me was that good brands consistently came out with better images. And like at first that seems really obvious. But then when you think about it a little more, that means somewhere in there, it knows what a good brand is. And that its understanding of what a good brand is matched by. And so then it became sort of an obsession with like, Well, how does it know that? What does it know? Can it quantify that? Because if it knows it, it means it’s quantified somewhere in there.

Then I was off the deep end, and was writing all this code and trying to extract what it knows, and I just got excited and I got excited about talking to people. So then I decided I was going to put an event on. So I’m going to put a conference on in New York City in May, and I’m doing a bunch of other things. And you know, just exploring the world of AI.

Drew Neisser: Awesome. So Noah started his career as a writer, but you wrote the initial code for Percolate, right? I mean, you taught yourself how to code, you sort of are one that loves getting into the technical aspects of these things and understand how they work, and then manipulate them.

I got 10 emails this morning about Chat GPT. I mean, just this morning. And everybody is talking about it. And I know that you’ve been playing with it, and a lot of interesting things. So if we keep it broad for a second, how big a game changer Do you think Chat GPT is for marketers?

Noah Brier: First, I’d say let’s sort of like zoom out of Chat GPT. I think talking about it as Chat GPT is a little bit of a tell about how you’re viewing it. Underneath all this is machine learning and these are a subset called large language models. And underneath Chat GPT is a specific approach called GPT. And, you know, most of my use, for instance, is not through the chat interface. But I write a lot of code that hits the API’s, which are the same API’s that the chat interface hits. Most of my interest, I’d say is that the sort of large language model level, right? I actually, like have very mixed feelings about the chat interface itself. But what these large language models can do is pretty extraordinary.

And, obviously, a lot of the focus has been on the writing and, you know, getting them to publish blog posts, or SEO content or whatever. And they’re fine at that as long as you’re happy with like, very average content, right? Because fundamentally, like they’re consensus machines, and they’re gonna give you sort of consensus answers, and like average writing. Where I have found them to be absolutely extraordinary is in doing like, much more basic tasks, though. So you know, things like categorizing content, taking survey results and categorizing them. I’ve been doing a lot of web scraping and summarizing. Those sorts of things that is like, already, in its current state, that’s a game changer. It has solved a problem that I have wasted many days and months of my life. Like maybe a whole bunch of problems, I’ve wasted many days and months of my life doing.

I don’t think that it’s at the point yet where it’s not ready to write great ads. You know, it can be an interesting companion. I think like, it certainly can be a helpful partner. It can get you started. And again, when you want consensus—like, you know, I was talking to a friend yesterday, and he had to write a formal letter to the city requesting something. Like your formal letter to the city is not judged on its creativity, right? It’s judged on its formality. And how close to the average it is. It’s like perfect for that stuff, right? It’s really, really good.

For me the day to day application where it’s all using these models for doing data extraction that like either I had to do, or I had to write a lot of code to do where I had to hire a bunch of people to do or put it on Mechanical Turk or something. And it’s just done. It’s solved. It’s like, absolutely extraordinary. I mean, it is a little mind blowing considering how much time I’ve wasted trying to do this stuff in the past.

Drew Neisser: We got to get above that for general for writing. But talk about what you’re doing with a survey. Because, you know, I know lots of CMOs listening to this they do surveys too, or they’re gathering data. How, specifically, are you using it that way?

I want to dive into that, because I think we’ve all sort of used it to try to write the newsletter. And I’m gonna speak to that first. Because you and I talked about this, so if I said write a newsletter for B2B CMOs on content marketing—or no—here’s a better example, because we’re talking about PR. Write one on how to get the most out of PR. What it gives you is 8 very basic and common things that are absolutely right. We’ll call it marketing 101. It’s not going to give you have an insight for the moment. But it’s still useful to make sure that you’ve covered the basics, right?

Noah Brier: You know, I mean, this is the kind of stuff it is just amazing at. If you wanted to do—so I recently did some really basic answer classification stuff, right? It’s an open ended question on a survey, and you want to classify it across 5 buckets. Is it a positive? What kind of answer is it? GBT 3 is pretty capable of doing that well, if you know how to prompt it to do it. And it’s all about prompting it and telling it, “Hey, here I want you to tell me which of these 5. Don’t tell me anything else. But just tell me which of these 5 it falls into. And if it doesn’t fall into any of of the 5, tell me you don’t know.” And there’s good ways. That is a thing where we’ve all had to have people code surveys for us, right? Like, it takes time and money. And it just doesn’t the cost of it is something in the realm of 2 cents per 1000 surveys. You know, it’s crazy, right? It’s sort of hard to comprehend.

Drew Neisser: I’m just thinking about it. We’re literally just working on a study for one of our clients, and there were 60 pages of verbatims. And the way we would have done that before, is said to an intern, “Try to classify these.” So now, what you’re saying is, you take all those verbatims, you give them general buckets. And you say—and this is an important point, I think we were talking about before the show, which is tell it if it doesn’t know, tell it to not do anything, right? And that’s an interesting thing in and of itself. But so we have all these verbatims, we give it the categories that they could fall into, and then it’ll just sort them all out.

Noah Brier: It’ll just sort them all out. It’s amazing. And if you want it to be really good, you can do a thing called fine tuning. And so that’s another service that’s offered through open AI. And what you do with that is like if you have past surveys, you know, you’re running a tracker over time, you have all these past surveys, you’ve already coded them all, then you’re perfect. You give them 4000 examples of how these things get coded. And then it becomes extraordinary at—it’ll never get it wrong, basically.

Drew Neisser: Fine tuning. Yeah. Amazing. So that’s a great example. So research, and that’s one that I don’t think people necessarily had, because that’s sort of an advanced notion.

Is there another thing like that? Because I think is this every personal email that you need for board of directors for the co op board or whatever. But what are the things are you seeing where this computational power is just awesome?

Noah Brier: Well, so one, small one, and it’s like along those lines, and it’s probably not interesting for everyone. But I’m guaranteeing some folks here. For doing scraping and data extraction, if you say for instance, You’re a B2B company, and you need to do research—like you have salespeople who are doing research about their prospects, and you want to be able to take that company and look at a couple key things and then pull it into a format, that you can prep the sales team or turn it into content. Doing scraping and data extraction from scrape text is extraordinary. Like, it’s absolutely amazing. It’ll give you key value propositions or whatever you want to walk in with—and include, again, I think, like, worrying about it writing the content is much less interesting to me than like, “Hey, go take my 500 prospects, go scrape their home pages, bring me back their key message, pull out all of the things you need. And then you know, we’re going to use it in our Account Based Marketing Campaign.” We’ll figure out how to include it, we’ll write the content that plugs in those pieces. But I’m gonna have it go and take that raw data, right? Like just a page of text, structure into whatever format I want, pull out key messages, pull out whatever it is, and it is extraordinary at that. I’ve done it 10,000 times now. I am yet to have a day where it works and I’m not just absolutely amazed. I just did it right before.

Drew Neisser: I’m going to emphasize because I want to spend some time on this. Because this is not a use case that I had been thinking about. But everyone listening in or watching in right now is thinking about—economies down—how do we get the most out of every single interaction that we have with a prospect, right? How do we do that?

One of the ways is to know the prospect better so that your interactions with them are more personal, more relevant. And again, you don’t have to pull up your screen and show, but how do you make sure that that is not just the basic? I’ve seen some in the past where the assessments were so basic that you went, “Okay, yeah, this is their job title and this is where they work and this is what the company’s vision is,” right? But there was no insight there.

Noah Brier: Well, again, I think it sort of depends what you want to be pulling out, right? I think that the way to use these tools is more as assistants than as insight generators. And that’s, personally, where I would start.

So I probably wouldn’t start with that question. I would start with, Hey, we have an ABM program going and we have 200 companies on our ICP. I know all those companies and have all their websites, I’m going to go hit all their home pages, I’m going to pull out their key messages. And then I’m going to go scrape their latest quarterly reports and pull out what were the key business drivers and the key business concerns. And I’m gonna put all that in the CRM so that when the salesperson calls in. Or I’m going to plug it into my ABM, so that I can actually like, pull one of these key things out. It’s still like this is a superpower, right? Like, it’s not a replacement for creative marketers. But like, you use that creativity to figure out like, what would I do if I could know anything about any of the customers in my ICP, right? That’s the part that’s crazy. It’s like, you know, you could go scrape every page of every customer in your ICP, and pull down whatever information. And it’s just all there. I mean, the thing I was doing right before this is, somebody asked me if I could do with PDFs. And so I actually figured it out on it all with PDFs. So I’m grabbing PDFs off the web, like for case studies, and then I’m sucking out the content for the case study. And then I’m summarizing it down. So I have case study summary. I know, what are their key customers? And what do they solve for their? You know, and then obviously, once I have that, now I know what I’m going to talk about with them, right? Like you put a case study on the web, and you talk about what you solve for your biggest customer. That’s the thing you’re going and telling all your customers. So that’s the thing I’m going to talk about with you.

Drew Neisser: The thing that really helped me as you were talking was think of this as an assistant. And anything that you might ask an assistant to do. It could do, and it probably can do it faster, better, assuming you give it really good guidelines, right? And that’s part of this. So being able to—Shelly Palmer uses the term prompt crafting. Which I’ve immediately adopted that that’s a new skill set, right? Is that you need to understand.

And so this means being able to direct your assistant to do that kind of thing. And I would challenge I suspect not a lot of folks have thought about necessarily looking at prospect case histories and sucking out that thing. It also occurs to me that there’s got to be a customer marketing application here, right? I mean, you could do the same scraping, right? And I wonder how many of the folks in CMO Huddles that are listening right now are thinking about, “Wow, this could really be a leg up for every single customer engagement?” Yeah. Amazing. It’s really amazing. And, and so not only do know them, and so forth, but you probably know what’s on their mind right now.

Noah Brier: Totally. And you have, you know, you even have content from them that you can add to them that you built for them that you can summarize. And I think take all this and sort of supercharge whatever it is that you want to do.

I’m also just going to paste into the chat, there’s a piece I wrote, since you mentioned prompting, I wrote up my favorite prompt, which for a while I was keeping secret, because I wasn’t positive that I wanted to share it. But I decided that like people are gonna figure it out. And I might as well try to get credit for figuring it out. Rather than keeping it secret. This is my sort of trick prompt for getting it to give me whatever I want in a very tight structure.

Drew Neisser: Your trick prompt. All right, I can’t look at it now. But I can’t wait to see what your trick prompt is.

How important—I mean, you know how to code. And coding isn’t daunting to you. And I know this thinking code too. And we could probably go there. But in order to exploit this. Does it help? I mean, should marketers be bringing in their coders to be thinking about this as well, so that they can help take the maximum advantage of it?

Noah Brier: Yeah, that’s an interesting question. There’s sort of two sides to that. My prompt, actually, my secret prompt is inspired by code. So it sort of inspired by the insight that these things understand how to code and they understand how to code just because there’s more code on the internet than any other content, right? Actually, I remember this was a guy named Charles, who worked at Renegade as well. And I talked in 2005, about the idea that like, you know, the thing that you know, is most prevalent on the internet is information about the internet, right? Like build the internet is the largest corpus of knowledge on the internet. And so it has a really good understanding of that. And so in fact, the prompt I use is a prompt that comes from programming language called TypeScript where I defined the output that I want it to give me in sort of code terms. And it’s able to perfectly understand that output and give me the output in exactly the way I want it. It’s because code has these ways to describe things in very exact terms that you’re able to make it happen. So, you know, I think on that side, that’s just that sort of interesting anecdote where, you know, I was inspired by the ability to write code to find better prompts.

The first thing I’d say to people honestly is like, you know, even before worrying about the code thing is move off Chat GPT, and play in the open AI playground instead. So the open AI playground is a lot like Chat GPT, it doesn’t have memory. Which Chat GPT keeps some memory. So you can say, like, continue and do other things. But the playground doesn’t it’s kind of a direct access to the API. It also doesn’t have some of the limitations of Chat GPT, where you ask it  for things, it’ll always answer. The reason actually, I suggest to use the playground, though is that I find it to be a much better reminder that this thing is not always like honest with you. Because it feels much more raw.

Somebody called Chat GPT, you know, the open AI API, the GPT three API with a necktie. And I think that there’s a lot of truth to that, that like this thing, sort of masquerades with a necktie and you sometimes forget that you’re talking to a robot, it doesn’t really know anything, right. And so the playground I think, just gives you a sort of more raw way to interact with it. And I think gives you a better sense of sort of like how to get better at prompt writing. You know, on the other side, I’m a huge proponent of writing code. And in fact, like, this prompt is all about the fact that you can get it to kick back responses in code compatible formats. And so you can integrate it into workflows in a way that is like amazing. Now, I also recognize that not everybody is sitting there writing code.

I suspect there’s going to be more and more tools that come out that make this process much easier. Because there’s a lot of stuff like, you know, one of the tricks that anybody doing any serious work with this is doing, for instance, it sort of come to be called prompt chaining. And so that’s where you’re taking a prompt, you’re getting an answer, and then you’re taking that answer and you’re feeding the answer back. So, you know, for my brand collab thing, for instance, what I do is I take these two brands, I get it to write marketing copy, and then I feed the marketing copy back, and I ask it to describe the product in the marketing copy. And then I take that, and I get it to write a prompt based on the description of the product from the marketing. So it’s a 3 step process, where you’re chaining prompts together. And I would sort of argue almost everything interesting happening right now is happening with chained prompts.

If anybody’s played with mid journey, for instance, which is an image generation tool. Mid journey is one of the invention generation goals. It’s like stable diffusion, or dolly if you played with those. But it’s way, way, way better. And the reason it’s way, way, way better is apparently because they’re doing a bunch of post processing on your prompt to make your prompt better before it goes in. And nobody knows the secrets of how it’s doing that. But it’s doing something amazing. What comes out is like consistently extraordinary. I mean, I can’t believe this stuff, you can kick out a mid journey. And I wish I knew what the secret prompt was. So I could just go run it on one of these open source ones. But you know, they’re not quick to give up those secrets.

So chaining these things together is I think, it’s like getting real work out of it. Even if you’re thinking about content building, like asking you to write long, one long article, rather than getting it to build paragraphs. You know, paragraph by feeding it back on itself, I think you’re generally going to sort of get better stuff with the ladder, but you have to figure out how to sort of chain them together.

Drew Neisser: Interesting. So before we get to visuals, because I do want to talk about visuals, the thing that occurs to me here is as a marketer, you know, the first thought—and I know a lot of CMOs are playing with it in themselves—but there’s a limitation to our ability to play with it, because we don’t really know what it can do and how to do some of the things that you know how to do. So I feel like that marketers will be smart to bring in somebody else who can think about this, either from a research standpoint, or from a coding standpoint, maybe someone from your data analytics team, just to make sure that you’re not just thinking about input/output content, right. And thinking more about some of these other more larger interesting things like the scraping that you’d find. I know that a lot of CMOs would love to be able to—first of all, a lot of CMOs` have BDRs and SDRs reporting to them, right. And they’ve got to come up with scripts every day. And they’re supposed to do research on the thing. Well the research takes time. So they probably don’t. Now what you’re saying is if you had the right prompts, it would just sort of…

Noah Brier: It’s solved. Like you could solve that problem in a few hours. And then write a BDRs script for, you know, 4 cents or something. You know, it’s mind boggling.

Drew Neisser: So yeah, but that is a really brilliant one, because that’s a time suck that doesn’t happen. And so you have these folks, they call the customer and the customer has no idea who they are, or their email isn’t relevant. And what we’re trying to do here is get to extreme relevance and ideally an insight, right?, that you can use to further a conversation.

So that’s one thing. Two, you talk about the cost of using this. And I know lots of the folks who have tried it here have only used the free tool. When you’re talking about the costs, what are we talking about? Is that with using open AI playground or some other tools?

Noah Brier: The open AI playground hits the opening AI API. So the weird thing about Chat GPT is it’s free. Even though it’s hitting the open API, that’s why they’re now throttling it and charging $20 a month for the Premium Package. They have a bunch of API’s, they’re all charged on volume, you get charged different amounts, depending on how much you consume. And which of the models that use. So if you use their newest best model in its standard format, it’s 2 cents per 1000 tokens. They charge everything in tokens. I can never quite remember what the token to word ratio is. But in my mind, I keep it as like 1000 tokens is 1000 words. I think it’s more than that. But like, whatever that’s the sort of basic idea.

When I was saying like summarization of our classification, you’d generally do it for like, 1000 for 2 cents, it’s because I’m thinking like, you’re gonna spend essentially one token per classification. And so you’re gonna get 1000 of them for two sets, right? Like, that’s just the cost of it.

And so, you know, these things can get expensive if you really, really start to do it at scale. Or if you start to I recently broke everything in my stack because I didn’t realize that if you—so I was saying Fine tuning AI. If you fine tune their most expensive model, it actually goes up to 12 cents per 1000 tokens.

I have a customer I’m working with right now, who we are generating 5000 AI generated dog breeds. And I had not realized that I was paying 12 cents per 1000 tokens. And I ran through my limit, and I had to ask them to let me spend more money with them. And everything was broken until they fulfilled my request.

Drew Neisser: That’s hilarious! Wait, so we’re talking about German Shepherd meets the Chihuahua?

Noah Brier: Yeah. And it generates a description, it generates a photo, it’s really amazing. Hopefully, it’ll launch in the next couple of weeks.

Drew Neisser: Oh, my God, I think that would be a Shep-huahua?

Noah Brier: Yeah. It’s got a naming model, too. So it’s another fine tuning method.

You can think of fine tuning like, you know, you prompt GPT with, here’s examples of the kind of answer I want you to give it. The way to think about fine tuning is imagine if instead of prompting it with 3 answers, you could prompt it with 5000. And so that’s what you’re doing when you’re fine tuning it. Essentially, that’s the sort of simple way to think about it. So imagine if, before your prompt, you had 5000 good responses that it saw. And so now it was really good and knowing exactly the kind of response and so yeah, I fine tuned a model to dog crossbreed naming. I didn’t realize that it was 12 cents per 1000 tokens, and I ran through a lot of tokens pretty quickly. I had to beg them to let me pay them more money. And they were kind enough to say yes.

Drew Neisser: So just thinking about the dog breed, because one of the things that I found was it really struggled with wordplay. It really struggled with any—like if you ask it to write a headline with word play or comedy, it just can’t do it. And my favorite one, and our team laughs, I said, “Write this in the voice of Jerry Seinfeld.” The response was, “Hey, it’s Jerry Seinfeld.” That was it. That’s all I could do.

So when you say that it’s coming up with names for these dogs, is that just because it’s sort of merging the best of the two? Or you have to give a pretty clear guidance, I imagine.

Noah Brier: Well, no. So I mean, that’s where I’m using the fine tuning technique, right? There are 1000s and 1000s of crossbreed dog names that are pretty cute, right? And I went and grabbed many, many, many, many of them. And I took the model and I basically taught it  where those came from, right? Because we know that you know, whatever, golden and poodle is a golden doodle or whatever.

And so you train it. And the way you train it is the same way you write prompts. Essentially you give it a prompt, but you give it the completion, right? So in this case, I gave it golden + poodle is a golden doodle. And so now it’s learned. And what these things do fundamentally at their lowest level is they’re pattern finders, right? They find patterns. Patterns we see, patterns we don’t see, all that kind of stuff. If you give it enough data, it finds the patterns in them. And so yeah, I mean, does it kick out the cutest best crossbreed names? No, not always. But is it like way better than I was getting out of the basic model? Yeah, way, way, way better.

Drew Neisser: Okay. Because you sort of trained it and taught it what to look for. You mentioned, Open AI playground, are there any other sort of copywriting tools, whether it’s Jasper or others that you played with it that you think anybody else should test at this point?

Noah Brier: There’s definitely others that I think are doing interesting things. So if you go to writer.com, I think they’re doing some interesting stuff. I’ve also got a whole landscape of companies, if you go to landscape.brXnde.ai, I have the biggest category is all these content generation tools.

I have to be honest,  I am not super impressed by the content output. And maybe it’s some bias being a writer. I’m personally much more interested in all these other things. I haven’t found something where I’ve been blown away.

I have done some SEO content. Because if you think about like, Where does it not really matter? Where SEO content is content written for machine. You know, you’re writing it for Google’s bought, fundamentally. So you know, I’ve definitely used it for that.

I don’t know, I haven’t been blown away by any of it. As far as the the tools that exist out there, I think I just have a high bar for what good content is, I guess. It’s been harder. I think, some of the stuff I can’t do, I’m much more excited about. So you know, I’ve like used a bunch of the image generation, I’ve done a bunch of the, you know, there’s image editing ones that are really cool. Like, you know, there’s one that uses AI to remove image backgrounds. And I’ve got it as part of a workflow where I need to remove image back, you know, it’s just like, these little annoying things, I use dolly and the collab thing. I haven’t found a single one where I’m like, “Wow, this is really significantly better than everything else.”

Drew Neisser: So the key takeaway here for all of these things, one is think about this just as an assistant for almost anything. Anything that involves taking data and collapsing it into something that you can use in one form or another as an assistant. Which is really very valuable framework to have, when approaching all of this.

I was going to move right into visuals, but you said something that I just I felt like I needed to address but I’m going to keep going. Alright, let’s talk visuals, in terms of… Oh, I know. I gotta go back.

Okay, for blog posts for SEO. And we’re talking about machines consuming it. But these blog posts go on your website, and someone might read it, and a customer is gonna say, “Well, that’s basic.” How would you manage that? I’m sorry, to give you that very specific use case.

Noah Brier: I don’t know, maybe there’s my own expectation, but I sort of always expect those blog posts to be pretty basic. They’re a way to get somebody to your website. And when I land on a website, it’s not necessarily that I’m looking at that specific content. You’re just trying to get people deeper.

So, you know, I did do a thing recently, where I took within the category that my landscape is in and looked at all the most common questions that the SEO tool that I use. And then had the assistant write content based on some existing content was not entirely sort of generic. You can includes some information about me and what I’m doing. I produce a bunch of SEO content with the goal of getting up in some of those key questions. I was looking at Question search terms in that particular one.

Drew Neisser: Right. So almost like FAQs and that kind of thing.

Noah Brier: And so I just set up a whole pipeline. I took the CSV out of the SEO tool and loaded all those question keywords in and then just had to churn through them all.

Drew Neisser: And boom, then those are blog posts. Amazing. Okay.

All right. Let’s talk visuals. What’s blowing you away right now and how are you using it?

Noah Brier: Basically, the way to think about—you know, I think the market right now as far as what’s out there the most is you have mid journey—which you can only use through discord—you have open AI with Dali—Dali is the one with the best API and the easiest to use—and then you have bunch of open source. You know, stable diffusion is sort of the biggest. There’s a bunch of other players too.

There’s a new one called playground AI. And playground AI has been integrating a bunch of additional models that are pretty neat. Like they were the first ones to sort of publicly release a new text edit model where you could take an image, write text, and it would edit the image based on what you asked it to do. Drew’s holding an orange and I want Drew to be holding an apple. And so I say, “Make him hold an apple.” And it’ll change the orange to an apple. And so  that’s a playground AI.

For all my collab stuff. Because I needed an API I’m using dolly. And what I’d say about dolly if you played with it—and it’s really true of all these things—except for my journey. Mid journey is just… you can get good content out of my journey without knowing anything about prompting. That’s why makes mid journey amazing. You can give it a pretty terrible prompt and it will—if you give stable diffusion or dolly a terrible prompt, it will give you reasonably terrible results. But if you give mid  journey, a terrible prompt, it’ll still give you amazing stuff.

But with DALLE it’s all about learning how to prompt it well. So you know, for that the trick is—and I’ll share all these links, too. But there’s a bunch of prompts search engines now, that’s the best way to learn how to do the image prompts is you just go hit those prompts search engines, you find images that you find interesting, and you look at how—I mean, and it’s amazing. When you look at some of these prompts that people are doing in some of these search engines, it’s totally extraordinary.

And you know, I did this thing, I taught a class over the weekend at University of Montana. And I’ve had this notion for a while since playing with these things that you could use these image models as a really interesting teaching tool, because the way to get good results out of them is to have a really deep understanding of both the vocabulary of aesthetics, but also the sort of history of the aesthetic drafter. So like, the more inspirations you can give it. And so when you look at some of these in the prompt engine, if you know a director whose style is famous for this, they’ll put that director in, right? And there’s all these things where—and so I did an exercise with the students this weekend where I gave them the vertigo poster, the famous Saul Bass Hitchcock poster. I didn’t tell them anything about it. I don’t think anyone had ever seen Vertigo before or knew about that poster, and I asked them to make a new one about a movie about a killer grizzly bear, because it’s the Montana Grizzly Bear. They immediately go in and they found who you know, the first thing you figured out is like, Okay, if I tell it Saul Bass, now it makes a better image, right? And so they figured out it was Saul Bass, and then they went and found other aesthetic terms. I think that part of it is amazing. I would really just kind of poke around on these prompt search sites and you know, find ones that you think are interesting.

The prompting can be very specific to the tool you’re using. So you just have to sort of be aware of that. And they all have filters. So you can say, “Hey, just give me mid journey prompts or just give me a DALLE prompts or whatever it is.”

Drew Neisser: Amazing. And, you know, it’s so interesting, because I think it’s the same with the writing. But the more you know about design—because you and I have worked on logo redesigns. And those people who do that have a million words for every little part of a letter, right?

Noah Brier: Yeah.

Drew Neisser: And so I’m imagining that that same kind of expertise in terms of just art and art style, if you bring that you’re gonna get more out of the tool.

Noah Brier: Totally. Yeah, I mean, because think about like, how does it know what it knows? It knows it from the internet, right? It’s scraped the whole internet, basically, right? And so that’s how it came to understand things. And so these words are all used on the internet.

And like, you know, you think about it, like it knows the aesthetics of Picasso. Because there’s so much Picasso. You know, it can do great Van Gogh’s because there’s so much Van Gogh, right? You just have to sort of put yourself in the model’s head in a way where you’re like, “Hey, you know, what we didn’t know a lot about.” It’s not going to know as much about this niche designer from then early 80s. You have to prompt it with things that it will likely know about. The more internet, the better.

You know, I mean, that’s been one of the really interesting insights from all the collab work I’ve been doing is it can reproduce brands—even really small brands that are very online, sometimes better than really big brands that are not as online. So it’s like, a friend of mine runs a small record label called Ghostly, which is big if you’re into electronic music, but it’s certainly a lot smaller than something like Costco. But I would argue that like it has a better sense of Ghostly’s aesthetic, because it’s sort of more on the internet than it does Costco. But if you think about what is the corpus of this thing, it can kind of make sense, right? Like this is Ghostly has a big ecommerce store and they put out lots of product. You know, Costco’s whole strategy is get you to come into their big warehouse.

Drew Neisser: And you know, that’s one interesting thing to just have it assess the design of your brand.

Noah Brier: Yeah.

Drew Neisser: And then say, “Do it.” And then you’ll kind of know as an interesting litmus test if your brand has a design and has a thematic and has a look and feel. You’ll find out right away, right?

Noah Brier: Yeah, I hope to have a tool that does that in the not too distant future. I’ve been playing with exactly that. And I think that question is fascinating. One of the things I find most interesting about this for marketers is, you know, I think obviously a lot of the conversation right now around chat GPT and education and all these things is about telling you false information. And the reason for that is like there is no true or false information to this thing, right? Like it’s just a consensus machine. It’s sort of like it’s building up strength of association. But in marketing, like, that is truth, right? Like this is a place where what’s actually factual is not important.

I sent a friend I’ve been playing with this tool to where it tells you what does the large language model perceive? What is the large language model perceive your competitors to be and why does it perceive them to be you? Why does it think they beat you, right? And so I sent it to a friend of mine and a brand. And it said that, you know, it said his biggest competitor was and it said they beat him because they have better selection.

And he’s like, “Well, that’s not true. They don’t have better selection.”

And I’m like, “Okay, but like, do people think that’s true?”

And he said, “Yeah.”

I was like, “Well, that is too bad.”

Like, in this particular problem, what’s what’s actually factual is not really relevant.

Drew Neisser: Interesting, because it’s just out there. And so I’m now getting to how you can use this to better understand your current brand, and where you are relative to your competition and what you think is really, truly differentiated. I mean, and what the perception. Because it’s going to feed back the perception, right?

Noah Brier: That’s exactly.

Drew Neisser: And perception is reality when it comes to brands.

Noah Brier: Yeah.

Drew Neisser: God, okay, that’s amazing, too.

Noah Brier: Yeah. So, I sort of talked about it, it’s a consensus machine, right? It’s gonna give you the consensus. And if you’re a marketer most of the time, like, that’s what you really care about is the consensus, right?

Drew Neisser: At least you want to know what that is because you got to beat it. I mean, if you want to change it, you have to do something.

Noah Brier: Yeah, exactly.

If anybody wants me to kick out their brand from this thing, I’m very happy to send them a—it’s not quite ready for primetime yet, but I’m happy to send them a readout from basically what is the large language model now.

Drew Neisser: We have covered a lot of ground in a short period of time. So we talked about using it to write software, using it to create visuals, using it to create FAQs. I’m imagining that it could do tutorials pretty well, right? It certainly can answer product questions easily. Are there other applications that marketers should be testing? One, we just talked about, which is your brand image. Do you have one?

Noah Brier: Yeah, I think that that sort of stuff is fun, trying to get a sense of what it knows about your brand. You can get a lot of other stuff out of it too. So the one other thing I will say on this and an area to explore—and I actually think like hugely underappreciated tool right now, because it’s harder to use still. There’s a thing called embeddings. And so this is another API that OpenAI offers. And embeddings, the way to think about it is, you know, what sort of sitting underneath all of these models, is they look at something and they break it into lots of different dimensions, right? And lots of different criteria. So they might snap a photo of you, Drew, and like blue is one and microphone. But there are many, many, many, many, right? And it takes all of those, and it crunches them down, and it uses to place you in space, right? In this like, multi dimensional space. And they have an API that lets you take out the data analysis directly. So they give you the placement of, in this case, Drew in space.

And what you can do with that is pretty amazing. So like that semantic search works, for instance. So like I do a thing on my landscape, where if you type in a use case, I will tell you which companies are best suited for that use case. And I do it by they all have a location in this multi dimensional space. And then I use a database that tells me which one is closest to your search and you’re able to locate it. But I’ve also been playing recently with I just give it a bunch of brands. Because I’m curious, what is its intuitive understanding of the brand space? Like how does it see brands arranged? And like how does it understand categories? And generally, it’s very good, like it understands categories, intuitively, it places brands of the same category in clusters together. But what’s really interesting then is like, why does it place other brands near those brands? And you can just do—I mean, the embedding stuff is like, you know, we’re just at the sort of beginning of that, and it’s still there’s not really a super accessible way to do it. But to some extent, I think like that’s some of the most mind blowing stuff that I have played with is like actually looking at that because there are a lot of use cases. And I could say something I’ve written about that as well.

Drew Neisser: Okay. Johan Abadie, from Processmaker. Johan you had a question?

Johan Abadie: Yeah, I had a question. Noah was talking about the content being produced by some of the tools that are used there was mediocre. I convinced my content team to use Jasper before everything was blowing up in November. And you had a lot of resistance from them, you know, content writers should be the ones writing the content. And I had a lot of pressure on producing more content. I just didn’t have enough resources to make it happen. The right cost. I pushed really hard and media requirement for them to test it on the blog post, email, and to summarize a white paper. Everyone was blown away. So now my content writer is all doing editing, more so than pure content writing. So we do ideation, we do write some paragraphs, they do editing, and then we publish it. So have you seen more people were successful with this? Because we are successful with that.

Noah Brier: Yeah, I probably should have been more specific. I think, to me, what you’re describing is sort of more along the assistant lines. Where, you know, you’re doing the editing, you’re having to do things like summarization. Yeah, those are what it’s amazing at.

I just don’t think generally, like you’re going straight from content to publish. And I haven’t seen it produced. I think that makes perfect sense. Like, having a come up with ideas for content is really good. For any of us who have ever had to publish a bunch of blog posts or whatever, like figuring out what your next one is going to be is often the hardest part. It’s often harder than actually writing it. Just putting 15 ideas in front of you. And just like giving you a head start is great. And so yeah, I think all of that totally makes sense. My comment was much more specific. I still think you’ve got to put in real work. But yeah, I mean, if you’ve got a content new team, you need to make them more productive. Like this is like every one of them having an assistant, right? Like, sort of junior writer who is a part of their team and it’s perfect.

Johan Abadie: Yeah, it’s the best writer’s block buster that I have found, for sure.

Drew Neisser: We have one more question, and then we’re gonna wrap up. So when you think about the skills that CMO needs to have on their team to take advantage of the various—we’ll call them generative AI tools—what should they be looking for?

Noah Brier: Yeah, I like that question. But, you know, there is a new job emerging called ‘prompt engineering’ that showed up in a bunch of places. I’ve been playing this role a little bit. And I think that this is something people will probably have full time in house. It’s like, someone who is sort of floating around and trying to figure out where can we do things with this and how can we integrate it better? And it’s probably like, they write a little bit of code and they are really good at prompting, and they can sort of like, tie it into the workflow of the organization. Honestly, I think you could hire that person today and see it have a huge impact.

My own personal experiences is that because I’m in the headspace. I’m now finding things every day where I’m like, “Oh, wow, can I really just pump all that through and like solve that whole problem for myself?” And, you know, “Oh, well, if I just had this one piece, then I can.” And I just say, you know, it can be a huge unlock. That’s the secret. I think this is why Microsoft is calling it—it started with GitHub for their coding assistant—they call it a co pilot, right? And how can you add these things as sort of a co pilot, an assistant that can sit alongside all these processes and help everyone on the team? I’m sure many of you are technology CMOs. Every technology company that promises like, “We’re going to let you focus more time on the things that matter more,” right? We’re going to sort of eliminate busy work. And, you know, this is that times 1000. Like this is very truly delivering on that.

Drew Neisser: Yeah, solve this problem, look at this data, do this. So I think the thing that’s so interesting and expansive of the way to think about this is we’ve been focused on copy, we’ve been focused on visuals—at least I have in the conversations we have—and we need to start with what are our biggest problems right now. And our biggest problems right now are, for example, BDRs getting the customer to demo. That’s a big issue, because if that lead doesn’t become a real, genuine prospect. Or our biggest problem is just figuring out what content is the stickiest or getting more content. Whatever those are, there is a chance that this can play a significant role in speeding up your ability to solve that particular problem. And that’s the mind blower here for me is just think more expansively about this. And when Noah and I started this conversation talking about, are we at the Explorer stage or the exploit stage? And it feels like we’re somewhere in the middle. Because you can just get into this vortex and keep playing and playing and playing. Let’s align it with your priorities and then use it for those priorities.

So before we wrap up, Noah, two do’s and a don’t when it comes to generative AI.

Noah Brier: Two do’s and a don’t. You just said it, but think of it as your assistant not your replacement for humans.

Probably the second do is just go play. This is gonna be so big and the opportunity right now to just get your hands dirty and tinker is massive.

You know, I think the don’t is probably just don’t get too caught up in the conversations and the hype. The reason we’re talking so much about content is because that’s where the conversation has gone. But I think the more you play with it, and get your hands dirty, the more of a realistic understanding you get from it. And so it’s like, read all the coverage and all those things, but make sure you temper that with hands on experience.

Drew Neisser: Yeah, use it and learn how to do it. All right, Noah Brier, thank you so much for joining us. Where can people find you and learn about your upcoming conference?

Noah Brier: I’ll put the conference link in but it’s brXnd.ai. The x is at the intersection of brands and AI. So there’s lots of stuff there. And I’ll be having a conference in May in New York City, and we’d love to have everybody there.

Drew Neisser: Awesome. All right, thank you so much for joining us, Noah.

To hear more conversations like this one and submit your own questions while we’re live. Join us on the next CMO Huddles Studio. We stream to my LinkedIn profile, that’s Drew Neisser, every other week.

Show Credits

Renegade Marketers Unite is written and directed by Drew Neisser. Hey, that’s me! This show is produced by Melissa Caffrey, Laura Parkyn, and our B2B podcast partners Share Your Genius. The music is by the amazing Burns Twins and the intro Voice Over is Linda Cornelius. To find the transcripts of all episodes, suggest future guests, or learn more about B2B branding, CMO Huddles, or my CMO coaching service, check out renegade.com. I’m your host, Drew Neisser. And until next time, keep those Renegade thinking caps on and strong!