How ChatGPT can make you a PPC automation wizard

Episode Description

A few weeks ago, OpenAI, the company behind #ChatGPT, conducted an event called DevDay to reveal major announcements and advancements that opened up a new world of possibilities — Custom GPTs, Copyright Shield, and speech recognition, to name a few.

After watching the event and playing around with some of the functionalities, I strongly believe that they’re incredibly useful to every #PPC marketer.

So how can we apply these functionalities to PPC?

To discuss that, I spoke to Nils Rooijmans once again, one of the premier names in the PPC scripts and automation space, in this episode of PPC Town Hall.

Nils shared his Custom GPT that can create some really powerful PPC scripts. We also discussed the announcements from DevDay with a lot of examples.

Episode Takeaways

Remote Work and Automation in PPC

• Nils Rooijmans manages a PPC agency remotely, leveraging automation and scripts to handle work efficiently while enjoying a flexible lifestyle.
• Automation tools allow for significant scaling in client management without increasing staff numbers, demonstrating the power of technology in remote work settings.

Developments in Generative AI by OpenAI and Others

• OpenAI continues to lead with innovations in generative AI, introducing new functionalities that enhance user interaction with AI models.
• Competitors like Antropic and Google are also advancing in the field, emphasizing privacy and model training quality, which enriches the AI ecosystem and provides users with multiple robust options.

Impact of AI on PPC and Script Writing

• AI advancements are revolutionizing scriptwriting for PPC, making it more accessible to non-programmers and enhancing the efficiency of ad management.
• Custom GPTs (Generative Pre-trained Transformers) allow for the creation of specialized tools that adhere to specific branding guidelines or operational methods, further personalizing the automation process.

Ethical Considerations and Future Directions

• Discussions around the ethical use of AI, data security, and the importance of understanding the underlying intentions of AI model creators are crucial.
• The future of PPC involves integrating advanced AI tools to refine strategies and execution, yet strategic control remains a human responsibility to ensure alignment with business goals.

Educational Opportunities and Community Building

• Nils Rooijmans offers workshops to educate PPC professionals on utilizing AI for script creation, highlighting a growing community eager to leverage AI for advertising efficiency.
• The ongoing development and sharing of knowledge within the PPC community help foster innovation and collective growth in adapting to AI technologies.

Episode Transcript

Frederick Vallaeys: Hey Nils, thanks for joining us again. And I hear that you’re currently working from a different location than usual, not your, water house.

Nils Rooijmans: I am, I’m, I’m, I’m dressed to the occasion. I’m in the Caribbean right now. Curaçao, it’s a small island just north of Venezuela, close to the Latin continent.

It’s part of the Dutch kingdom. And this is where I hang out in the winters to enjoy a sunshine, as you can see by the color of my face. And, nice temperatures, the sea and some PPC work.

Frederick Vallaeys: Yeah. Rough life. And I think the flexibility, and you being able to do anything from anywhere, clearly it has a lot to do with automation and technology.

And you’re one of the big scripts guys in the world.That’s what you teach people. So, remind folks a little bit about, like, what’s your main PPC angle decided like scripts and automation.

Nils Rooijmans: Yeah. So I basically run what I call a remote PPC agency. It’s a boutique agency with roughly 20 international clients and eight part time contractors that support me in doing the job.

And I was able to scale the agency to a very decent size, mostly thanks to automation and scripts so that I could create more value for my clients without needing to scale headcounts. I’d also create a lifestyle that allows me to enjoy a wintertime in the Caribbean and still, Go surfing, even in a Black Friday season.

Frederick Vallaeys: I think you are living the dream, my friend. So, yeah, hopefully, a lot of people watching this episode and learning from Nils and what he does with scripts and automation and tools. And the angle that we wanted to take today was actually talk a little bit about what OpenAI announced at their Dev Day.

And so obviously open AI, most people should know about it, but that is the company behind chat GPT, the thing that really popularized gen AI or generative AI. And, you and I, we talked about that a couple of weeks ago, a couple of months ago but at these dev day or at the dev day, they announced a slew of new stuff.

So I wanted to cover some of what they announced and what you think is exciting and what you’ve been doing with it. But yeah, so, so that’s the topic for today, gen AI and all the new stuff from open AI. But before we dive into like what’s new from open AI, maybe the latest thing that most people have heard about is all the drama that’s been happening in San Francisco.

So Neil, do you want to talk us through who’s currently in charge of open AI and does the company still exist? You tell me. Thank you.

Nils Rooijmans: You tell me, it’s crazy. Yeah, I think, I think Sam is back in charge, but you know, it’s been a crazy weekend. I think the short recap is on Friday, the board decided that Sam Altman wasn’t the right person to lead the company anymore.

so then he left. At some chat with his friends at Microsoft. Microsoft has a big share in open AI. I think it’s 49 percent where it was 49 percent back then, Microsoft decided that Sam and the other board member that left open and I could join Microsoft to start a new team there. when that rumor got out, a lot of the, staff at OpenAI decided to join both Sam and the other person.

New CEOs were appointed, new CEOs didn’t get answers from the board. and in the end, like roughly 90%, I think of the staff of OpenAI signed a. That state is something like if the board will not resign and Sam doesn’t come back, we will resign. So under that pressure, I think, in the end, Sam is back.

Frederick Vallaeys: Yes. So, but in a matter of really a weekend, the company almost. Came to an end because the board fired the CEO and I did a 700 employees. Basically all of them were willing to walk. So the good news is there’s other GPT players. So even if open AI did fail, which it’s clearly not going to, I mean, Microsoft invested 13 billion into it, but there’s some other alternatives.

Right. And so some of the ones that I’ve been enjoying the most, I guess, is Antropic and Antropic funnily enough, is actually a company that was started by some people who left open AI because they thought. They were maybe pushing the boundaries a little bit too much on not being private enough, not being safe enough.

And so they started a new company. And that company, the reason that I like it is because it’s very easy to attach documents. So it’s one of my things with GPT. Has been the limitation of attaching a file or a large piece of text as context for some piece of work I need to do. And in Entropic, you can literally go in and there’s a plus button and upload button.

You can upload five attachments, 10 megabytes each, and then you can make those part of your generative experience. So that’s one alternative. Obviously, there’s others. Facebook has their own dilemma model. That one is free for commercial use, which is pretty cool. Google has barred, but Google also has spawned too.

So, there’s a lot of activity in this space, but certainly an interesting weekend with open AI, and now back to normal, Nils, which other ones have you maybe played with and, and enjoyed?

Nils Rooijmans: Yeah, I’m really looking forward to Gemini from Google. They originally planned to release this, by the end of this year, but most probably it’s going to be the first quarter of the next year, but I expect that one to be really, really impressive.

Frederick Vallaeys: Lama, what are you saying about it? What’s, what’s supposed to differentiate it?

Nils Rooijmans: Mm, that, as far as I know, it’s all rumors. They, they don’t really publish that much information on it. I think they’re trying to really do an amazing job in fine tuning and training the raw model. Because basically with the LLMs, the large language model, the training happens in two phases.

First, you have like. A raw training process that creates a model that is really hard to communicate with, but then you have an additional training phase is called reinforcement learning with human feedback, where you basically train the model to create answers and communicate in a way that we people understand that we appreciate in our communications.

And I think Google is really trying to do a lot of effort in that part to make sure that the problems like for instance, hallucinations, will disappear. And that it will, comply with all regulations and doesn’t provide any answers that people will tend to. But yeah, so, because it’s Google and they have a lot of data, a lot of machine learning expertise and Of course, on the computing power, I think, we can expect some, some interesting results from that.

So I’m really looking forward to that one. But next to, next to, GPT, I’ve been playing with, with Lama from, from Facebook that you mentioned, it’s, it’s open. So that one is interesting, but in my experience, it doesn’t reach the level of GPT yet. I still think that, you know, OpenAI has done a great job with the latest GPT 4.

Frederick Vallaeys: I use it a lot and I think we should actually tell people for a second. When you start explaining how Gen AI works, like you actually have a background in this. Right. You studied this like what a decade ago?

Nils Rooijmans: Oh, LLMs didn’t exist back then. No, I studied artificial intelligence in the nineties. That was when feed forward networks with back propagation learning algorithms with the real thing. If you compare it to a neural networks these days, it’s it’s been is what we did back then. But ever since I’ve been following all the latest developments in artificial intelligence, me and the other alumni from my from the university, we’re in a group of discussing different sorts of applications of AI technology.

I’ve played around with it in search engine technology myself back in a Dutch search engine called Vilsa. So yeah, I’m still heavily involved with a lot of AI technologies, but I’m Honestly, I’m flabbergasted with, especially GPT 4

Frederick Vallaeys: and what’s happening. Yeah, it’s pretty amazing. And so, to give some context there too, right?

So Lama from Facebook, all of, all these large language models are basically trained on what they call, is it, it’s parameters or tokens, right? But which one is it? Has X number of parameters that it’s trained on. And so the Facebook one is relatively small. Then the Google one on two is five times as big.

And then GPT four is five times as big as the Google one. So it’s 25 times the size of the Facebook one. And so hence it’s not that difficult to grasp why GPT is probably a little bit better. And then there’s a ton of studies that are being done too right now, which basically look at what is the inflection point, how many parameters do you need to train on before the model becomes like really demonstrably good at certain tasks.

And so that inflection point, Facebook is at that level where they’ve hit that inflection point. So it is quite good. But science does matter in these things, right? And that’s why when you and I and anyone out there, It looks at how are we going to use these technologies and should we even consider building our own version of a large language model, which, by the way, is not that difficult to do anymore.

I mean, there’s a lot of open source out there that will let you do it, but to get it to be at the scale to be of high enough quality,that’s probably a losing battle unless you can also raise like 13 billion dollars for Microsoft for your own venture, like open AI did. Right, so yeah,

Nils Rooijmans: so yeah, so with respect to these base models.

They are trained on an enormous set of data across from the internet, and like you mentioned, they have a lot of parameters that need to be trained. Tuned by the learning algorithm, and that is basically what makes it so expensive, because if you have a trillion parameters and you need to update all the ways of these parameters, the ways in your neural network, that is, that is computationally very, very, very expensive.

So, but what you can do once the model has been trained, you can simply download these. The values for these parameters, the ways if you will, and then you basically have a copy of the model that you can run on a local machine. So you have like, there is a couple of open source initiatives where you basically take, for instance, the llama model, you download it to your laptop and you have a large language model running on your own laptop.

So the training is very expensive, but then using the model is less expensive and also something that’s for fine tuning the model or extending the knowledge that is being used to generate the answers. That is something that you can add on top of these base models so that you can basically create, agents, if you will, or GPTs as OpenAI calls them, that are specific to your needs.

Frederick Vallaeys: And so I think there’s a company called Hugging Face where you can look up sort of a ranking of these various models and they include the open source ones that you can then, as you said, make a copy of. So, and then I think, I mean, we are talking to a PPC audience. So one of the ways that I think about how does this all connect to search marketing?

It’s really about figuring out. Which model is the most effective for a certain task? Because the one thing we haven’t talked about is that GPT 4 and GPT 4 Turbo, like, the better the model gets, the more expensive it gets. And usually also slower. So if you’ve ever done a side by side comparison of a GPT 4 and a GPT 3.

5, like, GPT 4 is nice because you can actually read it as it’s typing it out. GPT 3. 5 will produce two pages like that, right? And then you have to sit there and read it. But if speed is of the essence, or if you’re just needing like some basic keyword analysis, you probably don’t need to spend the extra money on a GPT 4 model.

So I think that’s one of the ways that we as PPC practitioners can really help our clients by saying, Okay, listen, this model is really good for keyword generation. Maybe we use that one. Thank you. For doing some attacks generation, but then if you want to write the actual landing page or a blog, then you need to have the best one.

And maybe that’s where we spend some money on using GPT for. Yeah, I agree. Well, let’s talk about a dev day, right? So what did open AI announced and what is that next wave of the exciting things that are going to happen? So I wrote down a couple of things, but you just mentioned it. GPTs. So tell us about GPTs and what those are.

Nils Rooijmans: Right. So I think it’s a terrible name, but basically what they are, I don’t know why they call it GPTs. It’s like the word GPT in itself is already very difficult to pronounce. But anyways, so GPTs are sort of like your own version, mini GPTs that you can create. I’ve actually created some that we might demo in this session in a week or some time.

Yeah, so basically your own GPT that is very specific to a task and you can feed it information about the knowledge that it can use in providing its answers. But also for instance, the style of communication that it uses in its answers or the output. so for an example, if you have different, let’s say you have two completely different theories on how to write that copy.

You could create an agent that basically. You teach it both theories, so book A has theory A and book B has theory B. I can’t come up with any favorite or famous authors right now, but you feed it the content of the book, and then you ask it, all right, create some ad copy in the style of A and then compare it to the, similar ad in style B, and then you start testing that

Frederick Vallaeys: And then that sounds amazing for PPC marketers and marketers in general, because a lot of companies have sort of a playbook or a style guide or a brand guidelines.

And so it sounds like this is something you could feed into your own version of GPT and get it to produce ad text and ad copy that actually resonates or that, that that’s in line with your company stance. And one thing that’s interesting too about GPTs is a lot of what GPTs afford you to do Would have been possible to do in the past, but it would have probably been a bit more challenging.

You would have had to go into figuring out how to do fine tuning in context learning. In context learning, especially when maybe you can’t attach huge amounts of data, but GPTs, they are positioned by opening eyes, something that non developers can use. So you don’t need any technical skill to be able to do this now.

I know you have a lot of technical skill, but you did have a demo. So, show us. What the average person could achieve here. And then we’ll take a look at what Nils can help you learn. If you, you take his courses. Sure. Let’s try that.

Nils Rooijmans: So yeah, basically what I did is I created, I’ve used the GPT builder to create my own specific GPT bar and I’ve given it the name, the Google Ads Scripts Sensei.

And basically what this is, I’ve trained this GPT to help you create Google Ads Scripts. Now, what this allows us to do is create. Scripts that are pieces of JavaScript code that are specific to the Google Apps Scripts environment. So I’ve added a lot of examples. I’ve added the documentation from the Google Apps Scripts developer API, and I’ve added some custom instructions so that it creates code that is easy to understand and explains everything, how it works to the user of this GPT bot.

But what it can do is I think nothing but short of magic. So let’s try something I’ve prepared this morning. I’ve created a flowchart, which is an image. Actually, let me show it to Fred.

Frederick Vallaeys: So you’re doing multi modality here, so not even a text input, you’re giving it an image. Exactly. Show us that image. So is that a whiteboard flowchart?

Nils Rooijmans: Let me show you, let me show you, Fred. Because you’re also a developer, so if you see this image, You will probably understand what I’m aiming for, right?

Frederick Vallaeys: Yeah, awesome. Okay, so there’s the flowchart. I can see in the top left that you set a target CPA equals 100, and then there’s a flowchart happening. I’m glad you think I should be able to understand this, but I think anyone who does PPC hopefully gets this right. So basically it’s a big box on the left that says keyword stats from last 30 days.

And then there’s a decision box and in one direction, it says, if it’s more than 100 clicks and zero conversions, and then there’s a box that says, pause the keyword, and then the other direction goes and says, increase CPC bid, and there’s a piece of text that’s written on the whiteboard on the bottom that says, finish, send email only if change has been made.

Okay, so all visual, nothing typed out. And so you gave this to GPT to your,

Nils Rooijmans: so here I’m in my sensei and I basically say, create, I give it the instruction, create a Google, Google ads script that automate this process. So basically give it just simply. Photo of the whiteboards and this instruction.

There you go. So here we have the Google SQL Sensei.

Frederick Vallaeys: And so for people who might just be listening on the podcast version, so basically says, okay, we’ll we’ll have to break this task down into its subparts and a conceptual overview again. This is GPT talking about. So this is what it’s perceived from that image. But it says define a target CPA. Step two, retrieve the keyword stats.

Step three, evaluate the performance criteria. And then it has like two possibilities that were on the flowchart. And then number four, it says notification. And it says now let’s go ahead and translate this into pseudocode. And then it comes back with something that’s pretty close to JavaScript. it is actually JavaScript, but it’s not doing the functions themselves.

And it’s just basically the outline of these four steps. Exactly.

Nils Rooijmans: So this is how I instructed this sense I to work because in my experience that I’ve also learned giving training to other Google has professionals that want to learn how to code. There’s an intermediary step. It’s called the pseudo code that makes it a lot easier to understand what is possible, but also how the final results will actually do the work.

And if it is doing executing the instructions in the way that you intended.

Frederick Vallaeys: Because your human actually reads through this and and you should be able to make sense of kind of like the flow that’s happening without worrying about technical details. Okay, so then if you think this looks good, what happens next?

Nils Rooijmans: Yeah, so and the reason that I always instructed to to start with pseudocode is because if we are reading this. Basically, what it says for each keyword in the keywords, if the keyword clicks is above a hundred and zero conversions, then pause the keyword. That’s okay. But then it says else if the keyword has more than one conversions and the CPA of the keyword is above the target, let’s increase the CPC bit.

But that’s not what we want. Not what we want, right? We only want to increase the CPC bit if the actual CPA is below the target. So that’s why the Tudor code can be helpful. So let’s, let’s give an instruction.

Frederick Vallaeys: Okay, so it’s made a mistake from what Nils had put on the whiteboard. He’s putting in a message to GPT that basically says, for the most part, it looks good, but that they only want to increase the CPC bid when the CPA is less or below the target CPA. And this is interesting because you could skim through this pseudocode, And it’s literally just the directionality of the greater than smaller than sign that’s in the wrong direction.

So it’d be very easy to get that wrong and say, ah, it looks good. Go for it. And now you’d be raising your bids for the things that already spend too much money.

Nils Rooijmans: So it understood it. It has corrected the mistake. And now it continues creating the JavaScript code. Now, I’m going to quickly scan through this to see if it is doing what is intended here.

So you have to trust me on this, but I will also, I’m still working on the ultimate sensei. It will add a lot of comments so that you even need to know any JavaScript. Yeah, so this is, this is good. So this is literally something that you can copy paste and run it in the Google S scripts environment and it will work.

So we started off, we started off with simply this logic that you and your team created on the whiteboard. We took a picture, uploaded it to the Google Ads Scripts GPT that I’ve created. And there you go. Here you have a Google Ads Scripts that you can run in your app.

Frederick Vallaeys: Now this, GPT that you’ve created, can other people use this or how do you distribute this?

Or is this now for your team’s use? Right.

Nils Rooijmans: Yeah, great question. So currently this one is still private, and the reason for that is I’ve uploaded a lot of my own scripts. In the instructions, and you can hack currently, you can hack the GPTs to basically rip the, the original instructions out of this, this part, I think open AI is working on a fix to prevent this from happening.

But I don’t want to share all my scripts yet.

Frederick Vallaeys: Yeah, so you’re kind of alluding here to copyright actions for authors. And that’s actually another interesting thing that was announced at dev days. Is that GPT or open AI as a company, they have something they call copyright shield. So when you use one of their off the shelf models, or the enterprise models, they will actually protect you in court.

They will fight on your behalf if you get sued for copyright infringement. So that should add a lot of comfort for people using GPT. Now, I think when it comes to custom GPTs, I’m not sure that those would be covered by the copyright shield. And I think you and I both sort of feel that frustration because if you today go into GPT 4 and you ask, Hey, can you write me a Google ad script?

A lot of what it’s learned is from stuff I’ve written, from what you’ve written, from what, you know, a handful of us have written. and I’m actually quite happy. I mean, I think this is great for the industry that people can sort of push the boundaries. I mean, I think it’s fantastic. I think it just helps us as an industry evolve to that next level of PPC management.

And we’re all trying to figure out like how do we optimize for performance max, right? It’s different from how you used to do things. And I think the more smart people, Who can start writing like little pieces of code and then getting scripts together and hopefully share those scripts back to the community.

I think that’s only going to lead to, much better things for everyone.

Nils Rooijmans: I agree. So the competitive advantage. It’s definitely not in being able to write JavaScript. The competitive advantage is in being able to translate the opportunities of Google script automations. Into, ideas. So coming up with ideas on, on how to improve ad copy using, using scripts, using the GPT API, coming up with ideas on how to, use the GPT API to help you with, for instance, negative keyword mining, search term report mining.

Frederick Vallaeys: Yeah, I think it’s just, it’s so fantastic. I mean, You and I both were saying, listen, scripts are a way for anyone can automate their PPC accounts, but there was still that need for a little bit of technical savvy and being able to put the JavaScript together or grab like pieces of your code and combine it with a piece of my code and put a little bit of your own code and like that’s that requires some programming knowledge, not the super high amounts.

And that’s the classes that you basically do, right? You teach people how to get to that point where anyone can do it. Yeah. But now with GPT, it’s basically like you have your own private developer who’s got, who’s ingested everything there is to know about your type of code and you can have a conversation with that.

And it’s not just like, hey, here’s the, here’s what you need to write, but you can actually have a back and forth and it can make suggestions. It can say, hey, have you considered that? Maybe it’s the wrong approach. So one thing that I did was with OpenAI’s advanced data analysis, quote unquote code interpreter, so that they’ve been naming this thing back and forth.

They still call it code interpreter for the API, but they call it advanced data analysis if you’re using the chat GPT capability. But this is a system that writes Python code. And so this is great because you can have it ingest huge files of CSV of data from your account, your ads account. And then you can do things like say, hey, could you predict future ad spend based on the historical ad spend of these campaigns?

And it’ll actually talk you through different statistical analysis methods that it could use. And you can have a conversation. You can challenge it and say, I don’t like that one. Suggest a different one. Or then it’ll come back and it’ll say, Hey, did you notice that you have a little bit sparse data for these two months?

So maybe your prediction is not [00:26:00] going to be the most accurate. And again, this is the type of stuff where in the past. I would have had to sit with a developer or I would have had to do it myself. And then, you know, you email your developer, they get back to you the next day. And the next thing you know, it’s, it’s two weeks to develop a simple methodology, right?

But here, you can literally sit down at the computer with the GPT that never sleeps. And, and go through that same process and get something working within a couple of hours.

Nils Rooijmans: It’s actually, yeah, it’s, it’s literally like that. I remember. Doing some advanced data analysis that, you know, without ChappyT would easily take me two or three days to actually implement this regression model.

And thanks to Code Interpreter, I was able to do it in a couple of hours and I had a working model with the regression analysis. It’s crazy. Yeah. And also you mentioned, yeah, it’s, it’s a couple of weeks ago, I, I run a, an online workshop. It was a five day chat GPT and scripts challenge where me and roughly 40 people, where we’re using chat GPT to create scripts was a very nice experience, but I basically educated them on how you can use, chat GPT to get familiar with the basic concepts of JavaScript.

So for instance, what is a variable. What is a function, what is an object, etc. And once you have that basic understanding, and you’re learning in a playful way with ChatGPT, how you then can create prompts that allow you to create these scripts. With the use of chat GPT, but now that you can build your own GPT like the one I just showed you this inside, you don’t even need to learn how to create prompts anymore because the GPT developer has done that for you.

Frederick Vallaeys: Exactly. And then the other thing that was announced at Dev Day is the new GPT models are better at following instructions. And so this is part of what you’re talking about, right? It’s like usually inside the prompt, you would have to specify the output that I need is a Google ad script. And sometimes it would do it, sometimes it wouldn’t.

But so these new models are better at following specific instructions. And so if you ask for it to come back in a certain, programming language, it will always give you back that thing in that programming language. So that’s going to get better. And then the other thing that I thought was really cool is the integration of what’s called ‘function calling’ they also sometimes call it tools and models, but basically you can have these GPTs.

Like what we’ve shown you now, right? What you just showed us was go and get some data from Google ads. Now, the one thing is you’ve produced that code, but you still have to copy and paste it into Google ads to do anything because the Google ads data that sits in Google and that doesn’t flow into GPT.

but there are now starting to be integrations where you can actually. Call outside of GPT and pull data back in. So if you think of a simple example, like if you want to have a conversation with GPT and ask it, what’s the weather currently in Curacao? Well, it doesn’t know because and the latest model, by the way, has been updated to April 2023.

So GPT for turbo is not through April 2023, but still it doesn’t know what the weather was today. How does it do that? How can you go to Google? How can you go to Microsoft and ask a question and actually get today’s data back? Well, it’s because it does that in context learning, so it reaches out to an external API, and it says, give me the current weather temperature in that location.

But that’s a structured call that has to go to like, better underground or better dot coms A. P. I. That comes back and the G. P. T. Is able to use that Jason that comes back, pull out the temperature and then that piece of information can now be part of the verbal response that it gives for the written response that it gives.

And this is really cool because temperature is a very easy example. But yeah, what if we could integrate and directly pull Google ads data back in? What if we could pull in data? So you pull in your Google ads data and you push it back out to Canva to do like this really creative ad [00:30:00] design using the dolly three, which is now also in the API, right?

So you start putting all these pieces together and in the past that was a lot of work because you had to literally like figure out if I want the temperature, I need this API. If I need Google ads data, it’s this API, but the system is just getting better and it knows, Oh, that kind of data probably lives there.

And it goes and fetches it and you can do multiple fetches in a single call. Or you could. Put it, push it back out. So that’s really amazing what you’re going to be able to do in a, in short order here.

Nils Rooijmans: Yeah. The innovation is going at such a rapid pace. Literally it’s, it’s going to be hard to imagine what we’ll be able to do a few months from now.

Literally today, I discovered that there was a guy who was using the Zapier integration with an. I video to he basically shot a 30 second video of himself chatting like we are chatting right now. The platform did an analysis of his the mimics of his face, his voice, and he was able to send data through within chat.

Data through Zapier to the AI video generator, reintegrate the video in the chat and use that for instance, to send an email to his client where he himself, the AI generated video of himself was presenting the results of last month. I think that is just crazy. Nice. This is what is possible right now. It is a whole new frontier.

Frederick Vallaeys: And so it’s interesting too, because I think what you just showed us with the, the Google ad script sensei, that’s basically kind of replacing what you do. Right. Like, what do you, what do you think we’re going to be doing say, like a year from now? Yeah, hopefully we’ll still be chatting with them.

The bots are doing the work for us.

Nils Rooijmans: Yeah, in all seriousness I think, you know, like with the GPT assistance the unique strengths that we have with, when I say we, I mean, the PPC community is that we understand how to translate business objectives into the search marketing opportunities, right? So, search marketing still is something that is changing every day.

The platform is changing, competitors are changing, your client is changing. And the AI, the GPT assistance, they’re just going to be tools in our arsenal. Like, like for instance, Specialized juniors in your team. So if you think of, think of all these AI assistants as specialized juniors in your team, you can simply build them with the GPT builder and tools like that.

You’ll be able to, to create much more value for your clients in less time, but you still have to come up with the strategy that is, fed by all the. Well, the capabilities of all these different, AI systems. I think that is, that is where I see the biggest competitive advantage for PPC professionals.

Frederick Vallaeys: Yeah, I totally agree. I mean, at least for the next foreseeable future, you know, we’re in charge of that strategy and we still Tell the machines what it is you need to do to achieve that certain goal, right? And what are the steps involved in it? Because even as we saw in a simple example you just did, it got the directionality of like, is the target CPA good or bad?

Like, it got that wrong and it’s pretty simple stuff. We can pick that out. But if you do it wrong, then yeah, you’re going to, Waste money on your ads. I mean, I, so I live here in the Bay area, right? And one thing that happened recently, which was very unfortunate and a little bit more severe than messing up the PPC campaign, but it was what happened with Cruze is the self driving car division from General Motors.

It’s headquartered in San Francisco. And if you go to San Francisco, all of these robo taxis were running around, no driver in the front seat. And I’m sure you saw this in the news, but a woman was hit by a car. It was not a self driving car. It was not Cruze. And she got thrown into traffic and there was a cruise car coming towards her.

And that car, the automated car braked three times faster than any human could have. So that minimized the impact for the woman, but then the woman got stuck under the car and the car didn’t sense this. And so the car’s protocol, the car strategy for dealing with an accident is to actually pull off to the side of the road to clear traffic.

The poor woman was stuck below the car and got dragged. 30 ft to the side of the road. And that’s where most of her injuries were from. So kind of like two lessons here, right? Like you’re saying it’s about the strategy. What do you do if something goes wrong? Do you have all the right sensors in place when that comes to PPC?

Like, are we looking at the right metrics? And that’s where we often confuse ourselves, right? Like the example that you showed, like if your conversions are greater than zero, Yeah, but what is a conversion, right? Are you measuring the conversion that actually matters for your business? So focusing more on like getting that right and then the systems can automate handling that.

That’s good. So, but I think what we learned from this whole cruise example is we got to monitor the right things. We have to have the right strategies. We have to have some sort of like insurance in place so that if things go wrong, like it is handled correctly and the humans are there to step in to avoid things getting even worse.

So I think we can learn across industries here. But those are some of the human roles that I see and it’s still very much in line with what I put in my book, the PPC doctor, the PPC pilot and the PPC teacher, all of these super relevant in this open AI world, I think.

Nils Rooijmans: Yeah, I agree. I mean, you know, we’re the overlords.

We have the oversight and it is our job to make sure that the AI is not going haywire in the direction that we would not want it to go. Right. I think what we see happening with a lot of the automations within the Google Ads platform that are powered by Google AI, like for instance, B Max or Smart Bidding.

These clearly show that there’s huge potential, but there’s also at least a slight risk that, you know, the machine doesn’t always perform in our interest or makes mistakes that are costly, and it’s a hard job to, to, to spot them.

Frederick Vallaeys: And that’s really interesting because, so we’ve talked about the mistakes, but the other point you’re bringing up here is like the overlord.

And if the overlord in that case is Google, It is Google’s interest that’s being prioritized and not necessarily that of the advertisers. I think Google oftentimes does have advertisers interests in mind, but at the end of the day, they’re also beholden to the stockholders. So that’s an interesting point that you raise, like whoever builds the AI is in charge of saying what the AI should do in certain situations.

So, that’s what I think. Like, listen, we, as a company, as Optmyzr, we’re never going to fight against Google. We can not do AI better or in more real time than they can. They just simply have more money to have more resources and they can do option time manipulation and manipulation, not in a bad sense, but just setting the bids and like changing what the page looks like.

We cannot do that. So the best that we think we can do is give you a layer of control back. And say, okay, well, if Google did this thing, which they thought was the right thing, but we perceive it as being the wrong thing, then let’s quickly go and correct some of the settings that we can to bring them back to where we need them to be.

Basically we’re putting the guardrails in place and giving you control over those guardrails. Maybe that’s the only interesting thing is, as people start using different GPTs built by different people really understand, like, what’s that person’s motivation who built it, right? When it comes to NILS, yeah, he’s just trying to help the community and make all PPCers better, so you can probably trust that.

But if there’s some person you’ve never heard about putting a script out there or a script generator in GPT, well, God knows, maybe the instruction for every one of these scripts is, Point at the end, send all the data to my email address, right? Now they have all your data, so be careful what you have, what you look at.

Nils Rooijmans: This is actually a great example, but because there are some scripts out there, especially some older ones that sort of like an have an integration with Google Analytics, where every time the script is executed in the Google ads platform, it pings to Google Analytics API, telling the owner of that Google Analytics property that a script has been executed.

it’s not sharing any specific or Google has specific data with the The publisher of the script, but still, if you don’t know code or you didn’t ask chat GPT to tell you what the code is actually doing, then you might not see it.

Frederick Vallaeys: Yeah, and the nice thing, and you alluded to this is you can actually get a piece of code and then you can, because it’s like having a conversation with the developer, you could ask you the question, like, where is this data being sent and it’ll probably tell you there’s this email address right there.

Right. This is what’s being communicated back to it. So the same questions you would ask a human developer, ask your robot developer and get the answers you need.

Nils Rooijmans: I’ve actually shared a Google doc with some GPT prompts that you can use, especially, especially for this. I will, I will link to them in the, in the comments.

Frederick Vallaeys: So awesome. That’s fantastic. You know, crazy times, but I’m super excited. It sounds like you’re super excited too. And do you have some more classes coming up to, to teach people how to do scripting and GPT?

Nils Rooijmans: Definitely. Yeah. So, four weeks ago I run this, five day script challenge with chat GPT.

I’m going to repeat. It was a huge success. So I’m going to repeat it in February next year. So basically what that is, is five day. Per day online workshop where you and the other persons that are joining are gonna use GPT to create, Google Ad Scripts without the [00:40:00] need to learn a lot of Java scripts.

So we don’t need to become a ‘cause, we’re gonna have GPT create Google Ad Script.

Frederick Vallaeys: Okay, so don’t be afraid if you don’t have any technical skills so long as, you know, a little bit of p pc. Where can, people find that? What’s the url?

Nils Rooijmans: I’ll definitely link to it in the, in the show notes, but, you can find it on my website, which is my name, neil com slash jett challenge.

Frederick Vallaeys: Great. Awesome. Hey, well, Nils always, appreciate having you on the show and sharing some of the latest and greatest, really love that custom GPT that you’ve built. Yeah. So thanks for joining us. Thanks for having me. Yeah. Enjoy your winter in Curacao. I will.

Nils Rooijmans: Get some tan.

Frederick Vallaeys: Yeah. Make that head shine. All right.

And thanks everyone for watching. And, we’ll be back with another episode soon. So if you like this, go ahead and subscribe. See you on the next one. Take care.