ChatGPT for PPC: Use cases and impact of such powerful AI tools on our jobs

Episode Description

Just to be clear, ChatGPT did not write this description :)

We’re barely two months in since OpenAI opened up #ChatGPT for public use, but what a phenomenon it has already been!

While it has its flaws, there’s no doubt that you, as a marketer, can use it for some powerful use cases: creating ads, writing blog posts, scripts, and structuring campaigns, just to name a few.

But what does this tool mean for our jobs and the future of marketing?

In this episode, you’ll hear from Dave Davies and Amy Hepdon about all of that.

Tune in to learn:

- A few powerful applications of ChatGPT

- Is it really a “Google killer”?

- What are some other powerful AI tools for PPC?

- How do such tools impact our jobs and the future of marketing

Episode Takeaways

Powerful Applications of ChatGPT

• Enhanced Content Creation: ChatGPT facilitates rapid generation of content variations and can defend its creative choices, improving A/B testing for ads.
• Conversational Enhancements: Offers a conversational interface that allows users to explore different content angles and refine marketing messages dynamically.

Is ChatGPT a “Google Killer”?

• Uncertain Impact: While ChatGPT is revolutionary, its current impact as a “Google killer” is debatable; it might enhance tools or systems rather than completely displacing existing search engines.
• Future Potential: Continuous development could lead ChatGPT to significantly influence search technologies and the broader tech landscape, possibly reshaping how information is queried and processed.

Powerful AI Tools for PPC

• Jasper and Others: Tools like Jasper offer content generation with specific user-directed tonality and style, aiding in diverse content creation for marketing.
• AI Integration in Spreadsheets: GPT for Sheets can automate content generation directly in spreadsheets, enhancing productivity for marketers handling large data sets.

Impact on Jobs and the Future of Marketing

• Job Evolution: Automation and AI will shift the roles of marketers towards more strategic, creative, and supervisory duties rather than displacing them.
• Continuous Learning: Marketers will need to adapt to increasingly sophisticated tools that handle routine tasks, pushing the industry towards more strategic and less operational roles.

Episode Transcript

AMY HEBDON: I think one of the challenges of RSAs is they are so hard to meaningfully test because of how mix and match they try to be, it becomes really hard to like test the differences between them. But I had it create two RSAs for me that I said, you’ll create two different, two different RSAs and then tell me why this would make a good A B test.

And it was able to do that. It was able to defend its choice of having one focus more on exclusivity and one focus more on options, you know? And so the fact that it can do things like that I, I think that that’s just a good example of what we can use it for to, to kind of like help get something on the glass, help us get some ideas, you know, Look at things a different way, not, not to create something because we don’t know how, or because we literally don’t have time, but because it’s always helpful to get a new perspective on something that we can then move forward with and just be better as a result of.

FREDERICK VALLAEYS: 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 at Optmyzr. So today we have an exciting topic. We’ve been talking a lot about automation over these past couple of years. But the exact chatter about automation has changed quite a bit here in the last month or so, because a company called open AI has introduced a new chat bot that seems to be doing really amazing things with AI.

It’s writing movie scripts. It’s writing whole podcast episodes. It’s writing poems. And of course the marketing community has jumped on the bandwagon and has tried to figure out how can we use this amazing AI to do things for us. And make our lives a little bit easier. So what is chat GPT? How do people use it?

What is it good at? What is it better? That’s what we’re going to be talking about today. So welcome to this episode of PPC town hall.

All right. Then my two guests today are amazing experts. So we have Dave Davies and Amy Hebden. Welcome both. Dave, let’s let’s start with you. Tell us a little bit about what you do. And I mean, you’re really deep in the machine learning and AI space. You’re also an SEO, which is like, this is a great mix of things, but tell people who you are, what you do.

DAVE DAVIES: Perfect. My name’s Dave Davies. I am the amplification team lead. So dealing with amplification of content and stuff like that on the SEO tangent, but for the MLOps or machine learning operations company. Weights and biases. So yeah, you’re right. The people I deal with every day are doing this constantly and for a living.

And in fact, open AI trains some of their stuff with our with our tools. So, yeah, I, I love what we’re talking about. Probably one of my favorite topics. And I’m glad to see it finally getting sort of the light of day and, and, you know, The appreciation it deserves, I think

FREDERICK VALLAEYS: the appreciation and also anything that went wrong with it.

Basically, you guys are responsible too, right? That’s it. That’s it.

DAVE DAVIES: It’s actually my fault. Personally, if anything is wrong,

FREDERICK VALLAEYS: and then we’ve got Amy have been on the call too. So Amy, first timer on PPC town hall, just like Dave Davies. Thanks for coming on. Tell us a little bit about paid search magic and what you do.

AMY HEBDON: Sure. So I’ve been doing paid search since for about 20 years now. And paid search magic is a business that I founded and I’m the managing director of it’s an intentionally small Google premier partner agency. It’s myself and my business partner and husband James, and we really focus on paid search.

That’s Google ads, Microsoft ads, and then everything it takes to get that. To be successful, whether that’s improving landing pages and offers or creating custom reporting or growth strategy, just, you know, whatever our clients need to have that success. And we are currently in Tennessee. We moved here a few years ago with we’ve got a dog and eight chickens.

FREDERICK VALLAEYS: Nice, I’ve got three chickens. It comes in handy

AMY HEBDON: with the price of eggs now, right? Like, I know everyone’s talking about that.

FREDERICK VALLAEYS: I know, but mine stopped laying this last year, so, I guess chickens too. But so, hey. So you’re a small team. So with this AI stuff, it’s probably super useful to you having kept the team relatively small.

So I’m really curious to hear what has worked and whether this is actually helping you with that small team do more of the stuff that your customers want.

AMY HEBDON: Absolutely. But,

FREDERICK VALLAEYS: But yeah, I mean, so there’s been a lot of chatter about chat GPT. So why don’t we take a look at that first? And for people who’ve just heard about it, but haven’t played with it, let’s have the three of us.

Play with it. So here’s chat GPT did the signups are relatively open right now. Sometimes when you go to the website, it says you have to wait. I signed up with a second email address just now. It took me half an hour of waiting and I had access to the system. So Dave or Amy, what should we. Ask chat GPT to do,

DAVE DAVIES: I guess I’ll, I’ll jump in with, with some examples and, and we’ll keep them in the PPC realm. I mean, there’s, there’s all sorts of things you can basically dream up. You have it right. A haiku, if you, if you feel so inclined. But for example, you could type something like write me five. 30 under like with the less, less than 30 character headlines for a travel agency in Vancouver, right?

Like you could, you could type something like that and just see what it produces.

FREDERICK VALLAEYS: All right, so let’s do that. But since you mentioned haikus, I asked it to write a haiku about ads. And here we go. Ads on billboards, incessant noise in my ears, peaceful nature calls. Beautiful. But you, so you said right five in the red box here,

headlines under 30 characters.

We’re a travel agency, see what happens.

AMY HEBDON: This should be interesting because it actually can’t count. So we’ll see if it is able to hit those.

FREDERICK VALLAEYS: There you go, it actually, oh did you mean it can’t count five headlines or the 30

AMY HEBDON: characters? It can’t count characters. It doesn’t have that calculation function.

It processes language and so it doesn’t count characters.

FREDERICK VALLAEYS: Right. So basically you’d have to generate headlines, export them into a spreadsheet, and run a length calculation. Yeah, or know what 30

AMY HEBDON: characters look like and be able to tell if something’s off. Yeah.

FREDERICK VALLAEYS: These look pretty good, right? I mean, I think all of these would fit under the 30 characters.

So, yeah, let’s take a look at what it wrote. So, explore the world with us. Dream vacations made real. Where will you go next? Journey to your paradise. Adventure awaits. So what do you think about these? I

DAVE DAVIES: personally like, I like those well enough. What I would, and it would be difficult to do on the screen, but what I found also helpful is I’ll go to a search result page and pull off all of the headline snippets, not the full headlines, but you know how, like it’ll combine two or three and you can actually say, write me.

Five headlines or 10 headlines under 30 characters, whatever for travel agency inspired by, and I’ll actually just put a list of all of the headlines that, that appear up there, stripping out any brand terms or something like that. And it will actually use those headlines to inspire the type of language that That it’s going to, to construct in, in your sort of sample headlines under the premise of if it works for the top, then it hopefully will work for me, not always true.

So you have to put your thinking cap on, but it’s just one other sort of like twist, or if you have incredibly good headlines, you could dump those in and go, write me 3 new ones inspired by these are my best and right. You want to like pillage out your worst and find some, some ideas. Something I find interesting though, a lot of what we’re talking about here and a lot of what.

Sort of Amy brought up as well, or alluded to is, but. We got to use our brains because it’s a machine and it’s really wrong a lot of the time.

FREDERICK VALLAEYS: Let’s why don’t we do an example of something where we think it might actually be wrong. So I think it’s usually it’s it’s learning model is not that recent.

So I think it’s trained on data that’s about two years old. So it would not know what a performance max campaign is. So why don’t we ask it to create a paragraph explaining what is performance max?

Oh, wow. Look at that.

AMY HEBDON: Looks like it’s heard of it.

FREDERICK VALLAEYS: It’s heard of it now. So that, that is one amazing thing. So we did do some examples a couple of weeks ago, and it actually did not know at the time of performance max was and then this is maybe a key point. It’s machine learning, right? So it’s constantly ingesting new data.

It can be trained to, to learn about new things. So the limitation that existed just a few weeks ago, was just a technical limitation on the training data, but easily fixed by giving it newer information.

AMY HEBDON: So how would that work? Cause it still says that it’s trained up through 2021, that it wouldn’t include something that’s more recent than that.

So how is it getting that additional training data that exists later than that? It should

FREDERICK VALLAEYS: that sounds like a question for you.

DAVE DAVIES: Yeah, that’s funny. I’m like, okay, I’m just looking at the most recent example, because there’s, there’s 2 things at play here that go on with a model like this, right? Like, there’s the data that it has access to and then there’s the weights.

The model is using and these are 2 different things, right? Like, we will. Train well, we like, okay, people far smarter than me will train a model and it gets its weights. So it sort of understands, okay, this is how we are going to understand language. And this is how we’re going to decide what the next word in the sequence is.

We’re going to like, basically like, like Google does with their organic algorithm, right? We’re going to. Move all these levers. And now we’re gonna understand how to define what the next thing is and what our success metrics are. So there’s that version. That is a different training time than when do we inject new content in so you can refresh one.

And I’ll look that up while we’re like chatting about different things. When the last time they did that refresh was a good example is you dot com. Another search engine launched. You chat It was announced, I think, December 23rd. And it actually operates with some real time data. So like true inference, it’ll reference like a news story from like minutes ago.

Well, okay. Maybe not like last I saw was like hours before was, was the closest, but it’s operating in real time as it adds to their index, they have access to it. So it is possible to sort of like blend these two things together, but. That doesn’t mean it has a true understanding because the model weights were decided two years ago, so it may not understand what the entity is really.

It just understands sort of how things are piecing together in language, not necessarily how what chat GPT might be or performance max might be as an entity unto itself. It just understands that these words connect with these words. Because it’s seen that as an example before, but it wasn’t actually trained to understand it.

If that makes sense. I might. That’s that’s

AMY HEBDON: really interesting. Because I saw, I think last week someone had asked like, who owns Twitter? And the answer was Elon Musk. And I was like, how do you know that? If it’s not. You know, if that’s more recent, and then it’s like backtracking, right? Wasn’t able to explain what you just explained.

And so it was like, is this a conspiracy? And so like, just understanding that I think can kind of help to conceptualize what’s happening and where you might see more recency of data without necessarily having that be part of the training. You know, that precedes it.

DAVE DAVIES: Yeah. And I think this is the really, really hard part that Google will face sort of ties into exactly because we we’ve heard and sorry if I’m derailing here a little bit, but like chat GPT is like a Google killer, right?

Or we’ve all heard that in some format. A highly doubt that because as we’ve talked about here, it’s not right. A lot. So it’s, it’s, it’s built to do, you know, sort of a different thing. We can talk about it in relation to Microsoft. But what Google needs to accomplish is referred to as inference. And it’s basically meshing these 2 things with chat GPT and go.

You need to operate in real time, but with the accuracy of Google. And the mass of data, but models that might be trained in the past, but needing to understand what happened yesterday in the Ukraine or something like that to actually put the right context on on what it’s generating is an answer. So, I think that’s where we hit the real problems is balancing those 2 and then, of course, the compute power required, which is going to be a challenge, right?

For any company is going and you now need to run these calculations. Constantly on the fly, 10 billion people. I mean, I’m sure, you know, I’m overcomplicating that, but it’s a small challenge on their part.

FREDERICK VALLAEYS: Well, and that, that was Google in the early days too. Their data was very old because it took them weeks, if not months to crawl the whole web and refresh.

And that, that was on an actual cycle, right? But then obviously the technology gets better, machines get faster. So there is a point at which doing this for 10 billion people, Probably is going to be fine. It’s, it’s going to work just fine. But, but now to your point that the machine is often wrong, talk about that a little bit more because what you say it’s doing, it’s basically putting words together in logical or in sequences that it’s often seen before.

But unlike on Google search, where you get 10 results and nowadays, I mean, infinite scroll, if you see something that stands out as like, Factually wrong or different from all the other results as humans, you sort of picked that out, right? But here we get one answer. So how is it even saying like, this is the one answer I’m going to give you.

And by the way, let me show you something on screen. Let’s show the screen again but I’m going to ask the exact same question again.

AMY HEBDON: So the question is, yeah, what is performance max in Google ads? And it looks like it’s so far, it’s been very, very similar, but now it’s a little bit different of an answer. Exactly.

FREDERICK VALLAEYS: It started very similar and then it added a little bit to the end. So that was curious, right? So one of the things you could have done here is say write it in the style of Shakespeare, and so it probably would have taken those same words, but added a little bit of old English flair to it.

You could even turn it

DAVE DAVIES: into a haiku if you wanted to, or a poem. You know what’s

AMY HEBDON: interesting? I had I was playing with this recently, and I had it write a poem or a song from a vantage point of a dog, and we got into a philosophical argument about whether it’s okay to use animals for the sake of entertainment.

And I was like, it’s okay to be entertained by animals. It’s like, animals are sentient beings, and you shouldn’t just be entertained. Like, it kept every single time. I asked it anything about an animal or to do anything with an animal, it would come back to me to tell me that animals are important, that it had like almost a disclaimer around it, which was kind of weird, but it was very, very consistent too.

FREDERICK VALLAEYS: Yeah, and I’ve seen similar safeguards. So my my kids are four, seven and nine years old, and there’s a series of books where animals battle it out. And so it’s a shark versus shark or polar bear versus walrus. And so we were like, well, let’s ask Chad GPT, what would happen if a lion and a tiger would fight?

And the answer was something to the effect of, well, actually, they almost never meet in the wild. So it’s kind of a dumb question. But by the way, it’s also not ethical to make animals fight.

AMY HEBDON: Right.

FREDERICK VALLAEYS: And so that was like, yeah, I kind of respect that answer. It’s a little bit ethical, some philosophy in there.

But, but that’s basically the safeguards that are put in by the humans, right? This is not the machine making a judgment about, hey, animals shouldn’t be made to fight for human entertainment. These are rules that the developers have put in to make sure that we kind of stay within the bounds of morality.

DAVE DAVIES: Yeah. Well, and then that leads to it, to an interest, like a sort of different type of question is. Who decides that right? Like, and that’s a big question that we all need to ask is who gets to decide what those those boundaries are. Do we democratize ML and go? Things should be open source. So people can can do what they want with them and advance them.

Is that the path ahead? Or do we want? Corporations like Microsoft to be a gatekeeper to that information. I don’t think there’s a right or wrong, and I won’t pretend to have like a good answer, but that is a common question that you look at, like, say, Dali, which is also owned by open AI versus stable diffusion.

Very similar models, but that one is open source so that anybody can can build on it and turn it more into what they they want it to be. And you, you end up with the, which is a better path forward.

FREDERICK VALLAEYS: And that’s a really good point. So because chat GPT is an implementation of a chat tool using, is it using GPT 3 from OpenAR?

It’s using,

DAVE DAVIES: for right now, it’s on GPT 3. 5. Okay. And rumor, GPT 4 will be out within the next couple of months, hopefully.

FREDERICK VALLAEYS: Exactly. But so this is a tool built on one flavor or another of GPT, but we as a company, so Optmyzr uses GPT to do add text variation suggestions in our tool. And we get all the knobs and the dials about what is it that we permit and what is it that we don’t permit?

How close does it need to be? Like And so in that regard, it is a technology where we all get to make a choice, right? And then if we decide to use chat GPT, we have to live within the rules that that company has set, but other companies could set different rules.

DAVE DAVIES: Right. Or you could use, I mean, there’s GPTJ, which is also available and open source.

So you could now go, we’re just going to download that. And now we’re going to do what we want with it. Now you probably, because you’re running like a good and ethical company, probably don’t need the capability to go. I want this to generate anything that I could possibly train it on. But there would be applications where, where, you know, you might want to do that.

And then there’s this open source version, which is not as, as elegantly trained as, as this one, because there isn’t. Money and Microsoft behind it. But

FREDERICK VALLAEYS: let, let, let’s talk about that, right? So Microsoft and Google, where do they stand on this? So, Microsoft, Dave what’s what’s happening with them and Chad?

Well,

DAVE DAVIES: Microsoft in 2019 had put a billion dollars into open ai. And now they’re putting on the table another 10 billion to own 49%. With other stakeholders owning 49 and then 2 percent going to like the sort of charity that sort of like runs the whole, the whole kit and caboodle to give them some a great deal for Microsoft because if you read the arrangement, they get their 10 billion out 1st and then they just don’t 49 sort of sort of free values.

The company had about 29 billion. So that puts them in there. Now, 1 could think of this like, originally, when I read it, I’m like, oh, they’re, they’re really going to go aggressive at, you know, Google because you think of how you know, this is as an aggression sort of point, and you think of Google as a challenge.

But if you think about them actually integrating this in, and it could be a challenge on Google, but not from a search perspective, but going, we want to make office more valuable than Google suite of tools. We want to put this into spreadsheets because it generates code and go, Google. Okay, just make column C, add columns A and B, right?

Okay, now it’ll just do that or have it helping to write a document as you’re as you’re going along in Word. That could be a massive leap forward and and sort of maybe be one of those cases of Microsoft going. We don’t necessarily need to win. We just need Google to lose, right? To make this work. And this would be a good way to do that.

AMY HEBDON: I don’t know that Microsoft has the best track work record of acquisitions being turned into something especially useful. They put a lot of money in a lot of companies and that seems to be where it ends. You know, I’m still wishing LinkedIn did anything for, you know, like being, being the acquisition there did anything for us and it doesn’t.

And I just like a lot of, you know, Skype and all these different things. It’s like, Hey, great being in Microsoft invested in it, but how does it benefit the consumer? So I don’t have a lot of faith that anything is going to change as much as I would hope that it would, you know,

FREDERICK VALLAEYS: And sometimes it’s just about blocking the acquisition for, of going to someone else who might do a little bit more with it.

AMY HEBDON: Yeah.

FREDERICK VALLAEYS: Yeah, but also, I mean, so Microsoft did that investment, big stakeholder now. There’s some speculation, this is unconfirmed, but that chat GPT would be embedded. In being searched at some point in the future and you’ve already mentioned that GP to chat. GPT could be named a Google killer. I mean, is it going to be, we don’t know that to be determined.

But does this change the search landscape, Amy, if chat GPT, all of a sudden becomes part of Google or Microsoft? If it becomes part of

AMY HEBDON: Google? Oh,

FREDERICK VALLAEYS: well, I mean, or if it becomes part of a search engine, we already use, like, does it change our lives as digital marketers? Should we be concerned about this?

AMY HEBDON: Well, I mean, as a digital marketer, I’m pro the Google killer. You know, I I’m pro anything that disrupts it and makes it a better user experience. It makes it a better advertiser experience. Cause it’s neither of those right now. And I, I want it to be, but it’s not. So if we get some good disruption there and make it have to work a little bit harder on its initial initial mission, I see that as an upside.

FREDERICK VALLAEYS: Yeah, competition is good, right? So, and it’s interesting too, because the examples that were given about, hey, maybe Microsoft wants to use it to make office better. While most of us have been using Grammarly, we’re using Google Docs, and it’s already suggesting what to write. If we use Gmail, I type in hello and it’s like, Oh, you’re probably emailing Dave and Amy.

So it’s like, hello, Amy and Dave. And I don’t have to type that in. I just hit tab that’s AI in action. And nowadays, like we all use AI the same way that we all use electricity. It’s just a given it’s everywhere around us. And oftentimes it’s not really that visible. So this is just an extension of it. And, but what, what’s very fascinating here is like companies like yours, Dave, make this AI available to any company.

But even if we decide not to go that route and build stuff on our own, you can use a tool like Optmyzr. It’ll use AI to suggest headlines for you. You can use Google docs. It’ll suggest what to write. So, so I, maybe it’s not as big of a change. As people are making it out to be, it’s just a different manifestation that’s kind of like, very cool and different.

It’s the first time people have this hands on experience with AI. And maybe that’s why there’s so much talk about it right now.

AMY HEBDON: I think that’s really true. You know, you don’t get to be a digital marketer and not have AI in your rear view mirror all the time. You know, they’re, they’re successful. Paid search marketers who have never used a version of Google ads that didn’t have an AI component, right?

And there’s all sorts of third party platforms and bid management tools. They’re all using AI. There’s a lot of ways to leverage it for copywriting. So the fact that now as consumers, we get to participate in it rather than just saying yes or no. Like maybe we have the ability to say, don’t run that ad or do but we get to like.

Benefit from it more on a creative side I think I see the upside of this a lot more than I see any sort of downside because I don’t think that I don’t think that paid search marketers are really in the game of saying hey, you know My value add is that i’m old school and do everything manual and you know Everything’s made with love like we all embrace ai but now we get to participate in them more

FREDERICK VALLAEYS: Yeah, the example we love to give is if you hire an accountant Do you go with the one that has an abacus or the one who uses QuickBooks and spreadsheets?

AMY HEBDON: Well, exactly. If you’re going to a doctor, you don’t want the guy that’s afraid of technology, right? Like you want people who know how to use the tech to their and your advantage.

FREDERICK VALLAEYS: Exactly. So Amy, tell us a little bit more about how you’ve used it in the agency.

AMY HEBDON: Yeah, so I think that I have not used it to its full extent, right?

We’re on the very cusp of it just having been released very recently. So I’m still discovering new ways to use it. But the things that I think are interesting is how explicit and specific you can get with it. Because that’s really where the value is. It’s not, I mean, we use a good example to introduce how this works with five ads for, you know, five headlines for a travel agency, but that travel agency has a target demographic, right?

That travel agency has a specific differentiator that it should be communicating and you can put all that into. You can say, right, headlines, you know, and hopefully you’ll get the character count right for this product or service with featuring this benefit for this audience in this tone and it’ll produce it for you.

And you couldn’t even have it basically defend its work. Like I think one of the challenges of RSAs is they are so hard to meaningfully test because of how mix and match they try to be. It becomes really hard to like, Test the differences between them, but I had it create two RSAs for me that I said, you know, create two different two different RSAs.

And then tell me why this would make a good A B test. And it was able to do that. It was able to defend its choice of having one focus more on exclusivity and one focus more on options, you know? And so the fact that it can do things like that. I, I, I think that that’s just a good example of what we can use it for to, to kind of like help get something on the glass, help us get some ideas, you know, look at things a different way, not, not to create something because we don’t know how or because we literally don’t have time, but because it’s always helpful to get a new perspective on something that we can then move forward with and just be better as a result of.

FREDERICK VALLAEYS: Let’s take a look at the example you just gave there, right? So writing ads geared towards a certain type of persona. So the last two that I put in, I said, write a headline for a travel agency, targeted luxury buyers, and then the same thing, but targeting backpackers, and it came out with two pretty decent headlines.

Let’s read this

AMY HEBDON: for the, yeah, the people who can’t read it experience that’s like me, like experience luxury travel at its finest with us for luxury buyers and explore the world on a budget with our backpackers trips for for backpackers. So yeah, it does. I think, you know, that’s taking advantage of its, of its language processing, right?

Like getting really specific with how to use language.

FREDERICK VALLAEYS: Right. And now the big question becomes, is this a time saver or is this a time waster? Because now you have to go in and use the chat GPT chat functionality. It’s not like I can come in here with a big spreadsheet that says, here’s the vertical that I operate in.

Here’s like five audience segmentations. Mix and match them all together and spit out 5, 000 headlines.

AMY HEBDON: Would you want to though? I mean, I, I don’t tend to need a thousand headlines, I guess. So, yeah, but I guess if you’re, if you’re managing, you know, a client, then this is a really great way to do that. If you’re trying to manage a client at scale, you know, which is a different value proposition, then yeah, that.

You need to use it differently.

FREDERICK VALLAEYS: Right. And I think for me, it’s a little bit about stringing different technologies and processes together. Right. So, I mean, we all know we’re going to be doing RSA ads and that means we need headlines and maybe we already have 12, but we just need three more, right? So again, we, we just now need to go and get those 12 headlines.

We have formulate the right. Terminology. So chat GPT knows what to do with it and then capture the output and feed that back into Google app. And this might

AMY HEBDON: not be the right tool for that. There are other tools that leverage, you know, that are AI powered that are much more plug and play where you can just hit publish, where you can generate a lot at scale and then connect it to your Google ads account.

This isn’t that, but it’s, I, I, I prefer it. Cause like I was talking about before, like just the amount of control you can have and say, Oh, that’s not quite right. Try it a different way as opposed to accept or reject and then move on. You know?

FREDERICK VALLAEYS: Exactly. And that’s what we see a lot too, is that humans, advertisers, they still want a tremendous amount of control, but they want the benefit of the machine speeding up the work that they do.

Right. And so it’s that final approval stage. And in some cases, you may have a client who is in a sensitive vertical. Finance, for example, you know, great, you, you can write certain things, but if you’re selling mortgages, you have to be very specific about the rate that goes in and better get the headline one or headline two, etc.

So you can’t just let the machine run it because it’s going to get you in trouble. There’s some other tools that you can think of besides Chad, GPT that are having some momentum for advertisers.

DAVE DAVIES: Sure. There’s, there’s a couple, one that we, we sort of talked about. Well, it actually is based on, on GPT.

But for, for folks who want to, want to. There’s actually a fun and free extension called GPT for sheets which actually would, would sort of fill what we’re talking about here, where you can fill all of column a with like, write me a product description for, and then just fill entire column a with that and then fill entire column B with like shoes, bracelets, you know, combs, whatever.

And then just. let it create in column C all of the combinations. So it’ll actually just generate in a sheet all of your all of your different combinations. So that’s a fun one. But then there are other, other tools outside of GPT. Jasper is one of them that came out a little bit before chat GPT had come out.

And it does fundamentally the same sort of thing. It’s, it’s content generation but with a little more, It’s it’s paid, but it’s a little more user friendly, I think, in a lot of ways in that it lets you see your options, right? Like it has a little drop down for what tone do you want it to write this in?

Do you want it to be friendly or formal or and it gives you a little prompts for, like, in the tone of like, you know, Snoop Dogg or Shakespeare, right? Like, it sort of lets you lets, you know, sort of a little bit more about what you can do. And then, of course, we were talking about Dolly or stable. This isn’t just about you know, the generation of.

Textual content. There, there’s, you know, Harmon AI is doing some work into generating music and tunes. And they’re also under the stability AI umbrella, which are the folks that created stable diffusion. You know, and then there’s, yeah, Dolly and, and stable diffusion themselves generating images, which I think we’ve all played around with, or you, you’ve certainly seen on Facebook, people who’ve like gone, I’m now a superhero.

And there they are. They’re a superhero on the moon. You know, in a spaceship, but with a unicorn, right? Like, or whatever it is that they’re, they’re dreaming up to put in there. For most companies, does do the, the image generation work? You know, I, I work in an ML company. Yes, it does for us because a bunch of ML people expect to see a bunch of ML generated images, for example, or, or, you know, in their featured images does a standard travel company.

Like if I’m, you Was was working with a travel company. Would I go? Yeah. And somebody wants to see a stable diffusion generated image of the location that I’m talking about. They don’t. But there are some twists that you could do. For example, if you do use stable diffusion, you can go to the dream studio and you can go there.

You can actually train it. You can give it samples and go like, Okay. My house looks like this and just dump a bunch of these like vacation rentals in there. You know, if you were Airbnb or whatnot and go, here’s 20 images of this one. Okay. And it’s like, when I say vacation rental 1, 2, 3, I mean, this one and actually train it on that and go, okay.

Now we’re putting this one on the moon, right? Or whatever it is and sort of like have some, some fun. But I think it, it more in those applications leans more to the social side than on the direct you know, sort of sales side.

AMY HEBDON: I think there’s potentially an opportunity there. Cause like, I, I’ve been playing a little bit with MIT journey and it is, it is amazing.

It’s like, I had this dream when I was 10 and it looked like that. And it’s like, this is what your dream looked like. Yeah, that is, you know, but, but as far as like the, the application for my clients, it’s not there, but I, I can imagine a time where like, you could just drop in your product and say, Hey, I want it, I want it, you know, these different vignettes of this being in use or, you know, having a good product shop, because that’s the one thing I’ve seen, I’ve noticed so far that it’s not really good, like you would have to like, It just doesn’t natively do that.

And so if you wanted to quickly kind of clip your thing in and kind of create this thing, I, I feel like that’s where this is headed, that that could be a good commercial application for it.

FREDERICK VALLAEYS: So you’re both, oh, go ahead.

DAVE DAVIES: Well, yeah. And if we look like, I don’t know if people remember Dolly mini, like just because we’re, we’re sort of chatting with Dolly, but it took off last spring.

And if we look at the images, it was generating. They were, it was, it was good. It was the best we’d seen. And that’s why it sort of went viral and people were doing it. And the New Yorker had it on the cover of their, their magazine. But it was no matter what you put in and how well you described it and tried to prompt engineer engineer your way through it, you ended up with kind of nightmare fuel or like people with like 84 fingers.

Right. And that’s just sort of what we had. If we flash forward, just like. 7, 8 months when Dolly 2 was like, you know, put on to public beta and we could start to use it and the stuff we’re doing a text generation now versus a year ago. And we look at that gap. I think we, as marketers, and I know everybody here is thinking about it and go, okay, let’s flash forward before I even hit 2024.

We’re going to be in a way different scenario with what these things are going to be capable of doing. GPT 4. 5 will be out at that point. Dolly 3 point something will be out at that point and we’ll probably be able to connect the dots and go make me a book about and it’ll just make a book about a thing connecting like images and text generation, right?

Like so I think prepping for that is a top of mind, I think, for a lot of marketers as well.

FREDERICK VALLAEYS: And let’s talk about that more from an SEO or a PPC perspective, but to Google eat criteria. Okay. So about authoritativeness and expertise experience, what happens with that when you can go and say, write me a book on the topic of PPC.

All three of us, I mean, we have a brand name in this. So we have the authority. We can slap our name on it. Is that going to fly with Google? What’s, are we still going to have to do work? What do you think is going to happen, Dave?

DAVE DAVIES: In my mind, the threat here isn’t to Google. And it’s not even from each other, like somebody writing a.

You know, a book and then going in the tone of Dave Davies, right? I’ll go, then you’d end up with the kink singer, right? And then off you go. And it’s a completely weird SEO book. But I don’t think that’s the, the threat in this case. The mall had mentioned earlier that Google’s working on something.

It’s called Sparrow. They’re working on their like sort of chat GPT. I’m sure it’ll be named something different when it comes out. I believe that’s, that’s their sort of like. Working name of it right now. So they’ll come out with one based on Lambda and stuff. It’ll be a chat GPT thing. So now, rather than thinking of me generating it, or you generating it, or Amy generating it, or our competitors generating it, think of Google going, why am I even sending anybody to you?

I already understand everything in the world in real time. I’m going to generate it. And that’s where I think the threat is. To us more comes in.

AMY HEBDON: Well, can I, can I ask you something about that? Because as it seems to me, like at this point, anyone who’s going to rank on SEO and I’m not an SEO expert, but it seems like anyone who ranks in the SERP, you know, for Google, they have an incentive.

Every entity has an incentive to be there, right? They either paid for a technical SEO person to do it because they have something to gain from people visiting their site or, you know, they’re, they’re a domain expert because again, they have something to gain from people believing them or whatever. But a lot of those, A lot of the listings that appear are there because they get there’s there’s ads on their page, right?

We all know that when we go click something and it’s like you’re just bombarded with ads I I have a hard time imagining Google creating content where also post ads on it and I could be wrong and maybe you know They they have like this authoritative content and say by the way and have all these pop ups It just seems a little bit off putting Brand to them if they’re like, Hey, we’re trying to make the world’s information accessible.

And by the way, watch our pop ups. It just, it feels really, really like a regression to me. So I’m having a hard time figuring out why they would benefit from that unless they had ads, but why they would want to do that to have, you know, if they’re, if they do or don’t have ads. So I I’m, I’m curious what you think is I’m definitely not an expert in this domain.

DAVE DAVIES: No. And this is like sort of a. Like, who really knows? We all just like watch you know, and we read what Barry Schwartz is doing, going, okay, there were like 14 different tests on the pages this month. Right. And like sort of how were their, their fine tuning stuff. My guess is, well, we have sort of two problems, one, which would require Google’s foresight that I don’t know if they, they have it because they’ve never really shown me.

They have it in this one, which is, well, if you take away the incentive for that publisher to sell advertising on their page, Why are they producing that content? So now what are you training your system on? Right, like, if all the publishers just sort of go, there’s nothing for me now, I’m bankrupt. Okay, that’s that’s a different problem.

So now they’re they’re training set or knowledge set is gone. But additionally, what I see them sort of going, we’re going to create a content page. My instinct would be no, but what it like, all of us are skimming really fast. All of us are like. Looking at structures of pages with all sorts of affiliates and going, okay, we know Dave’s looking up this location.

We know the last thing he’s looking at that. We know Dave has like an attention span that is actually going to want him to have, like, if he’s read this far now, he wants a video, right? Like where they’ll be able to personalize that content layout and go. Yeah, chat GPT. This version, let’s just put one paragraph and then give him a video.

Cause if he doesn’t see it in one paragraph, he’s probably going to want a video. Let’s truncate that to three minutes, right? Like, because he’s like trying to solve a problem and then move on to like lists with affiliate links, but they’re actually just paid ad people, right? They’re not affiliate links, but working more on that.

Cause I think that’s one of the problems that both Microsoft and Google would have is how do you put ads in it? Yeah. You can insert a chat function. But how do you get ads in it? And I think that’s going to be super fun. Like Frederick for, for your company is up because they’re going to have to pivot really fast.

And then all of a sudden there’s going to be a bunch of people going, I can’t pivot that fast. And then they’re going to look to you guys.

FREDERICK VALLAEYS: And to me, at some level, this feels like voice search, right? So voice search for the longest time has been a buzz that this needs to be monetized. There need to be ads in it.

But it’s a fundamentally different problem because again, you don’t have 10 results. You have the one thing that this is the answer. How do you weave something into that answer? That’s maybe an ad. So if I’m saying, how do I fix my toilet? Well, ChatGPT can probably give me the instructions and then weave in the fact that, Oh, and there’s Ace Plumbing who could do this for you in case you didn’t want to do it.

Right. There’s also a level here where do we even need to formulate the query or is Google going to know based on everything that’s happening? Like, okay, this is on Fred’s mind right now, which they’re already amazingly good at, right? I mean, how often do we see ads for something that we’re like, I didn’t really search for this.

Like, was, was Alexa listening to me in the room? But it’s like, okay, people behave in certain ways. And even if you don’t formulate it, well, there’s so much data floating around that they know what you’re going to be interested in. It might start feeding you some of those ads and then you’re like, Oh yeah, let me go deeper on that and research it.

And then the ads come back in. Now, the concern that I would have is us as marketers in that vision that you’re painting, Dave, we are becoming unnecessary, right? Because Google itself has all the answers. Google knows all the businesses that are out there and what services they provide. And so they will know how to weave them into the right searches.

So what is it we end up doing? A plumber for now until there’s plumbing robots. Sure. They can go and do their thing and they’re going to get money for that. But us as digital marketers, having connected these dots for so long, What is it that we will bring to the table? And that’s maybe the scary vision.

Like, are we still necessary?

DAVE DAVIES: I would argue yes, but that’s maybe because I’ve just heard SEO and PPC are dead for just so long. But I think in the PPC realm, like, since we’re talking about PPC right now, I have a hunch it’s going to get maybe a little more technical, like going, okay, how do we work with the mechanics to make sure that They understand that.

Okay. They gave Dave some instructions. Now they’ve heard some very colorful language. That’s when we need to inject in and we Dave need somebody right now because we can hear the water hitting the floor. Right? So, you know, on the paid search marketing side. We’re still going to need to market and human beings, at least for the foreseeable future, we’ll, we’ll actually inherently understand each other better than a machine will but also being able to connect the dots and go, this guy or girl is available right now.

And Dave needs them right now. And they’re three blocks away connecting those dots for, through the ad system. And that’s

FREDERICK VALLAEYS: technology. That’s technology we’re working on. But so to say that you’re a plumber, how do you connect your availability of service technicians to the demand that’s out there? And how do you like scale it up and down?

Ultimately, these are all technical things, right? And so these are the things that the machine, if you show it enough examples, it can start doing this. You don’t have to do it anymore, but you touched on the human elements. And maybe that’s where I’d love to hear from Amy, like when we’re shopping for vacations.

Like when I went to Hawaii and I saw this beautiful sunset, how did that make me feel? That’s sort of the thing that we started this whole conversation with is that the machine can put the words together, but it can’t have these experiences. And so we as humans, we can probably write something. We can be creative about this in ways that the machine just doesn’t fundamentally get.

That still makes that human to human connection. And so based on those examples, maybe over time, the machine will learn to replicate that. But for now it can’t.

Amy, I’d love to hear from you. Oh, sorry. I thought you had that,

AMY HEBDON: James. I apologize. I mean, right now it can’t, right? Like, I don’t think there’s any question about that. Like a, a good copywriter can outperform, you know, where we are at with AI and AI can outperform a bad copywriter or someone who’s not paying attention or someone who doesn’t know that doesn’t have those skills.

And I, I use copywriting rather than just marketing because I feel like it’s, it’s a discipline within marketing that gets you So overlooked, like so much of the time is just like, well, we just have to have a landing page. Just have to have someone, you know, somewhere for them to go to. And in the example that you were just using, like, yeah, you know, a, a plumber, like that sort of demand, like that sort of like immediate need there, there’s no landing page there, there’s just that, that connection.

That’s really what you want. But a lot of times there is a decision making process and you need to be able to show your prospect a better version of themselves through your copy, through how you’ve arranged all the elements. And because a you know, machine learning and AI only know about you, what you tell it, you have to know what your benefits are and your key differentiators, and you have to know how to tell that story in order for it to be told.

Like, I, I don’t think we’re there yet. Will we get there someday? Hopefully, maybe, you know,

FREDERICK VALLAEYS: And let me go back to the example that we ran a little bit earlier about the luxury vacation versus the backpacker vacation. The machine came back and basically included those words in the headline, which is not that hard to do, right?

But if you’re a backpacker, maybe you stayed at hostels that weren’t super safe or that weren’t very clean. And we understand that that’s, that’s the thing that we as humans, we like a safe and a clean place. So why didn’t it use those words? As something for backpackers. And when it came to luxury, why did it just name luxury as opposed to that feeling that you get it?

Like I might stay at a hotel where celebrities, A listers are staying. Maybe that’s what I’m into. I want to see that. Right. So, and that’s the human connection that it hasn’t quite made.

AMY HEBDON: But you can ask it. You can say, why did you choose something so basic as to just put the word backpack in? I mean, you could go right now and ask it and it would give you a reason.

It might be good, it might not be a good reason, but that’s still available. Like, I’ve asked it, why did you choose to put the domain URL in description one and description two? And it told me, it’s like, actually, that’s not really a good idea to do, but I did it because I was trying to write, you know, for this reason, and here’s another way you might want to do it.

And it is very conversational, even though it’s AI, it’s still, it’s not just static, like accept or reject the results and judge it, it’s like, you can still collaborate with that, you know?

FREDERICK VALLAEYS: Yeah, so here’s the collaboration. So I asked him, why did you use the word backpacker? And it says, I use the word backpacker in the headline because it’s common term used to describe the type of traveler who typically travels on a budget and carries all their belongings in a backpack.

This type of traveler often seeks out more budget friendly options. May be more interested in experiencing local culture and immersing themselves rather than luxury and comfort. Therefore, backpacker strips sounds like a perfect headline for a travel agency that targets and specializes in budget friendly and cultural trips.

So that’s interesting because now it’s saying, okay, the local cultural experience is maybe what this type of traveler wants. And then maybe if we’d asked it, give us five headlines for backpackers, those terms would have eventually been used as well. Very fascinating.

AMY HEBDON: You know, budget friendly cultural trips and you can start to use those, like give me an ad for this.

And then you start to build on that. So you’re, you’re starting to get away from just like the very basic mechanics of, I think when we’re defining an audience and we did that here, right, with luxury and, and backpackers, we’re looking at demographics and psychographics and no one, Makes decisions like purchasing decisions based on that, you know, there’s that quote, like you don’t buy the wall street journal because you’re 42 and a white male, like, there’s other reasons that you make decisions.

And we kind of use segments to make it easier for us to make targeting decisions. But the reason that someone’s making those decisions. It’s complex and nuanced. And so the better we can understand that, I mean, then the better job we can do, but I think that it’s not so different that we can say, Oh, AI will never know why someone decides that.

I think, I think we can use it along the journey to help us on that process, if not today, then in the near future.

FREDERICK VALLAEYS: And I really like that, how you’re thinking about the AI. As a colleague that you can bounce ideas off of and sort of have a conversation about why they went that direction. It doesn’t mean you have to agree with it, but it can open up some new thoughts, like those additional words for the backpackers that maybe I hadn’t considered.

And then I can go down that path and explore it deeper.

AMY HEBDON: Right.

DAVE DAVIES: Fascinating. I really liked, sorry, I really liked something you mentioned, Amy, that I think is just worth Sort of focusing on as a value prop for humans which is it does understand everything in the world. Basically, like at the time of training, plus anything it’s acquired since then, it does understand everything.

So is it smarter than me? Yes, hands down. It has access to far more information than I do, but can it create an original idea? And that’s where a human Can offer its value, not just in understanding sentiment to a level that that a machine can, but also in going, these people are tired of seeing every ad they’ve seen for the last decade time to come up with something new and and a machine just at this stage, like, we don’t have artificial general intelligence and we’re probably a good chunk away from that.

You know, for the foreseeable future, it’s going to take a human to actually create something original. Now, the machine will like immediately rip that off and start using it. But for a short duration, you’ll have that original idea and be the only one with it.

FREDERICK VALLAEYS: And people who’ve seen me speak before, I’ve heard this example, but I talk about AdLibs.

And so this was an early project in Google Ads in the early 2000s. It was not artificial intelligence, but it was basically saying, hey, too many local service businesses struggle with how to make a good ad. So why don’t we just give them a template that says if you’re a restaurant, tell us the type of cuisine, the location you’re in and some other like basic facts.

And just like in Mad Libs, it was fill in the blank in the template. The problem was very quickly that it had all these ads, but they all started looking exactly alike. And consumers, the users of Google, they hated it. They stopped clicking on the ads because they felt cookie cutter, felt like it was always all the same.

And now we’ve seen that where we have, Responsive search ads and we have automatically generated attacks and it is A little bit more variety, but like you’re saying, it could go down the path where eventually it just, all the ads look the same as humans. We end up not differentiating. So we just click on the first one and that’s a big problem for Google, right?

So there needs to be variation to let evolution happen. That’s evolution. That’s how evolution happens. You need to make mistakes. You need to try things for the best to be able to rise to the top.

AMY HEBDON: Yeah. I think one, one interesting thing that I don’t, I don’t. Totally understand. I’ve just, you know, experienced it again from a consumer side is what motivates a I to give the responses that it does.

You know, I think I mentioned I went back and forth several times over several days saying like, What is the character count for this phrase? And I kept on giving me the wrong one. And then finally, it’s like you tell me what it is. So I said, Okay, that’s 44 characters. It said, You’re right. It’s 44 characters.

I said, If I had told you it was 43 characters, would you have agreed with me? It said, Yes. If you had said 43 characters, I would have agreed with you, and that would have been incorrect. So, just, you know, it’s, it’s, It’s like

FREDERICK VALLAEYS: a customer service rep who wants to get you off the phone, and they’re like, yeah, whatever you say.

AMY HEBDON: Right. Let’s not talk about

FREDERICK VALLAEYS: numbers here.

AMY HEBDON: Exactly, but it kind of speaks to, like, where, like, You know, the quality of, of inputs, of prompts, of questions gives, you know, determines the quality of the outputs. If you say, tell me why AI is good, it’ll tell you why it’s good. If you say, tell me why it’s bad, it’ll tell you why it’s bad.

So we really need to think through, like, not just ask anything, but like, think through the kind of problem we’re trying to solve in order to get back the an answer that’s really useful to us because it’s trying to give us the answer that we want not necessarily the answer that is right, but the answer that we’re looking for.

FREDERICK VALLAEYS: Sorry, say that one more time, the last thing. So it’s going to give us the answer we want, not the right answer. So the echo chamber effect, that’s such a big problem right now. Basically, this is going to. Make this even worse is what you’re saying.

AMY HEBDON: That, that’s been my experience so far is that it gives answers back.

I mean, and Google does this too, right? If you say food, it’ll show you pictures of food. But if you say, where can I get Thai at Saturday night at 10 p. m., you know, in my city, it’ll give you those results. It tries to tailor. And that’s fine because it’s still

FREDERICK VALLAEYS: food. It’s still food. But if you ask it is, is the moon a planet?

The factual answer that humanity has, you know, Mostly agreed on is no, it’s not a planet. Well,

AMY HEBDON: I don’t know, you know, I, I would hope, you know, I, I have seen from my, my usage that you know, it is willing to argue with me. It is willing to defend, sometimes an indefensible proposition, but it is willing to defend the things that it believes to be true at a certain level.

Math just doesn’t happen to be one of those things, but, you know, animal rights does, and, and good for it for taking a stand about writing, you know, songs from the perspective of a dog. But I think there’s probably at some point that it’s like, I will only give this factual information, but there are also like how it, the, the lens through which we interpret it and like the point of view and the conclusions, it does seem to be that there’s some flexibility.

FREDERICK VALLAEYS: Wow. It’s a strange new world. All right. So this has been an amazing conversation. Amy and Dave any final points or tell us tell the viewers how they can get ahold of you and how you can help them. Amy, why don’t we start with you?

AMY HEBDON: Sure. Well, I’d say my final point is to think of a chat GPT right now, like a very precocious intern that you will need to spend time, you know working with it, but it’s also free labor.

So at least for now, right? So you don’t, you don’t just like say, do this and walk away. Like you need to review the work and, and kind of spend time there, but it’s, it is valuable, it is helpful and useful. And you can find me, I’m on Twitter at Amy PPC. I don’t tweet a whole lot, but I’m there. LinkedIn.

And then we’ve got a Facebook group called Google ads for savvy digital marketers. We try to keep it really clean of spam and like the, you know, things you just don’t want in a, in a community. We try to keep all those clean and just have it like a really useful answer for, you know, where we can find answers to like their questions.

FREDERICK VALLAEYS: I love the whole intern example, and this is more of a technical question. I don’t know if either of you will know the answer, but if you treat it as an intern and you train it and you have conversations with it, the next time you log in, will it remember your preferences?

AMY HEBDON: It, it says it won’t, but I don’t know if that’s true or we will remain true, but it it, it resets with each, each unique conversation.

FREDERICK VALLAEYS: So every morning you have a new intern at the company.

AMY HEBDON: Right. Or every time, every time you have a network problem, then it all starts.

FREDERICK VALLAEYS: Dave, any final thoughts from you or where do people find you?

DAVE DAVIES: Yeah, my, my final thought, we, we talked about performance max earlier, or you referenced it in one of the examples. And I think I’ll just use that to sort of highlight the problem with these, these sorts of systems from a search and why I don’t think we’ll see a world where Google or Bing or any of them would or should use them.

And if you ever look through your assets in a performance max campaign, you’ll see that Google at one point goes, yep, those are the three. And then the rest of them have like one impression, two impressions and a bunch of them have like thousands. They sort of will go. Like they did in the example of the local ads that you were talking about, they’ll go, that one works, and then just hammer it.

That’s fine when it’s just my ad, and a different human built another ad, and a different human built another ad, and we all had our creative juices, and nobody’s going to see the same thing regurgitated. But if they tried to pull that off en masse by going, yes, let’s unleash our AI on this entire system.

We’d end up with some basically like three or four ads and eight images that got used for everything across across the web. Anybody wants to reach me, the easiest way is probably on Twitter with the handle online inference is yeah, happy to connect.

FREDERICK VALLAEYS: Great, amazing. Thank you both for being on.

Thank you all for watching. If you’ve enjoyed PPC Town Hall, subscribe, and you’ll get updates about the next episodes we’ve got coming up. We also have the blog at Optmyzr, and if anyone is looking for a PPC management tool, Optmyzr has a two week free trial. Look at that. Amy, Dave, thank you very much, and we’ll see you for the next Town Hall.

Thanks

AMY HEBDON: so much.