Powerful ChatGPT applications for PPC and SEO marketers

Episode Description

One of the viewers of that episode, Nicholas Baken shared some powerful use cases he was using ChatGPT for in the live chat. So I reached out to him later to discuss more on that.

In this episode of PPC Town Hall, you’ll watch Nicholas and myself discuss the applications of ChatGPT beyond PPC and into SEO, content creation, and general digital marketing.

Tune in to learn:

- How can it generate schema markup for SEO

- How to use OpenAI’s playground to open up more possibilities for GPT

- ChatGPT’s limitations

- How ChatGPT can create Google Ads scripts and discover negative keywords in PPC

and much more

Episode Takeaways

Generating Schema Markup for SEO

• ChatGPT can effectively generate JSON schema markup tailored for specific business types, enhancing SEO by helping search engines better understand website content.
• Users simply input basic business details, and ChatGPT provides the full schema code, which can be directly integrated into HTML, simplifying the process for those without prior coding experience.

Utilizing OpenAI’s Playground

• OpenAI’s Playground offers advanced settings to customize the behavior of GPT models, such as adjusting creativity levels (temperature) and response likelihood (top P), facilitating more precise and relevant outputs.
• The API allows users to train models on specific content, enhancing relevance and accuracy in responses, and integrating AI more deeply into business applications.

ChatGPT’s Limitations

• While adept at generating content, ChatGPT struggles with maintaining specific character counts and may require manual adjustments for precise content length requirements.
• ChatGPT may not excel in long-form content creation without additional inputs or guidance to refine its outputs.

ChatGPT in PPC and Google Ads Scripts

• ChatGPT can automate the creation of Google Ads scripts, making it easier to manage ad performance data and integrate with tools like Google Sheets.
• It can effectively identify and suggest negative keywords from search term lists, helping to refine ad targeting and improve PPC campaign efficiency.

Additional Takeaways

• ChatGPT’s ability to simplify complex information into more understandable content can transform detailed technical documents into easily digestible formats.
• The tool is evolving to ask follow-up questions for clarity, showing improvements in generating more contextually relevant and precise content.
• As AI tools like ChatGPT continue to integrate into various business functions, they offer significant time and cost savings while enhancing content quality and marketing strategies.

Episode Resources

Check out Search Marketing Academy, our new podcast: Search Marketing Academy

Episode Transcript

NICHOLAS BAKEN: Like, I think a lot of people need to understand that these tools are there to help us, right? They’re not to replace us. They’re to make, they’re there to make our jobs easier.

FREDERICK VALLAEYS: I had some ad groups running and I looked at the search queries, the search terms that Google had been showing my ads for. And in some accounts, these lists get really, really long.

Now, I want to go through this and I want to say which negative keywords should I be adding because maybe Google showing my ad for something that’s not that relevant. Hello and welcome to another episode of PPC Town Hall. My name is Fred Vallaeys. I’m your host. I’m also one of the co founders and the CEO at Optmyzr.

So the topic of the day seems to no longer be performance max, but a little bit more AI and GPT. And now even Google barred GPT 4 obviously just came out not that long ago. There’s even calls about freezing all the AI development that’s happening these days. So so clearly this is hitting a button with people.

It is potentially going to revolutionize the way Everything works. And so that’s why we want to talk about it as PPC practitioners, digital marketers, how does GPT and similar generative AI fit into our worlds for that? We did an episode a while ago. And one of the most active participants basically on the chat was Nicholas Bacon.

So we reached out because we thought he’d have some interesting things to share. He’ll be our guest today, and we’ll be talking about GPT. So let’s get rolling with PPC Timeout. Nice

NICHOLAS BAKEN: to meet you. I’m Nicholas Bacon with Nyfty Labs.

FREDERICK VALLAEYS: Well, thanks for coming on Nicholas and thanks for being such a, well, first of all, a viewer of the last episode we did on GPT and for actively participating. Yeah, we thought we had some amazing insights and it looked like you’d really been playing with the technology as opposed to being on the sidelines watching and sort of speculating what it could do.

So that’s why we wanted to bring you on and hear from you. But tell people a little bit more about Nyfty Labs and what you guys do at Nyfty Labs.

NICHOLAS BAKEN: Nyfty Labs is a first full service marketing agency and we support hyper local, local and national campaigns. We’ll deal with basically from small non pop, if they just want to throw up a, you know, a smart campaign within Google up to You know, full scale, hundreds of thousands of dollars a month.

So we can basically cover most of. Labeling as well. So you know, been in the industry for over 20 years and have experience with you know, a myriad of different things as the industry has progressed one being obviously AI now, which is seems to be the, the new toy of everybody. And, you know, a lot of people discuss, like you have all these.

AI finders you know, they’re looking at the content saying, Oh, Hey, this was created by chat GPT or, you know, some of these other iterative AI models that are out there. And, and you know what, I mean, there’s been a lot of conversation, especially on the Google threads whether Google’s is, is going to, you know, cause havoc for companies that are using AI.

And I just really don’t think so. It wouldn’t, it wouldn’t benefit them in any way.

FREDERICK VALLAEYS: Exactly. And they basically said, listen, it’s not about who wrote the content. It’s about whether the content is useful and obviously accurate. But let’s maybe dive into that topic a little bit more, right? So GPT 4.

Radically better than GPT three and a half. There’s also Google Bard, which is now available. So give us a perspective on what you’ve tried. And even maybe going back before GPT 3. 5, which is the the chat GPT. Like what is each one good at? And what have you found?

NICHOLAS BAKEN: So for us, we’ve used a lot of different ones.

We’ve been testing a lot of, of different AI based platforms. And some are good, some are okay. You know one of the ones obviously is ChatGPT. BART, we haven’t really focused on very heavily just because one we haven’t been accepted into the program as of yet. So we haven’t been able to use it.

Secondly for us, the ChatGPT side you know, it’s been one of those scenarios where we’ve been working on you Trying to integrate it into a chat solution, right? So answering questions that normally a human would answer to be where we can ask it, like for example full scale company information, right?

Like going into a knowledge base and be able to literally. Provided information just from a question. So being able to create the open API and integrate with chat. GPT has been, you know, something we’ve been playing around with. We did implement it originally in some of our sites, and we had to quickly turn it off because it started doing some weird stuff.

And we were teaching it. Hey, don’t do certain things. Then all of a sudden, you know, It would do something that,

FREDERICK VALLAEYS: yeah,

Maybe that speaks to the whole issue that you were saying, which is Wozniak and Musk basically saying, Hey, we need to put a halt to this. And there’s two levels of fear, right? I think one level of fear is that it sounds very authoritative.

And so it could feed misinformation that people start believing. And that’s obviously been a big problem across the globe for the past couple of years. And then there’s the other issue of does it become smart enough where it starts? To to improve itself to the point where we no longer control it, and it starts doing things to humanity where we’re getting more into the science fiction doomsday sort of scenarios.

I believe the first scenario a lot more than the second one. But I mean, obviously there’s debate that can be had around that as well, but, but, but, but, so, I mean, to talk a little bit about, you know, you guys marketing agency, right? But now you’re using AI. And you say you put it on people’s sites. So is this like a landing page, CRO sort of optimization play?

Like, like what are you doing with that?

NICHOLAS BAKEN: We didn’t do it on any customer sites. We did it on all of our own internal stuff. So we own a mortgage company as well, a number of other ventures. And so we wanted to utilize the technology to kind of simplify the process of people engaging with our client base, right.

Or I’m sorry, with our, with our company, like for an example, on the solar side. We wanted to give more information like on what, you know, solar panels we were using in, in different, you know times for install, et cetera. So,

FREDERICK VALLAEYS: so you were literally putting a chat bot on the basically landing page if somebody was looking for solar panels.

So they, and what were you doing before? What was that landing page? Just there was no chat. So the information.

NICHOLAS BAKEN: No, there was no chat. You know, we were, we were basically just modeling it through lead, you know, just lead flow. So from Google or Facebook, you know, other, other sources and we were feeding it into our landing pages.

We didn’t have any like chat based solutions in place that would be something of that nature. You know, you could go into the knowledge base and ask questions and stuff like that. That’s one thing. But we wanted to be more humanistic and engaging and it worked well when you were in the very first few conversation pieces, right?

So when you were asking your first few questions, it was perfect. It’s when you could ask it about anything else. Aside from that, it would literally go down a rabbit hole and you’re like, Whoa, this is, this

FREDERICK VALLAEYS: is right. So you couldn’t keep it in the box of, Hey, we’re selling you solar panels. Like, please only talk about that.

Yeah, not yet.

NICHOLAS BAKEN: And that’s, and that’s, you know, the models are being built right now still. Right. So that’s still I mean, we tested it just to see what it would do. You know, and, and it’s when we started seeing some of the feedback, we were like, Oh yeah, we have to turn this off. You know, and, and it just from the excitement of it too, but I mean, we’ve used Jasper that’s, it’s a really cool tool.

You know, just because of the fact that it’s, you know, the marketing material, like if you need a specific size, content, stuff like that, if you need it to wordsmith something the way that we’ve used chat GPT though, and any of the other AI models has been, we’ll create content. So it will be unique for us.

We’ll do all the research, everything along those lines, and then we’ll take that content and then we’ll throw it back into Sage hat GPT and say, Hey, make this more concise, make this more simple to read. You know, make sure it sounds like it’s a human. Don’t use reiterative, you know,

FREDERICK VALLAEYS: funny, actually written by humans, but you’re like, make it sound a lot more like,

NICHOLAS BAKEN: well, it’s, it’s because, you know, one thing I can tell you is most content written by humans has a lot of fluff.

Right. We do that naturally. It’s, it’s part of our writing style. It’s, it’s just the way that we think it’s part

FREDERICK VALLAEYS: of the guideline. Hey, write me a blog post. It needs to be at least 1200 words. Okay. So

NICHOLAS BAKEN: what we do is we take it to basically clean it up. Right. And not to mention, you would be surprised. If you compared it.

The chat GPT version to a human, how many times people have chosen the chat GPT one in comparison to a professional writer. And we’re not talking like someone that, you know, is just writing because they’re wanting that, but like, this is what they do as a living. So it’s very interesting. And I, and I think the tools, like, I think a lot of people need to understand that these tools are there to help us, right.

They’re not to replace us. They’re to make, they’re there to make our jobs easier. And. You know, if you’re looking for these things to replace you, then I think you’re going about it wrong. You know, nothing in life is easy and there’s a reason that’s the case. And, and nothing that’s worth having is easy either.

And if you want to have great content, you have to do the research. You can’t expect. You know, a platform to do it all for you. And then you don’t do any editing. You don’t do it. It’s just, it’s not at that level yet. And by the time it gets to that level. We’re probably going to be having very different conversations at that point.

FREDERICK VALLAEYS: And Google is still a very instrumental part of that, right? If you, when you’re doing the research, you need to start digging deep into the web and what’s out there. Great place to start. But then yesterday I had a question where I was like, my daughter was asking me what good books are there about ballerinas for like eight year olds.

And I was like I could Google that, but now I have to look through a bunch of links and then go on Amazon. And that’s, so I just went to chat GPT. I was like, give me the summary. And then I just showed it to my daughter. I was like, okay, that’s. The answer. Yeah. So they did all of these search technologies have their place, right?

NICHOLAS BAKEN: Well, you know, it’s interesting about that. And I’ll show if we can show my screen real quick. You know, one thing is that the platform is extremely good at funny enough, it’s like code base, right? So how to apply use context use like It literally will walk you through the process. It has the information for this.

Now, this is, this is chat GPT 4, right? And then just recently we were talking about, you know, saying, well, this is 3. 5, right? So it’s not the smartest of, of the bunch, but it’s pretty interesting because You could ask a different questions and say give me social content pillars for men’s haircut or hair care company, right?

And so it will go into like before and after so it’s really good at providing potential opportunities, right? Where maybe you didn’t think of something right? And then like say for an example You want to have layout template code base with Next. js, right? So JavaScript and it will literally provide you detailed information relating to it.

As you can tell, we’re implementing with MapQuest and, and some other really complex stuff. It does understand this stuff. It truly has taught how to do.

FREDERICK VALLAEYS: I just want to pause on this, right? Because a lot of folks here may not be programmers. So if you go back to the top of that a lot of people may not understand what your prompt or the top of the specific one, the code that you were showing for the template layout, right?

So So what’s nice here is if you’ve heard someone like Nicholas talk about this and you don’t even know how to set up a server or where to put the different pieces of code, what amazes me is that GPT is amazingly good at saying like, okay, here’s step one, you go to this place, make this file, put that code in.

And then here’s step two. And so it’s actually quite a nice way to learn to code. If you’ve never done it, like I know how to write code in PHP, for example, which is kind of old school at this point. But if I wanted to know how to do something else, it will actually explain where I can get a server, how I set up the configuration.

And that’s, what’s amazing, right? It can really teach us as novices. To become programmers at some level

NICHOLAS BAKEN: absolutely and not to mention like, something I want to show as well Something we’ve been doing recently and this was just to test it out again, right? Is let me see if this is the right one or not we actually had it create a basically an outline for different companies, like short descriptions and stuff like that.

Now, it wasn’t the great greatest, but it was definitely something that we could work with. And so that, that we found that to be pretty exceptional. So like, for example, like the solar style book or style guide we asked Chad GPT, write and create an associated press style book for Nyfty solar.

Right. And so what we did is provided all the detail, right? So the business synopsis, the vision, mission, the core values, et cetera. And what it did, it took all of that information and it, and it basically created an association. And this is something that’s more on the brand side, but it’s very, very representative of like language tone, how you want your company to be represented, you know, in the marketplace.

And I think that’s a really huge use case because a lot of people spend a lot of money for this type of stuff. Right. And if you can follow a template and it can basically provide that template within the confines of, you know, what your company is about. Then that just saves you an enormous amount of time, effort and resources, right?

And I find that, you know, when you can simplify my life by providing me 80 percent and then we edit it from there, that saves me a lot of money. And we did that for the brand guides. We’ve done it for, I mean, we’ve been literally testing out everything that we possibly can to see where there’s that limitation.

Now there are some limitations, right? So let’s talk about that for a moment. Some of the limitations is if you tell it to create a character count paragraph or sentence size, it is not good at that. Why it’s not good, I’m not entirely certain why but it definitely has an issue with being able to address that particular, you know, framework.

It can get close. So, you know, if you’re going to ask it to do something and say, for an example, Hey, provide me 10 iterations of a description, right? You can just say, Hey, but make sure it’s under 350, knowing that it’s probably going to go under, it’s going to go over. Right. So it’s not great at that, but some of the stuff that it comes up with is, is, is pretty good.

FREDERICK VALLAEYS: So again, let me, so to address the. Lack of Operations, Mathematical Operations for a moment. So open. I actually did an update that made it a little bit better. But the way that this generative AI works is it figures out what’s the probable next word in the sequence. And so it’s good at that sort of prediction.

It’s not good at counting and doing. And that’s sort of the problem, right? And so one technique that we found is if we want headlines that are under 30 characters, like Nicholas is saying, it does not know how to get you exactly under 30 characters. But if you give it a couple of examples towards the end, it is basically a completion algorithm.

And so it’ll tend to give you headlines that are similar to the ones that you’ve provided. So it’ll be more likely. That they will fit inside the Google Ads character limits than if you didn’t give it any headline examples, because now it’s kind of like starting from a blank slate and it may go longer than you thought it should.

NICHOLAS BAKEN: Absolutely. So something that we did is we would basically say, no, that is not the character count. And then we’d ask, well, how many characters are there? And we’d say, and then we would tell it 260 characters. And then be like, thank you for letting me know. It does get closer. You’re absolutely right. And it does learn.

So, you know, as time goes on you know, it will. It will learn more and it will be able to instinctively be able to provide better information. It’s just going to take time. And I think, you know, when you start seeing, you know, 4. 5 and some of these other more iterative models that are specifically built for that, more like Jasper.

You know, the Jasper. Let’s

FREDERICK VALLAEYS: talk about Jasper for a minute. So a lot of people may not have used Jasper. So what is it good at?

NICHOLAS BAKEN: So Jasper is actually really great at creating marketing content. And not like long, you know, it can do some long stuff, but, you know, for the most part, we kind of use it for the short stuff, like descript, I’m sorry titles and descriptions and be able to do something real fast.

You tell it basically what you want it to do and how you want it to do it. And then it will spit it out. Now.

FREDERICK VALLAEYS: And so that’s the one thing, and kind of to weigh in on that, right, so a big thing, and you kind of alluded to it way back was prompt design. The way that you get this generative AI to do a good job for you is by knowing how to formulate the question in the right manner.

What Jasper does really well is they basically have templates. So they say, oh, are you looking for a business plan? Are you looking for ads? And it’s pre formulated what the inputs to that should be. Well, how you should phrase that prompt and then it puts those pieces together. And then the other thing it is my understanding from having talked to Jasper people, but they basically leverage different generative AI systems and they will know, okay, this type of problem that’s asking for a business plan is best served by maybe Google barge API.

And this other type of categorization of keywords is done really well by GPT model three. Okay. And so it sends it to the right place with the right sort of prompt so that you get the answer that you want more easily. You could have gotten there yourself by using these various generative AIs and using kind of or experimenting with the prompts, but they give you basically a headstart if you will.

NICHOLAS BAKEN: Absolutely. And, and not to mention, it’s just like what we’ve been talking about, right? Like providing a detail of. Understanding of what it needs to be able to write about. That’s why we take the content that we have, some of our content writers, right. And we throw it into chat TPT and we’ll throw it into different platforms to make it, see if we can come up with other iterations.

That’s another thing, right? Like it’s not a matter of just like rewriting what they wrote because we think it writes better. It’s Hey, can we come up with other iterations based similar to what that person wrote rather than having that person, right. 15 different variations for different seasonalities. I can throw it in there and say, Hey, come up with different, you know, say five different iterations, say, you know, one for, or one for summer and one for spring.

And you know, throw me something right. Like in the context and it does. And sometimes it’s good. Sometimes it’s not great. You know, but it all requires editing. And I think the, the, the moral of the story is, Don’t ever remove yourself from being involved in the process of doing content creation because that’s what it’s really good at Right, but you need to be able to edit it yourself or have people edit it for you that has knowledge of that industry and then that way you have a great piece of content that you can push out In fact right now we’re in the process of creating a real estate site And that entire real estate site was actually created first by human in terms of content.

And we’re having chat GPT literally rewrite the entire site for us. And we’re going to do a test case on it to see how it ranks in comparison to others within the marketing Phoenix. Which is going to be really cool because There’s no current site up there, so we get to see it from start to finish.

And we don’t need to compare it to anything else other than their competitors. And if we start competing and we start getting better, then it just shows you that the content that we’re providing is more succinct and it’s easier to read. And that’s really what it comes down to, right? Also, one thing that ChadGBT is exceptional at is dumbing down content.

really complicated information. So if you have like a really thick document, It can write it out for you in a way that it can be read by say by you can actually tell it to write it in a third grade understanding like someone that, you know, writes in a third grade level, right? And chat GPT is really general.

You know, stuff like Jasper and, you know, and other ones out there are more specifically lately, right? Lately, we just we just got turned on the lately and it’s essentially for social media, right? And social media as I’ve noticed You know, everybody does the same thing, right? But it’s using big AI models to do some submission for doing some AI models for social media.

So,

FREDERICK VALLAEYS: So let’s let’s go into some marketing examples here, right? I think a lot of what you’ve talked about is related to marketing, but let’s get specific. So one example, and let me start here. So if you can show my screen. We, so I’ve written a lot of Google ad scripts and the other day I was like, well, let me ask for a simple ad script.

And so my prompt to GPT model for here was write a Google ad script to create a Google sheet with all the keywords, but over a hundred impressions in the last 30 days include clicks, impressions, and conversions in the report. This is a pretty simple one, right? But I didn’t think it was going to be able to do it because GPT basically learns from all the information that’s.

out there. And ad scripts are really specific. There’s not that many of them that are written, but it came back and I was like, okay, well, here’s the instructions. And then it started writing out the code and I copied and pasted this code into my Google ads account and it worked perfectly. I didn’t have to make any edits.

And it was nice because even like it says, replace this variable. In the code below where your specific spreadsheet URL. And so this is the type of stuff that I would have spent hours writing in the past, and then I would have blogged it on search engine land, and I would have had to write the instructions here.

It’s all done for me. Now I was like let me see if I can get more complex, more advanced. And I asked it if it could write an RSA spreadsheet with the RSA ad components. But I was pretty lazy and I was like, well, I’ve already created I asked my developer to write, help me with the script. And I literally sent an email that I sent to the developer.

I didn’t edit that email at all. And I didn’t think it was going to get it. But again, it just understood what I was trying to do. And boom, here’s the code that comes out of it. And then I, and then the next thing I said was, actually, I would like to This spreadsheet, which has a listing of all the headlines, the headline components and the description components of each RSA.

If I’m not using all 15 variations of headlines, use GPT itself to suggest new variations that are similar to the ones I already have. So I made that the followup question. And again, it was like, sure, I can do that too. So Basically said, you’ll have to do a few things. You’ll have to add this new function to the code and boom, that work was done in less than an hour.

Something that would have probably taken a couple of hours of coding in the past. So that’s one PPC example that really impressed me. And just, I mean, if you want to keep it simple, That first example that I gave, take your data, put it into a spreadsheet because what do we love to do as marketers?

We’d love to work with spreadsheets, right? So give me the data. I’ve all put it in the spreadsheet so that I can do something else with it. That’s one example I loved it. Nicholas, what about one of your favorite examples that you’ve done?

NICHOLAS BAKEN: It’s been basically rewriting. Titles and descriptions the application on your particular end is a little bit different just because I used Optmyzr as well.

So but yeah, The

FREDERICK VALLAEYS: Optmyzr includes AI.

NICHOLAS BAKEN: Yeah, right.

FREDERICK VALLAEYS: We use the GPT model. So if you just want to use our tool to ask for variations and then put them in, that makes it super easy. Yeah,

NICHOLAS BAKEN: it’s a little bit of a plug, right? So but yeah, that’s absolutely accurate. And, and, and it does a really great job too.

You know, and one thing is, I mean, let’s talk about Optimal in a moment and how the platform does work and why is because, you know, like some of the aspects of going into a, an account and trying to figure out, well, what changes made certain things happen, right? That’s. Would be really hard in, in an Excel sheet, right?

I mean, you’d spend hours doing that, but being able to, you know, be able to, to display that information visually in a, in a manner that makes sense you know, that, that, that is probably one of the main solutions we use internally with even our clients when they come in, it’s because we get to show like, Hey, Some of these changes that we have made, this is what the outcome is, both positive and negative and being able to see that and understand that is, is crucial.

So, and that’s a really great way of utilizing AI. And for us, you know, how we’ve used it is probably a little bit different than you because of the way that your company and platform work. For us, it’s mainly on the content side because we’re a content heavy company. And, you know, for the SEO stuff and development.

So that’s. predominantly how we use it. So a perfect example schema. Let me see if they if I still have it. So we basically asked it to provide us an outline of schema for a website, right? And what we wanted to do was make sure that it was compliant with, you know, schema for Google. And so what it did is, well, it provided us all the code for that purpose.

And then we could literally fill in the blanks. So it has an understanding of really, really basic questions on what you want it to see. And let’s take a look. Let’s, let’s do that real quick. Right. So, let’s do write schema for a, what’s this do to do that for Nyfty blending

lines with,

FREDERICK VALLAEYS: and sometimes This is GPT 4 which is a lot slower than three and a half. A lot slower. Also, it’s been at capacity a lot lately, so might have to wait a bit.

Okay, and so for people listening, not watching, so It is now producing schema markup. It says for a mortgage company called Nyfty Lending that complies with structured data guidelines from Google. It’ll include information about the organization, its services, contact details, and location. And you can add this schema to your website’s HTML to help the search engines better understand the content and improve your site visibility and search results.

It’s coming back with all of that from a simple prompt that Nicholas put in, which is just to write a schema for Nyfty Lending, which is a mortgage company. It’s basically all of Nicholas was asking, and now it’s writing the actual JSON code and it’s putting it into a little box where at the end of it, you can just literally copy that code in a clean text or not annotated.

So you can then just paste it back into your HTML or your web development software and save it and be done.

NICHOLAS BAKEN: Yeah. So it’s pretty exceptional, right? When it comes to this and say, for example, you just have No prior experience on how to do this or even where to start because, you know, the schema schema in general can actually have quite a lot associated with it, right?

So you can have subcontext to the type of company. Like for an example, a car wash, right? So, you know, being able to provide it so little, but have it do it for you, simplifies your life. Now it may not be absolutely perfect. That’s why you want to go through it and just make sure everything is in here.

I didn’t give it any other context, other, you know, Hey, this is what it is. I didn’t provide any of my address. I didn’t provide any of that stuff, which you would need in that in that process. Right? But it gives you the context of how to do it. So you just fill in the blanks, the postal code, the country, the state, you know, right?

FREDERICK VALLAEYS: I wanted a cool thing here is It’s actually, so you probably have to make sure that those numbers like the phone number there, it looks like it’s incorrect. So you’d have to fill that in. But just the fact that it knows how to put, like, is it telephone number or is it phone number or is it phone? Like how the schema specified that attribute.

Like I’d have to go and research that. I’d have to dig into Google’s documentation. That alone would probably take me 10 minutes if I’ve never done this before. And here it’s basically pre written it. And I’m sure you could ask it, are there some other attributes that I could put in? And it’ll tell you.

And similar to my Google ad script code you know, what does Google call impressions these days? Is that with a capital I or a lowercase i if I’m asking for the report? Okay, that’s the level of detail that That’s not meaningful. Like, I’m still thinking about what is it that I as a human need, but now the implementation detail is the thing that often takes so much time.

And that’s where GPT can really help save, save a good amount of time.

NICHOLAS BAKEN: Well,

FREDERICK VALLAEYS: not

NICHOLAS BAKEN: so good. I wanted to show we were working on doing redirects, right, for Cloudflare and integrated because we were doing rewrite conditions and something where you can have, like, all these different variations, right?

Can you simplify it? And basically what it does, and this is using 3. 5, so this is before 4 came out, and it would basically rewrite the condition and the information that I would need in order to utilize it as a Cloudflare worker. Which was pretty exceptional considering you know, so it’s one of those scenarios where it can do a lot as long as you know how to provide the necessary data.

Like I stated, I mean, when it comes to mark, it can give you some really, really great questions or great information. Like if you’re stumped on, you know, hitting the brick wall on content, like, Hey, what do you want to write about? Ask it to provide you 20, you know, 20 ideas about, you know, how I could use hair gel, you know, and it would so not all of them are going to be great, but it gives you a starting point,

FREDERICK VALLAEYS: right?

And now okay. So you’ve written the landing pages about how to use hair gel with 20 examples. What if we wanted some keywords for that, right? So now we could just simply give it that landing page, or we could even say it for the text that you’ve just written. Suggest a couple of relevant keywords that you could put in Google ads.

I did a session and we’ll share it at some point, but it’s basically How do you build a PPC campaign in Google ads from scratch using only GPT? So if you’ve never figured out how to get a keyword if you’ve never built an ad group if you don’t know how to write ad Texts if you don’t know what relevance is GPT will do it for you.

Let’s go to my screen share for a second. I want to show you another example that I really liked. So I had some ad groups running and I looked at the search queries, the search terms that Google had been showing my ads for. And in some accounts, these lists get really, really long. Now I want to go through this and I want to say which negative keywords should I be adding because maybe Google showing my ad for something that’s not that relevant.

So I asked a simple question. Put these keywords in a table and sort them by relevance for a company that makes PPC management software. And then I just pasted the list of all the search terms from an ad group. And I ran that. And and this is actually a really cool capability in GPT is that it can work with tables.

So if you ask for a table you can see all of these search terms, but then it comes back with its response right there. So it’s a two column table that it’s produced. The first column is the keywords and the second column, it’s put in relevance for PPC management software company. And again, this is not perfect, but if you look at the high relevance keywords or search terms and down to the the low quality ones here, you know, there’s something like PPC AI.

That’s a low relevance keyword, right? So that’s still about PPC, but it’s not really that relevant of a keyword for someone who’s selling software. And so it really helps you focus. It helps me focus my my attention on the most likely low relevance keywords that I probably should be adding as negatives.

And what’s cool here too, is. I never specified what it means to be relevant. So GPT inherently understands constructs like what is a keyword? What is an ad group? What is a tweet? What is relevance and how is relevance to keywords? How are those two related? So it automatically comes back with something that’s quite useful.

And that again, surprised me at how deep the understanding can be of that system.

NICHOLAS BAKEN: Absolutely. And that’s kind of something we’ve been noticing as well. The difference between 3. 5 to four is. It seems smarter, right? I mean, it is smarter, but just it’s, it’s more thoughtful with the information that it is providing which is interesting because, I mean, I didn’t think that last we spoke, you know, this is prior to for coming out.

I didn’t think it was going to be such an increase in capability, but it is especially on the writing side. Very, very different. It, it, it 3. 5 would reuse words a lot. It would be very repetitive. 4. 0 does not do it as much, not even close. And it asks more questions too, which is interesting. We’ve noticed where it will like, we’ll ask it a question and be, and then it will come back with a follow up question, trying to learn what we’re trying to say which is really cool.

And, you know, again, AI is going to iteratively change it’s going to apply in different ways just like we were showing on the PPC side. Yeah. I think it’s going to be extremely assisting in terms of providing tables and trying like for an example there was an article that came out about four weeks ago six weeks ago where She basically went through chat gpt provided all the keywords had it start Grouping them all together and creating campaigns through it and then uploading it into google so it has a lot of capability and it’s it’s pretty You It’s pretty smart when it wants to be, and you know what to ask it.

FREDERICK VALLAEYS: So. Right. So, again, it comes back to prompt design. And we’ll start producing a lot of content on the blog about how to design proper prompts. But if you go to my screen share again, I want to show you something else. So, a lot of people are going to be playing with GPT. through chat GPT and chat GPT, you basically get to formulate your prompt, but that’s it.

Now on the backend, there’s a whole API. So you can register for an API key from open AI in this case, or some of the other generative systems. But once you do that, it opens up quite a few possibilities. First of all, you can start training the models to be more in line with your content or how you like to respond, but it also gives you access to simple things like temperature.

Temperature and top P, these are basically measures of how creative you want the system to be. So with a top P, that’s basically as it’s writing text, it tries to predict what’s the next token or the next word to simplify that. What’s the next word that’s logically should appear in this sentence.

It then scores each of these next tokens with a probability of how likely it is it is the top P limits the creativity, because it says we should only take the 10 percent most likely tokens to continue with, right? So you’re limiting the creativity. If you say we can Any token, even if it’s much less likely to be the next one, then it’s going to go down these paths and start writing in a completely different style.

The temperature is basically a measure of that as well. So the temperature is again, a measure of how creative so lowering the results in less random completions. So as the temperature approaches zero, the model becomes more deterministic is what they say. And so what that means is a temperature of zero.

means you will get GPT to write something that’s probably already been written before. If you put the temperature all the way at one, it might come up with something completely different. So, and these are just some of the dials that you have, but if you really want to get into you know, preventing repetition of words or repeating sentences, this is the playground.

The API will enable you to to control some of that. And so here as an example, I said right about Mount Everest. And we’re going to put the temperature at zero first of all. So very deterministic. So hopefully something that’s highly factual about Mount Everest. So as we asked that it says Mount Everest is the highest mountain in the world, standing at 8, 848 meters located in the Mahalangur Himal sub area of the Himalayas on the border between Nepal.

Tibet, China. Okay. So in a lot more context, and then I’m going to change that. I’m going to ask the same question and I’m going to put the temperature at one, which is much less deterministic. Oops. We’re going to ask that again.

Okay. And we’ll submit that.

So you get a little bit of a different output. The facts still seem correct. But again, that’s sort of a, a way that you can get the system to come up with new stuff. And then of course, if you regenerate this response now, if you have the higher Temperature. Then it’s more likely to come up with something completely new as opposed to what it just gave you before.

NICHOLAS BAKEN: Yeah, it’s pretty extraordinary. And not to mention you know, something that you can do, and I don’t know how much time we have left, but on WordPress, right? There, there’s actually a platform that you can use. If you want to share my screen real quick I’ll show. Where you can actually integrate the API.

So right here,

FREDERICK VALLAEYS: WordPress plugins that are using this sort of AI.

NICHOLAS BAKEN: Yeah. So it, it basically is, is going through like it, it updates it. They’ve been doing a lot of upgrades, but as you can see, we have our key and everything which we integrate with. And so it will do title suggestion, excerpt suggestions.

Like there’s a lot of potential with this platform. And this is the one that we’ve been testing out and working on. And you know, you can basically say like if you wanted to put a web, you know, a chat bot on your website using chat GPT, this, this is one of the ones to do it. The problem is that the fine tuning so we had to remove some of that.

Some of the fine tuning, but you can do a fine tune model or you can do a database builder or a data set builder where essentially you give it information like directly. So we’ve, we’ve kind of pulled away from doing that at this particular point in just trying to focus on our models and the playground as well as what you were just seeing right there.

And you can integrate. Pretty much what you want here. And this is where open AI stuff and, you know, the API error rates and so forth. So highly suggest you play around with it. If you guys use you know, WordPress it’s a pretty cool pretty cool plugin like Fred said, you know, it’s probably going to be a lot more out there that come out.

This is one of the ones that we found to be the most complete as of this point. And. You know, it, it, it works for us and we’re going to be turning this back on the next few weeks to see how it works on, on the solar side. But this is on the lending side, which we had to turn off because it started providing things that it shouldn’t have legally speaking.

We can’t have it like that. So but it’s, it’s, it’s a pretty cool platform overall and being able to integrate it into the workflows. Is gonna be extremely important. I think overall, as a company, as a business being able to provide knowledge based information and being able to have your internal company communication and, and, and be able to ask questions of the platform to provide that data is hugely important because I mean, how long do you go through some of your own knowledge based stuff just to find what you want to see?

It’s really hard sometimes, especially if you have a huge knowledge base. So that, that, that’s where I think that a lot of the AI stuff is going to be hugely beneficial on a business case on the marketing side, obviously titles, descriptions code writing, rewriting, you know, making it more concise.

You know, even on the social side, right? Blog content. One thing that chat GPT is not good at is blog content. Long, you know, long form content. Then you have content at scale, which uses like three different AIs that, you know, builds out really long content form. So anywhere from 1200, it could be less, but 25, 3000 characters.

Right. And it’s really good. You know still you need to have human editing, but it does provide some incredible value there and if you look at this year amount of costs for having a content writer Well, you know you can do a lot with with with little in terms of time. So

FREDERICK VALLAEYS: yeah Well exciting times ahead Nicholas.

Thank you for sharing some of the ways that you use gpt in your business And we’ll definitely be doing more episodes on GPT. So if you’ve enjoyed PPC Town Hall, subscribe, so you’ll be notified of upcoming sessions. Nicholas does a lot of SEO, but as he said, he also does PPC and he uses Optmyzr for that.

If anyone else wants to try out Optmyzr, we got a two week free trial. So sign up on our website, optmyzr.com. And then of course, if you want to get in touch with Nicholas from Nyfty Labs, Nicholas, how should people find you?

NICHOLAS BAKEN: I mean, they can contact us via phone. They can email us on the website.

If they want to contact me directly, they can hit me up on LinkedIn or Twitter. I’m available. I kind of try to say this is very unique for me ‘cause I don’t, I’m not usually in the forward facing. But it’s so exciting what we’re doing right now with the AI stuff and I think where the market’s going that, you know, it’s, it’s, it’s a lot of fun.

So, you know, I’m out there they can, they can reach me at different areas if they need.

FREDERICK VALLAEYS: Yeah, so that’s Nicholas Bacon on LinkedIn. And so thank you for being with us today. I’m Fred Valles. Thank you for watching PPC Town Hall. See you for the next one.

NICHOLAS BAKEN: Thanks, guys.