How to level up your PPC game with ChatGPT and PPC scripts

Episode Description

With the launch of GPT 4 and the added functionality of plugins, what are some new applications of ChatGPT? And is there a way we can combine the power of GPT and scripts?

To learn more, I spoke to Nils Rooijmans, one of the premier names in the PPC scripts and automation space, in this episode of PPC Town Hall.

Nils shared his hands-on experience with GPT and the scripts he built with a lot of examples. And I spoke about how you can make ChatGPT work for you with custom instructions and some powerful plugins. I hope you enjoy this conversation :)

Tune in to:

- Learn how Nils created a graph of n-gram analysis of Google Ads search term data using ChatGPT in only 5 minutes

- Learn how to get better responses from ChatGPT with custom instructions

- Get hold of Nils’s list of over 400 PPC scripts

and more

Episode Takeaways

• Improving ChatGPT responses with custom instructions:
  • Setting custom instructions can significantly tailor ChatGPT’s responses to specific needs, enhancing relevance and accuracy for professional use.
• Access to Nils’s collection of over 400 PPC scripts:
  • Nils offers a vast library of PPC scripts, which can automate and optimize various aspects of PPC campaign management.

Additional Takeaways:

• Integration of AI in PPC:
  • The discussion underscored the transformative impact of AI, particularly generative AI, in streamlining and enhancing PPC campaign management.
• Empowering PPC professionals:
  • Advanced tools and scripts are enabling PPC professionals to perform complex tasks more efficiently, potentially reshaping their roles to focus more on strategic oversight and creative tasks.

Episode Resources

Nils’s list of 400 scripts: https://nilsrooijmans.com/free-google…

Episode Transcript

NILS ROOIJMANS: Some of my clients would, would send me a request to create a script for them. And if, if I thought that it would generate some interesting scripts, I would say, okay, I’ll, I’ll take on this job. And I could either do it myself or I would write the requirements for a junior developer to write the script for me.

Right. But currently with GPT, I have like an army of interns or junior developers. Because literally, so in the past, it would cost me like two days or maybe even more to create a script myself. I can do it in two hours right now, or I would outsource it to a junior developer. But now I can simply outsource it to chat GPT and it only costs like 20 bucks a month.

FREDERICK VALLAEYS: My take is that a lot of agencies, unfortunately, tend to do a little bit of a cookie cutter job and they make money through skin. Right. I mean, so they, they might be able to help. A company that doesn’t have an in house team that doesn’t have a budget for an in house team But so they get the benefit of the scale Working with an agency, but the downside is that sometimes agencies tend to do the same thing for each account.

And so that’s a little bit what you’re going into now that like the real value that you bring is like the ideation, the creativity, like the new strategy, like how do you connect the strategy to your business data? You’re the world that you operate in the industry that you operate in. And I see in house teams being much more active in that field.

field, they’re always thinking about like, Oh, we have this data set, which could influence PPC in that way. And then they’re thinking like, how do we connect that? Right. How do we automate it? And that’s where scripts are so brilliant because rather than having to do this manually, you can turn it into a process and make it repeatable.

It’ll probably then tell me something along the lines of, yeah, but I do need some data for that and actually gives me a step by step breakdown of how that would work. Right. And what’s interesting is in that work itself, you can open that up and that shows the Python code. So you can copy and paste that Python.

You can actually run it on your own system if you wanted to. But the, the other nice thing is you don’t have to run it on your own system. It’ll just run it inside of GP. 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 CEO and Optmyzr.

So for today’s episode, we’re going to talk about something that’s really near and dear to my heart. It’s the topic of automation in PPC. And what it means for PPC professionals in terms of their careers. Now, for a long time we’ve been talking about automation, but there’s one huge change that has happened in the last year.

And that’s the introduction of generative AI, a new type of AI that does some completely different things than what we’re used to. And to talk about that, as well as talk about scripts and everything else related to automation, Who better to bring in than my guest today, Nils Roimans, who’s been writing scripts for a very long time.

He’s a huge expert in that field. He speaks at SMX Advanced in Berlin on that topic. So let’s hear what Nils has to say. Thank you for joining and let’s get rolling with PPC Town Hall.

Welcome Nils. Good to have you on the show again.

NILS ROOIJMANS: Yeah, thanks for having me Fred. Always fun to chat with you about our favorite topic, PPC and automation. And yeah, it’s great that you mentioned Berlin because we’ll be meeting up again, right? Berlin.

FREDERICK VALLAEYS: Yeah, so I don’t know if people are going to see this before or after we are in Berlin, but it’s a huge honor, obviously, for anyone to be invited to speak at SMX Advanced the more advanced show where you really have to know your stuff.

Otherwise you get booed off the stage. So, and I see you’re doing a session. Is it, is it just on scripts or what are you going to be talking about?

NILS ROOIJMANS: GPT and scripts. Yeah.

FREDERICK VALLAEYS: GPT. Okay. And I think I’ve got the session right before you. And guess what I’m going to be talking about?

NILS ROOIJMANS: If I’m correct, you’re going to be talking about large language models, right?

FREDERICK VALLAEYS: Yeah, GPT and scripts. Well, they’re all supposed to be scripts. But basically the, the angle on GPT is how to, Influence the large language model to be more grounded and hallucinate less. So some business examples of that, but I mean, so we’ve jumped kind of straight into it here, right? Like large language models and generative AI.

Give me your take on what is this new flavor of AI that we’re playing with and where is it useful? What have you tried?

NILS ROOIJMANS: Right? Yeah, I think it’s amazing. I’ve been active in, in, in, in artificial intelligence since the early nineties. Sort of showing my age here, but I studied artificial intelligence back when when feed forward networks and multi layer perceptions still were the thing and i’ve been playing with artificial intelligence ever since And to be honest, I did see something like gpt2 or gpt3 coming But once I started playing with chat gpt, and especially gpt4 I was really caught by surprises and I think we’re really in a special time Era when it comes to ai developments and the power of a large language model and what they can actually achieve Especially with some of the latest developments in gpt the the code interpreter I think I can show you a couple of examples of what i’ve been doing with that I think it’s just amazing how far these things are going to get and also the rapid pace of innovation It really has surprised me a lot.

FREDERICK VALLAEYS: Yeah, and I mean there’s a lot to unpack in You That opening right there, but the pace of innovation, I think that’s a, let’s start talking about that first, because I’m feeling a little bit of buzz that people are like, Oh, well, chat GPT was school and I had fun talking to it, but like now what do I do, or it’s not been that good at answering my questions.

They’re like, it’s not getting better. You know, what’s the real business applicability. And I think to your point, I mean, it’s evolving super rapidly, just like technology always does. And something that does, that’s not working great today. Give it four weeks. It might work. Fantastic. One specific example that I’ve kind of seen a lot was accounting and math like it’s not that good at math, but it does seem to be getting better.

At it all the time and that’s pretty important when you do ppc, right? Well, we’re so numbers focused very much Yeah, but but what what’s your take on the the pace of innovation? What what do you think is going to drive it forward?

NILS ROOIJMANS: Right. Yeah, so so definitely the combination of more data more computing power more fine tuning of both the the training algorithms and the models itself One of the topics you’re probably going to talk about today and hopefully in Berlin will be the fine tuning of existing models versus training your own models.

But then also again, the companies themselves like OpenAI and Google, they will probably release different versions of these large language models and other generative AIs that will surprise us. The next few years.

FREDERICK VALLAEYS: And I think these are all really important things to go into, right? Because we all, most people listening are PPC experts.

So like TRO, STCPA, CPC, these are the things we speak every day. We know what these things are. But now we’re talking about fine tuning. Now we’re talking about embedding grounding temperature. Like this is a whole new vocabulary and it’s really important, right? Because these are ultimately the levers that we have to get the thing, the GPT thing.

For large language models to, to output what it is that we want. So I mean, let’s talk about that a bit, right? And I’ll start here, but fine tuning is basically taking an existing large language model and is giving it more examples of the type of prompt and response that you want. And so very simple example would be to take as a company, as Optmyzr, we would take all of the support tickets that we’ve had.

And we would say the question that the person asked over email, that was the problem. And the response that was written by my support team, that’s the response. And so this is a structured data set that gives examples and it teaches the model. Okay. If this is kind of the question, this is sort of what the answer should be like.

So we did this and one very fascinating thing started happening. Whenever we asked a question to this fine tuned version of the model, it would always sign its name. So it would respond as if it was a human, like a support person. And it would always sign the name Juan. And why was it doing this? Well, it’s because a lot of the support tickets that it was evaluating were answered by Juan.

So the, the large language model started assuming the role of the the personality of Han. Right. And so that’s great because he’s actually one of the people. He’s a great person on our team. And that is better than having it be some random response. So that’s fine tuning. Do you have any thoughts on fine tuning before we talk maybe about building your own LLM?

NILS ROOIJMANS: Yeah. So, so one of the funny things we noticed with GPT is that a lot of people started playing around with it, but then they discovered that It certainly has its limits, right? For example, you can use GPT to create Google ads scripts for you. I think that is a use case that is very exciting. And I’m actually in the progress of creating a workshop for people who want to learn how to do that.

But to be honest, the default version of GPT for currently has a rough time in coming up with. Certain parts of Google ad scripts that are relatively new to the platform. So, so keeping up to date with the latest possibilities of the Google ads query language the latest API releases that is very difficult for GPT and they, the scripts that it generates surely have some mistakes and omissions there.

So I’m trying to come up with a way to train the model to actually make use of the latest documentation that Google has published on Google adscripts so that you can make better use of GPT to create Google adscripts for you. And fine tuning is one of the ways to do that, but most probably creating a different large language model with, with its own embedding for Google adscripts is, is the preferred way to go there.

But I’m not there yet, but

FREDERICK VALLAEYS: there And let’s explain that to people who are, you know, maybe not our most technical listeners a little bit, but so building a large language model is extremely expensive. But the good news is the, the way to do it is pretty well published. There are open source ways of building your own large language model.

And so to build a large language model, basically it’s very simply you ingest whatever. written text, code that you want to teach the machine. It’s, it’s your teaching set, basically. And so by the way, generative AI is a subset of machine learning. So it’s, it does work on learning, right? But you teach it whenever you want to teach it.

Now this is kind of fascinating because one example I always like to share is if you ask the GPT 4 model, a question about gun control. and a question about religion. The religious answer will tend to be sound like a republican person, a republican conservative in the United States, whereas the gun control question will tend to be answered as if you were a democrat, more of a liberal person in the United States.

And now why is that? Well, it’s because more stuff has been written by liberals about gun control and conservatives about religion. So that’s why this large language model starts to emulate these patterns. And so if you build your own large language model, You get to choose what goes into it. So you could very well say, like, listen, I’m only going to put in the liberal viewpoint or the conservative viewpoint, or in your example, just add scripts that you validated that, you know, work.

And now that’s, it’s the, it’s learning set And so from that, you’re gonna get something that then tends to output responses that are better. And also it circumvents the problem that GPT, its learning is up to, what is it sometime in 2021? I don’t know if they updated that yet, but that’s exactly the reason why any of these new Google ads query language queries, it may not know the latest things.

So that’s a problem.

NILS ROOIJMANS: Yeah, that’s a, that’s a great explanation. Yeah. So, and one of the other things that. In the past, it was relatively hard to, to use your own data in GPT. So for instance, if you were to ask GPT give me some keyword suggestions for this URL. It could not actually fetch the content of that URL.

It couldn’t scrape your landing page and then look at the content of the website or the page to, to help you with the keyword research, but thankfully now, thanks to a lot of plugins in GPT and also the code interpreter, it is able to actually use data. That is of your own, your own PPC data or third party data from the web.

And you can actually use that data to feed the GPT AI with content that will increase the quality of your of the answers it gives to your questions.

FREDERICK VALLAEYS: Exactly. And this is the pace of innovation, right? So if you are still used to the old chat GPT. For, for the most part what you’re talking about, you do have to be a subscriber, so you do have to pay your $20 a month and then you get access to betas.

And so two betas that we’ve talked about here that absolutely you should try right, are the, the code interpreter and the plugins. So the plugins is the other one you mentioned? Yeah. There’s another one. Don’t know if this one is part of the paid version, but it is the custom what is it called?

It’s the custom.

NILS ROOIJMANS: Right. Yeah. Yeah. Custom instructions. Yeah. I think it’s, you have to pay for that as well. Well, that is a no brainer because GPT it’s like you have, like for instance, you and me, we spend a couple of hours every, I don’t know, week, month creating Google ads scripts. And in the past, sometimes some of my clients would, would send me a request to create a script for them.

And. If, if I thought that it would generate some interesting scripts, I would say, okay, I’ll, I’ll take on this job. And I could either do it myself or I would write the requirements for a junior developer to write the script for me. Right. But currently with GPT, I have like an army of interns or junior developers at my fingertips.

Because literally so it in the past it would cost me like two days or Maybe even more to create the script myself I can do it in two hours right now, or I I would outsource it to a junior developer but now I can simply outsource it to chat gpt and it only costs like 20 bucks a month, which is really cheap, right?

FREDERICK VALLAEYS: kind of the the thing here is When you work with a junior developer, you have to give them very specific instructions. And sometimes you have to assume that they are not well versed in what Google ads is and what, how does a campaign relate to an ad group? And right, if you don’t have that specificity of how things interact, that junior developers get it wrong.

And so if you’re going to spend that time writing those instructions, Anyway, and then you’re basically just you got the cost of paying that person And you got the two days of waiting for them to write the script and then give them feedback and go put this back and forth Right, that’s where you’re wasting time And that’s where if you’ve written the instructions already you might as well give them to gpt And i’ve literally done this i’ve taken some instructions that I wrote months ago this was actually my presentation of friends of search in amsterdam.

I think you were there But I took literally the email that I’d given to my developer and I gave that email to GPT and it just wrote the script. It didn’t need any other information and it worked. It was amazing.

NILS ROOIJMANS: Yeah. Great example. And, and to be honest, you touched on a point that I think is very interesting also for a lot of PPC professionals.

Because you and I, we are both able to sort of like. Right. Pseudo code, right? We sort of know how to create the script so we can give a high level description that has enough details so that junior developers can create the script for us. And that skill in itself is becoming more valuable, I think, thanks to the current developments in AI.

So the seniors that do know how to very explicitly create requirements for it. It could either be ad copy or creatives or data analysis or, or code the ones that will be able to, to clearly define the requirements will, will, will, will gain superpowers with these new AIs.

FREDERICK VALLAEYS: Exactly. And a good example of like the ability to write pseudocode.

So I had a new set of instructions and I wanted to, and this script is on search engine line so you can go get it. But basically I wanted a narrative description of a change in performance, one period versus the previous period for Google ads accounts. But I literally didn’t want to do any work on this.

Right. So I went to GPT and I was like, you’re going to write the script that pulls the CSV data. Which I’m then going to feed into GPT to produce the text. Now, when I went to GPT and I said, pull two date ranges and compare them to each other, for some reason, it said, I’m going to write one Google ads query asking for two date ranges in a single query.

And you and I, and any developer knows that that is not possible. That’s not how it works. Right. But so again, and again, it would produce the code. I would put it into Google ads and say, there’s an error. There’s an error. There’s an error. And so. You know, I’ve used it enough that I could read it and be like, that’s kind of weird.

You shouldn’t be asking for two date ranges in a single query. So I very simply updated my statement. I said, First pull the data for the last 30 days, then pull the data for the 30 days before that, and then join it together on the campaign name as the common key. With that instruction, it was like boom, boom, boom, script was done.

It just worked. And then, but that’s the power of pseudocode, right? It’s like being able to take it to a level of specificity where there’s very little ambiguity about what needs to be written. And I think that’s also a little bit the art now of writing GPT or using GPT to write scripts. But like, you don’t want to go super specific from the beginning because a lot of stuff, it just knows how to do, but it’s knowing when to go a little bit deeper.

That that’s really quite helpful.

NILS ROOIJMANS: Right. Yeah. And you could actually also ask GPT itself on what will be a logical next step to improve the quality of this script. So one of my students actually sent me a prompt and they, they, they just, they just plucked a, they got a script, a third party script from the web and they asked GPT, how would Nils Reumanns Improve this script.

And it gave a flattering answer as in presenting me as, as, as a Google ad script authority or something like that. And then it continued in the way that I would, how I would refactor the script or rewrite the script to make it to make it even better. And these suggestions were just, it was like, it’s reading my mind.

I’m out of a job here.

FREDERICK VALLAEYS: All right. Hey, people being out of a job. I mean, I think that opens up the next great topic of discussion, but you know, I’ve, I’ve written a book about it and I think, do you have a copy of my book?

NILS ROOIJMANS: I’ve got it here, Fred. I was lucky enough to to receive the manuscript. For review before it was published a little bit of the camera that we can’t see the contrast there There you go.

You can see I was actually reading you can see the underlining. So so I actually read it and I did my job Yeah, yeah Yeah I mean Just like just like you I feel like automation layering is the way to go with when it comes to ai

FREDERICK VALLAEYS: Exactly. And so I mean and thank you for all the feedback you gave on the book.

Yeah But yeah, I mean, so let’s talk about automation layering and then let’s also talk about, so when, when I wrote that book, that was a number of years ago, that was before generative AI. And one of the points in the book is like the last bastion of human, like the thing that you absolutely will have to do that the machine will not be able to do is like be creative.

Wow. Eat my words, right? I should be eating the actual paper that it was the book was written on because here we have gpt It could probably have written the book for me and I didn’t see that coming. You clearly did see it coming because you were involved in gpt2 way back when but like talk about that nuance, right like How is this creative component and it’s writing scripts for us and it’s writing a text for us and it’s like you can even use it to build webpages.

Like how is that changing what we do? Is automation layer the answer?

NILS ROOIJMANS: Well, well, seriously, let’s take let’s say we were professional PPC, Google ad script developers, and that that is all we did, right? If we define our job as merely generating JavaScript code, Then surely we would, we will have a problem because we are competing with with with GPT, right?

It can write similar code. But if we redefine our job as creating value for our customers and coming up with new ideas to automate, So repositioning ourselves, not so much as being the one that is doing the actual coding, but the one that is translating these PPC objectives into automation opportunities, then we can still create a lot of value there.

And that is the creative part that I think you’re referring to. It all starts with with business objectives business goals Audiences that we would like to reach and then coming up with different strategies to actually Yeah, reach these audience and generate clicks that actually convert into value that is profitable for the business

FREDERICK VALLAEYS: Yeah, and I think let’s talk about agencies versus in house teams for a minute.

So my take is that A lot of agencies, unfortunately, tend to do a little bit of a, a cookie cutter job, and they make money through scale, right? I mean, so they, they might be able to help a, a, a company that doesn’t have an in-house team that doesn’t have the budget for an in-house team, but so they get the benefit of the scale of working with an agency.

But the downside is that sometimes agencies tend to do the same thing for each account. And so that’s a little bit what you’re going into now that like the real value that you bring is like the ideation The creativity like the new strategy, like how do you connect the strategy to your business data?

You’re the world that you operate in the industry that you operate in and I see in house teams being much more active in that field They’re always thinking about like, oh we have this data set which could influence ppc in that way So And then they’re thinking like, how do we connect that? Right. How do we automate it?

And that’s where scripts are so brilliant because rather than having to do this manually, you can turn it into a process and make it repeatable. And I think agencies actually have an opportunity now because of the, this army of junior developers, like you said, to actually go and do something a little more custom for each of their clients, which increases the value they bring.

And we would, which also, I think. Insurers job security at least for a couple more years until like that next wave of evolution happens,

NILS ROOIJMANS: right? Yeah, and to touch on your point with with with in house versus agency agencies I think also a lot of the in house advertisers they have sort of like domain Expertise that they can combine with the artificial intelligence.

A couple of years ago. I I did a talk in Austria the great day event and the title of the talk was keeping an eye on ai You Keeping an eye on Google’s AI because I think it’s especially with the way That Google is pushing us to make use of a lot of these AI technologies It’s very important to keep an eye on that and using your domain knowledge or domain expertise to guide the AI Can be very very very important interesting to, to, as you would say, unlevel the playing field and create a competitive advantage.

So I think there’s also a lot of opportunity there for the agencies to, instead of, you know, playing the the guacamole in the Google ads interface to really dive into the business of their clients and try to learn how to, to, to become more competitive.

FREDERICK VALLAEYS: Exactly. And so a little bit the PPC pilot where you’re monitoring the automation, the machine, and if it goes off course for some reason, knowing how to bring it back to where it needs to be.

I think that’s going to be especially important now going into Q4. So obviously a huge time, the cyber weeks happening for retailers is going to be huge. That’s not the time to change anything, right? That’s not the time to implement PMAX for the first time, but it is probably a time to have some anomaly detection, to have some scripts that are constantly monitoring things going right.

Have a good tool and there’s a number of them out there, right? So we use this technology, use this automation to stay on top of things.

NILS ROOIJMANS: Which if I may plug an example last week, I published a script that is using one of the latest releases of the Google API reporting API. We can now finally access the search category data in the P max through scripts.

So I created a script that monitors the trends in the different search categories that your P max campaign are being shown for. And these trends, they change over time, of course, and insights in The type of queries that your pmax campaigns are being matched to can be of high value, especially in a season like like the black fridays Christmas when it’s coming.

FREDERICK VALLAEYS: Yeah, and this is brilliant because I think if you look at the google ads interface to get to the search term categories You have to basically open every PMAX campaign and go to the insights page and go to the right section there. And then you start to start to see these things right now in an ideal world.

I think you shouldn’t just have one PMAX campaign. My thinking on this and we can talk about this, but you probably have multiple products with different margins. And so hence, if you want to maximize profits, you should have different TROs for each of these margins. So even with that. At the very least, you probably should have two PMAX campaigns.

Now if you scale that up, all of a sudden you have so many places you need to go to, and that’s where a script is really handy, because it just puts that in a spreadsheet, and then it sounds like your script actually monitors changes week over week, or what’s the time range that you look at for the changes?

NILS ROOIJMANS: This, this one is week over week. Yeah.

FREDERICK VALLAEYS: Nice. We’re putting it up on the, on the show notes so people can go and find that.

NILS ROOIJMANS: Yeah. And so, so if you start thinking, and that’s always what I’d like to teach my students as well, If you start thinking about these kinds of monitoring opportunities that script can create for you, you will, you will come up with completely new ideas, like, for instance, with the max, it’s a black box.

And we don’t know where our ads are being shown and what products it is actually showing and how that is changing over time. But if you have a script that alerts you if a certain product has an increase, a dramatic increase or decrease in number of impressions or clicks, That is something that is very hard to detect if you manually go through the interface, but the script can simply alert you if there is a change, let’s say 25 percent difference in last week impressions versus the week before.

So then you will be in the know of the type of products that PMAX is promoting for you.

FREDERICK VALLAEYS: Right, and that reminds me of a really cool script that you’ve written, I’ve written, Google’s written, everybody’s written it, but basically like rising and trending search terms. And it does the same for search campaigns, right?

So if all of a sudden there’s a new keyword or a new competitor that comes in and your broad match starts showing up on these new queries that you haven’t seen before. Good thing for you to know. There’s not necessarily a bad thing, right? I mean, I think a lot of people often assume like, Ooh, Google is broad matching to like really horrible things.

Broad match is bad, but no broad match just like identifies new trends. And if you knew about that trend, you could actually build it into a landing page. You could have messaging relevant to that. So that’s a great way to get more results from something like that. And then on the flip side, one of your popular queries that you were sort of banking on, all of a sudden the volume might drop because there’s a competitor who comes in and I mean, like Teemu, I don’t know if Teemu is in Europe yet, but Teemu is huge in the United States.

The amount of money they’re spending on e commerce ads, like they’re displacing Walmart, they’re displacing Amazon. And I still talk to a lot of retailers and they’re like, wait, what now? Who’s Teemu? But if you look at the numbers, like. You should know who Teemu is. They are impacting your auction dynamics.

NILS ROOIJMANS: Yeah, yeah, yeah. Great, another great example. Yeah. Yeah. Which actually brings me to, to, to maybe, maybe if the audience will appreciate me sharing some of the scripts that I use to actually take benefit of these AI algorithms with with everyone to

FREDERICK VALLAEYS: show on screen. The I think we have a chart here, right?

Do you want to talk about this one?

NILS ROOIJMANS: Yeah, let me for the people wondering what the, what, what the light is, this is a reflection of the, the sun in the canals. So I’m just going to close, close something here so that the water is better. Let’s see if I can. Okay. Quickly share this. Can you see the, yeah, there you go.

Okay. So this is a a campaign setup that I currently use to actually reap the benefits of one of some of the AI algorithms in the Google ads platform. And it also shows how you can use scripts to reap these benefits. So the green parts are what I’ve automated. Through scripts and the blue parts are two different campaign types.

So on the left, we have the alpha campaign. It is basically a standard search campaign with exact match keywords. And on the right, we have the boost campaign or the beta campaign. And that contains the ad groups with broad match keywords. So what we can do now is if we shift, like, like if we balance the budgets between the alpha and the boost campaign, we can sort of like dictate and help Google explore new opportunities through the broad match algorithm and a DSA ad group.

And then we can exploit all the learnings in the alpha campaign where we run the exact match keywords with some other smart bidding algorithms. And this boost campaign, of course, is generating search terms and the search terms have conversion data. For all the search terms that have had two conversions, my script will automatically add it to the alpha campaign as an exact match keyword.

And it will also add it as a new broad match keyword in the boost campaign. That latter, the latter is also very important because Unlike the other match types, broad match keyword relevance, matching also looks at the other keywords in the ad group. So if you add more converting broad match keywords to the same ad group, it will actually improve the quality of the broad match algorithm.

Another thing that the script, another script does is have a look at all the search terms and if it has had over, let’s say a hundred clicks and zero conversion, it will add it to the negative keyword list that is attached to the boost campaign so that you don’t spend money on a lot of the, Irrelevant matches that broad match is also notorious for yeah You can even go a bit a step further and include levenstein distances to circumvent the problem of a negative keyword match types but then you could also use different smart bidding algorithms inside these campaign types So for instance, I like to use the maximized clicks with a bit limit in the boost campaign So that for instance if you have a limited budget for exploring new opportunities, you’re going to maximize clicks And you’re going to update the bit limit also again with scripts To max, either maximize the clicks or make sure that the complete budget is spent.

So if you’re not spending your full budget, the script will increase the bit limit, but if you’re already spending the full budget, it will decrease the bit limits so that you get more clicks. And then of course, another script will have a look at the keywords. And if they have at a hundred clicks, there were conversions, it will pause them, et cetera.

This, this is typically something that is very hard, if not impossible to do without your own automation, but thanks to scripts, you can actually come up with these creative new strategies to create more value for

FREDERICK VALLAEYS: for your. This is brilliant. I think it’s time for a shameless plugs for a minute.

So if people want to learn how to do scripts like this or they want to get some of your scripts, tell them where to find it and how to get ahold of you.

NILS ROOIJMANS: Right. So if you, if you Google my name, you will, you will find my website where I share the knowledge on these kind of automations you will find some free scripts, you will also find a list to roughly 400 scripts that are out there you can simply copy paste for free, a lot of your scripts, by the way, Fred and every now and then I run an online workshop.

So people can join the workshop is twice or three times a year. I’m currently working on a workshop on how you can use chat GPT to create these kinds of scripts for you and how you can actually use GPT to learn the basic of the JavaScript in a way that is a lot of fun. Because GPT is also a very good teacher.

FREDERICK VALLAEYS: Yeah, that’s brilliant. So that’s your shameless plug. I’m going to do my shameless plug. So obviously Optmyzr started as a scripts company, but we’re way beyond that. So we have a rule engine capability. So that flowchart and conditional adding of keywords, adding negatives, changing budgets, all of that can be very easily Set up using those little boxes.

So there’s no coding required. And you can build your own methodology and you can put it on automation, right? So it’s, it’s a different way to it’s like scriptless or codeless scripting. You can do through Optmyzr and we have a two week free trial for anyone who wants to try that. All right, let’s go back to GPT and let’s I’m going to share my screen and we’ve talked about some of these things.

But I want, I want to show them to people. Okay. So the first thing I’m going to go into settings here it’s shows you right there, custom instructions. Okay. Let me go on the right place here. So you hit the three dots next to your name. Okay. Where is settings? Here we go. Settings, custom instructions.

So custom instructions is a way for you to specify. Kind of like the things that need to go into the prompt every single time and remember if you go to chat gpt You basically you frequently tell it speak in the voice of like someone who’s authoritative don’t use these words, et cetera, et cetera, these custom instructions, which looks like they’re currently not loading on GPT.

But this is a way for you to basically set those conditions and every single time they’re going to go into the chat. Now, sadly. I had my custom instruction set up and it looks like GPT has just lost them. But I, I, what I said was like, listen, I’m a, I’m a PPC expert. I’ve been doing it for this long.

Like these are some of the things I’ve written, but this is the tone of voice that I like to use and it’s kind of flattering because now whenever I ask it any question related to PPC, it’s like, Oh, Fred, while you’re, you’re a great expert on PPC, so you must already know this, but here’s my answer. I don’t like, okay, GPT, like stop flattering me.

But it is kind of funny how you can change the tone and you don’t have to repeat these same things so that’s custom instructions. Okay. So the first thing I want to show you here is custom instructions So if you click on the three dot hamburger menu next to your name, you have a box for custom instructions and these custom instructions basically allow you to not have to repeat the same sort of thing in the prompt all the time.

So it says, what would you like chat GPT to know about you to provide better responses? And so I gave it a little mini biography of who I am, what I’ve written about, what I’ve done. Also, where I live, Silicon Valley, right? So I can bring in examples related to things that I might have experienced living in the Silicon Valley.

And then there’s also a box for how would you like ChatGPT to respond? I’m not currently using that one, but this is more about how do you want to be addressed? Should it have opinions on certain things? et cetera. Now the one downside, I think as an agency, if you use this because you have different clients and they might all want a different tone or sort of different custom instructions, this may not be that useful to you because these custom instructions, they are for your whole chat GPT account, and they will apply to any future chats that you instantiate with the system.

But that’s right there. So, and that’s available both on the paid and the free plans. So try that out. The next one I want to show you, Is go to settings and beta to get beta access, you do have to be in the premium plan. So you have to pay your 20 a month, but then you get some beta features. So one of them, as Nils was mentioning was plugins.

And the other one is advanced data analysis also known sometimes as code interpreter. Now, let me so you turn these on and then you go back, you have to be in GPT four and then plugins you click the plugins right there. And then you enable which plugins you want. So I’ve installed a few of them.

They do various things. If you go all the way to the bottom of that list, there’s a whole plugin store. And so you can start searching on here for the various things that it can do. And like Nils again was saying, one of the plugins would be like a web scraper. So you can say, go and look at this web page and summarize it or find me some keywords for it.

Because as you all know, the large language model is based on data prior to 2021. So if you ask it, like, what is this blog about that I wrote last week? It’s going to make a pretty good guess just from the title of the URL and the title that it sort of assumes that it’s for the blog post, but it’s not actually going to be able to read what your blog post is.

Any plugins that you’ve particularly liked Nils?

NILS ROOIJMANS: For this concrete example, I like the both the link reader and the web pilot. They’re very good at accessing third party data. If you stall them

FREDERICK VALLAEYS: link reader, or maybe

NILS ROOIJMANS: they

maybe they renamed it again. They search for web, web, web, web pilot, web pilot. There you go.

FREDERICK VALLAEYS: To browse web page and PDF data. So yeah, and some of these new plugins, they get launched. So Microsoft had one, so you could do a search on Bing, but then that got pulled back because it found that it was like a great scraping tool.

And that was the intent to make sure that a copyright was protected. They’ve pulled that one back while they sort of rework some of the privacy settings. So anyway, these are plugins. And then once you have a plugin, so okay, let’s make sure we turn on the web pilot and Neil’s give me an example. Of what you might want to do with that pilot.

So you can run up at the same time. So I’m going to turn this on. Right. So what might I actually,

NILS ROOIJMANS: yeah, so actually one of the scripts I’m currently creating is a script that pulls the search terms that your ads are being shown for, and then it’ll ask GPT, how would you improve the copy on the landing page for these search terms?

So we have a list of search terms and you have a landing page URL, and then you ask GPT. How would you improve the copy on the landing page?

FREDERICK VALLAEYS: So I’m asking, how would you improve the landing page copy of Optmyzr. com? I mean, what you’re doing obviously relies on writing a little script to pull the search terms from Google ads first, and then it feeds them.

Into you know, this plugin, by the way, do the plugins work in the API? No. G-P-T-A-P-I to go, right? No.

NILS ROOIJMANS: So, so this, this concrete example requires me to first fetch the HDML of the landing pages and then feed the HDML as part of the prompt. Yeah, but that is yeah, maybe a bit too technical. to go into the details.

But in theory, what should happen now is that GPT calls on WebPilot to actually scrape the website Optmyzr. com. And then

FREDERICK VALLAEYS: So it seems to be currently a little bit stuck. So we’re going to ask it what is on the URL with Optmyzr. com.

So yeah, unfortunately Chad, GPT does hit its limits. Okay. Now it is actually using web pilots, so you can expand that box to see what it’s doing. It’s still generating. Okay, boom. So there you have the content of that page and it uses its large language model to summarize what is on that page. It includes a nice little image that was on the page.

That’s pretty cool. So that. Basically connects GPT’s capabilities to what is on the web today, as opposed to what’s in a large language model from a couple of years ago. Awesome. Let’s do another thing. Let’s use the different plugin.

I’m going to start a new chat for this, and we have to use GPT 4 again, but this time I’m going to use the advanced data analysis. So one example here about one important thing is you can upload files So generally what you would do is you would download a data file Say all of your keywords or all your search terms from Google Ads and then you can upload that as a structured data file I don’t think I have any connected here on this computer, but then with that you can write Python code.

And so we can talk to it in normal language, but can you write a script that draws a chart of, CPC costs it’ll probably tell me something along the lines of, yeah, but I do need some data for that. And it actually gives me a step by step breakdown of how that would work. Right. And what’s interesting is in that work itself, you can open that up and that shows the Python code.

So you can copy and paste that Python. You can actually run it on your own system if you wanted to. But the other nice thing is you don’t have to run it on your own system. It’ll just run it inside of GPT.

NILS ROOIJMANS: Right. I mean, maybe this is a good time for me to share an example graph that I’ve created with Code Interpreter in as little as you already see it’s made up some, some, some numbers and it shows a graph.

But yeah, let me Let me share my screen. Okay. So what you see here is a graph that I created with chat GPT’s code interpreter in as little as five minutes. And what this graph shows us is an Ngram analysis of search term data that I took from the Google ads account. So I simply downloaded the search term report, uploaded it into code interpreter, and then I asked it to do an Ngram analysis and focus on bigrams.

And I asked it to have a look at the top 20 of these bigrams and create a graph that shows the number of impressions, the click through rate, and the average CPC for these bigrams. So on the x axis, you will see the bigrams. On the y axis, you will see the impressions, the line is the click through rate, and the color of the bars indicate the average CPC for that diagram.

So what you can immediately can see is the red one on the left, this one, has a relatively high CPC and a low click through rate. Whereas this first dark green one, Has a relatively low CPC and a very high click through rate. So this is the type of insight that you can create in like five minutes, where in the past, literally this would take me hours, if not two days.

To create a graph like this, but

FREDERICK VALLAEYS: with code and with math, so you can trust that these results are correct. And you, you do have to validate the code a little bit, right? You have to make sure that when it looks at CTR, it knows it was clicks divided by impressions if it didn’t already have that data. But it’s not making stuff up, which a large language model will tend to do.

Like did what you saw on my screen. It just made up some random campaigns when I didn’t give it data. It’s like, yeah, I know what a campaign is. I know what a CPC might be like. So here’s a potential chart. Yeah. So excellent point.

NILS ROOIJMANS: But, but, but here’s the good thing about the code interpreters, since it is generating Python code, you can actually validate that.

What, what it’s doing and that is correct. And the nice thing about code is it’s, it’s logic, right? And logic is something that is either completely true or completely false. And whereas with, with the, with the free language, like if you, if you ask it to generate some, some prose or poetry, then it can generate a lot of confabulations.

FREDERICK VALLAEYS: Exactly. Some things you want it to be creative, right? So writing poems, be creative, writing code, no, don’t be creative. And then here’s what you’re talking about. So you can actually open up in a GPT. What is the code that you were using to produce this work? And so you can see here while the CPC costs, because I didn’t provide any, it just made an array of its own CPC costs.

Obviously this is where I’d want to plug in my actual campaign names with those correct costs. Now let’s actually go and do this Neil. So. Okay. You said you have to get your search query data. Wouldn’t it be nice if we had a script that got us that data and puts it in a Google sheet and sends us an email.

So what would we ask GPT to do?

NILS ROOIJMANS: Right. So I would start with the system prompt. So act like you’re a professional Google ads script developer.

Okay. So. Now we’re going to ask it to create a query that returns search term data from our Google account.

FREDERICK VALLAEYS: It seems to be extremely slow right now. So that’s a bit of a downer, but okay. So you create a query that returns search terms data from an account. I’m still thinking about the previous response. Where we just asked it to be a professional developer Hey, maybe you think professional developers are always slow in responding and that’s why it’s That

NILS ROOIJMANS: is domain expertise

FREDERICK VALLAEYS: Okay, so it looks like gpt it was actually at capacity right before we started recording here So I think it’s just too many people using it.

That is a bit of a potential concern. But yeah, I mean so You Listen, we’ve reached the end of time for this episode, at least not the end of time in general, but Neil, so again, remind people if they want to learn how to do scripting, where can they get a hold of you because there’s some really fascinating stuff, and hopefully you’ll join us for another episode as well.

But how do people get a hold of you? What’s the website?

NILS ROOIJMANS: Yeah, so it’s basically my name, Nils Roimans, but I’ll also link it up in the, in the description. In the comments or the footnotes. And yeah, we definitely go look for the, the, the list of free Google ad scripts that will give you some inspiration and get your mind thinking about what is possible with all these kinds of automations.

And also if you’re interested in learning more about how you can use GPT to write your own scripts and teach you how to, to learn the basics of JavaScript, be sure to find me on LinkedIn and I’ll I get in touch on on Twitter. The next workshop opportunity.

FREDERICK VALLAEYS: Awesome. Well, Nils, thank you so much.

You’ve been a fantastic guest and shared a lot of your expertise and hopefully people can now go and use GPT to write their own scripts a little bit. If you’ve enjoyed this episode and you want to see more, please use the buttons at the bottom of your screen to subscribe. We have PPC Town Hall roughly twice a month.

And of course, if you’re looking for some prebuilt scripts or some other automation tools, Check out Optmyzr. We have a two week free trial that you can use at optmyzr.com. We also have great e commerce capabilities. We just introduced a new PPC vertical insights report. So lots of cool new things being added all the time.

And when we see a cool script we tend to also turn it into an API tool so that you don’t have to do scripting. If you don’t want to, but I highly recommend scripts because they’re a great way, like Neil said, to to get an edge and unlevel the playing field and get better results than your competitors.

So thank you for watching. Thanks Nils for being here. And we’ll see you for the next