Google Explains How to Leverage Its AI Tools for Profitable Ad Campaigns

Episode Description

Tetsuo Konno, Search & Performance Lead at Google shares how Gemini and predictive AI are changing the way marketers approach Google Ads campaigns.

He also teaches how to set up your data properly for Google Ads and why that matters more than ever.

Here’s what you’ll learn:

• How to use predictive and generative AI in Google Ads
• Why feed optimization is key to better performance
• Practical examples of AI in action, like ad copy and data insights
• The future of PPC roles and skills

Episode Takeaways

AI is making big promises in PPC, but is it actually helping advertisers drive better results?

Fred and Tetsuo cut through the hype to reveal how predictive and generative AI are changing bidding, search behavior, and campaign strategy.

The power behind predictive AI in marketing

While everyone’s talking about generative AI these days, the conversation highlights how predictive AI remains incredibly valuable for businesses. Tetsuo shares a compelling example of how predictive AI solved a major problem for a Dutch fashion retailer with high return rates.

Next Step👉If returns are cutting into your margins, don’t just accept them as a cost of doing business. Work with your data team (or a tool like Vertex AI) to analyze return patterns and build a predictive model. Then, start adjusting bids based on return probability to protect your profits.

Value-based bidding is more than just ROAS

The conversation reveals how advanced marketers are moving past simple ROAS (Return on Ad Spend) metrics to more sophisticated value-based approaches that factor in true profitability. This allows for much more accurate optimization.

Next Step👉 Shift your strategy from chasing high ROAS to maximizing actual profit. Feed Google Ads the real value of each sale by including margin data, return risk, or even expected lifetime value in your bidding model.

How user search behaviour is changing

Tetsuo offers fascinating insights into how generative AI is fundamentally changing the way users search and interact with results. This isn’t just a minor update to search but a complete reimagining of the user experience.

Next Step👉 Audit your website and ad assets to make sure they’re AI-friendly; meaning structured data, high-quality images, and clear, informative content. If you sell products, make sure Google Merchant Center feeds are optimized with detailed attributes to improve discoverability.

Higher quality traffic from AI-enhanced search

Contrary to what some might expect, AI-enhanced search drives more engaged website traffic, not less.

Next Step👉 Stop fixating just on CTR and start focusing on what happens after the click.

A slow, confusing, or irrelevant landing page can tank conversions, no matter how good your ad is. Speed things up, refine your messaging, and make it easy for visitors to take action. If bounce rates are high, your page might not match what searchers expected.

And if your landing pages aren’t working, neither are your ads. Use tools like Optmyzr’s URL Checker to catch and fix bad URLs before they waste your ad spend—so every click has a real chance to convert.

AI’s impact on campaign creation and management

Beyond just writing better ad copy, AI is changing how campaigns are conceptualized and managed at scale. This opens up possibilities for creative work that was previously too resource-intensive.

Next Step👉Test AI-generated creative assets for your campaigns. Try different variations of ad copy, images, or videos and see which resonates most with your audience. Then, refine your messaging based on real data.

Feed optimization: a hidden growth opportunity

AI isn’t just for ad copy; it can also enhance feed quality. Tools like Google’s FeedX and FeedGem use AI to generate structured, optimized product feeds, improving impressions and conversions.

Next Step👉Ensure your product feeds are complete and well-structured. Test AI-driven optimizations to improve match rates and lower CPC.

Democratization of data analysis

One of the most profound impacts of AI is how it’s making sophisticated data analysis accessible to non-technical team members.

Next Step👉Make data-driven decisions in real time instead of waiting for formal reports. Use AI-powered analytics tools to pull insights on campaign performance, audience trends, and optimization opportunities.

The future of PPC specialists

The conversation concludes with thoughtful reflections on how PPC specialists need to evolve their skills to stay relevant in the AI world.

Next Step👉 Don’t just manage PPC campaigns and take a step back to focus on strategy, cross-channel expertise, and AI integration. Learn how to make the most of AI, automation, and data analysis to drive bigger business impact.

Take control of AI-powered PPC with Optmyzr

AI is changing the way PPC marketers work. However, success isn’t just about using AI, it’s about using it well.

That’s where Optmyzr comes in as it helps you use AI without losing control.

If you’re tired of guesswork, wasted spend, and time-consuming manual optimizations, these tools can make a real difference:

• Too many optimization tasks, too little time? Optmyzr Express gives you instant, one-click suggestions to improve results
• Not sure what’s hurting your account? Run a PPC Account Audit to find wasted spend and missed opportunities
• Seeing performance swings but don’t know why? PPC Investigator pinpoints the root cause so you can take the right action

AI is powerful, but only if you use it to solve real problems. With Optmyzr’s suite of tools, you get the best of AI-powered optimization while staying in control of your strategy.

Not an Optmyzr customer yet? Now’s the best time to sign up for a full functionality 14-day free trial.

Episode Transcript

Frederick Vallaeys: Hello and welcome to PPC Town Hall. My name is Fred Vallaeys, and I’m your host. I’m also the CEO and co-founder of Optmyzr, a PPC management software suite. In today’s episode, we’re going to talk about generative AI and marketing— a topic we’ve discussed quite a bit, but what makes today’s episode unique is that we have Tetsuo Konno from Google.

Tetsuo is here to share Google’s perspective on how generative AI can help marketers with better keywords, targeting, and ads. And of course, being from Google, we’ll dive into how Gemini fits into all of this. I’m really excited to hear Google’s take on generative AI, and I hope you are too.

With that, let’s get started with this episode of PPC Town Hall. Tetsuo, welcome to the show! It’s great to have you on.

Tetsuo Konno: Hey Fred, yeah, thanks for having me. Glad to be here.

Frederick Vallaeys: You’re calling in from Google in the Netherlands, right? Tell us where you’re located.

Tetsuo Konno: Yes, exactly. I’m based in the Google Amsterdam office, which is a bit towards the end of the day here. My role is a performance specialist, and I focus on helping our largest advertisers and agencies in the region get the most out of Google Ads.

I’ve been doing this for about six and a half years, and prior to that, I worked at iProspect for about four years. So, I’ve been in the PPC game since the days of manual bidding.

Frederick Vallaeys: That’s impressive! And some of the world’s largest advertisers are based in the Netherlands, like Booking.com, which is a Dutch company. They really know what they’re doing, and they tend to be pretty sophisticated in their strategies.

I always enjoy talking to someone from the Dutch market because it’s such an interesting scene. There are so many agencies based there, and a lot of tools are actually developed in the Netherlands as well. It’s a market where some of the smartest minds in PPC are all gathered together. I’m really excited to hear about your experiences, both from your time at the agency level and now at Google. And, yeah, I think some of the guests we’ve had on PPC Town Hall before are actually friends of yours, right?

Tetsuo Konno: Yeah, exactly. As I mentioned, I used to work at iProspect, and a few of your former guests, like Leinand from TrueClix and Marcel from Roots, I’ve worked with them. They taught me a lot about PPC and how it works. We used to work with Optmyzr quite a bit during our agency days.

For me, it’s a full-circle moment—having used your tools and read your books, and now being a guest here. I’m really excited to be here. Also, the way I got here is that I connected with your colleague, Neva, last year. We were both speaking at Friends of Search, a conference here in the Netherlands.

Also, I think a very great conference based out of, again, the tiny little Amsterdam.

Frederick Vallaeys: Yeah, it’s really interesting because I was just in London for SMX London, and it seems like in the United States, a lot of conferences have largely gone virtual. Which is great for me because I can sit in my studio, record something for an hour, and get it out to lots of people without having to travel. But in Europe, there’s still much more of that one-on-one connection at conferences.

For anyone who hasn’t checked out Friends of Search, it’s quite a large event and very well organized. They have great food, excellent speakers, and it’s in a fantastic city. The event usually takes place in late winter or early spring, so definitely check it out.

I’ll be speaking there this year, and they tend to rotate speakers, so I think you spoke last year, and you’ll be taking this year off and then coming back the year after.

Now, you can sit back in the audience and enjoy the show instead of being a bit nervous about speaking! I’m sure you’re looking forward to that.

Tetsuo Konno: I’m definitely looking forward to sitting back in the audience and enjoying the show instead of being a bit nervous about speaking! But yeah, I’m very curious about your talk as well.

Frederick Vallaeys: Yeah, my talk will be about AI, what talk these days isn’t, right? Tell us, what was your talk about last year at Friends of Search?

Tetsuo Konno: My talk was partially about AI as well, but we presented a broader framework on how we think marketers and PPC in general could be more successful by following a three-step framework.

We called it the AI marketing flywheel. We focused on generative AI, but also emphasized the importance of predictive AI, setting the right goals, having good data, and implementing a solid testing framework to continuously validate strategies. That was the core of my talk last year.

Frederick Vallaeys: Yeah, and that’s interesting because you just mentioned predictive AI, which was the state of the art a few years ago, before ChatGPT showed us what a generative transformer can do. But honestly, Quality Score, which has been around at Google for what now, 20 years? That was a form of artificial intelligence. It was a machine learning prediction system, and it still is to this day. So, in some form, we’ve all been using AI for much longer than we realize.

But now with generative AI, it’s opened up whole new doors, right? It’s no longer just about giving something to Google and letting it use AI to predict what will happen. Now, we can go to Google and say, “Hey, I don’t know what to give you—help me with ads, help me with images, help me with keywords.”

So, maybe you could give us a broader perspective on the state of AI right now and how you see it evolving.

Tetsuo Konno: Yeah, exactly. So I think, exactly like you said, generative AI is obviously revolutionizing a lot of the steps in the marketing journey, but as I alluded to earlier, I think there are still a lot of powerful opportunities out there that are maybe not generative AI and are still being used by a lot of us, maybe without realizing it, like smart bidding, for example, which has been around for years but is still very powerful, and also predictive AI in some other areas.

When you think about the steps to be successful in AI, I think as a PPC marketer, setting the right goal to start with, having the right data in place, and the right KPIs is a very interesting starting point. Then, feeding that to AI and enhancing it with generative AI to improve is the framework that I use a lot. But when we think about those goals, I think predictive AI can have a very impactful role in that by predicting future outcomes, for example.

So, if I can give an example, I used to work on a case, so maybe I’m a bit biased because I worked on it myself. I was actually working with a Dutch fashion retailer, Omoda, and their goal was obviously to sell a lot of clothes profitably. However, a big problem we were facing was that 50 percent of their bought items were actually returned, but you don’t notice until about 30 days after the fact. So how can we make sure that we feed bidding with the right data? What we did there was use Vertex AI, our enterprise AI environment, to predict, based on historical data and BigQuery, the likelihood that someone would return a product at the moment of a transaction. The value that was predicted at that point was fed into bidding.

It’s obviously not a hundred percent accurate, but it was around 75 percent accurate and actually drove a lot of bottom-line profitability for the company. I think that just speaks to the fact that generative AI is definitely great and we’ll go into it much deeper in this podcast, but don’t forget, I would say the power of some of the predictive AI that’s out there as well to help predict what a business outcome might be.

Frederick Vallaeys: And that’s super interesting. And I think a lot of experts talk about this need for haring the correct goals with the machine and think of the machine as a colleague and a colleague couldn’t do a good job if you didn’t communicate what it is you want. And so sometimes you say, I want lead forms being filled out on my page, but that’s not really what you want, right?

You want those leads to be high quality. And so we see problems at both sides of the spectrum where some advertisers are simply too small to get enough true conversions, even like a good lead form fill. And so the struggle then becomes how do you go higher up in that consumer journey to maybe detect that they downloaded a white paper or engaged with certain pages.

And maybe that’s a good signal. What you’re talking about is that other end of the spectrum, or maybe it’s a bigger advertiser and they already get like actual sales data. They get actual profitability data, but how much of that is being returned? And to talk a little bit more to like, when you, when it comes to using Vertex AI, like how big of a project.

Is this truly reserved for enterprise-level advertisers or what does it take? What staffing do you have to have? How much data do you need to have to make this work?

Tetsuo Konno: Yeah, I think in general, personally speaking, that was definitely the case a few years ago, but there have been a lot of developments recently, not just in generative AI, but across the board, making these models and their implementation much more accessible.

For example, we have models like the returns prediction model, which was once very custom and required data scientists or close work with an agency to get off the ground. That worked well, but now when you think about things like predicting lifetime value for a specific audience, we have templatized solutions available, like Crystal Value, which is even hosted on GitHub. While you’d still need some understanding of how to navigate a GitHub repository and deploy the model, the barrier to entry has lowered significantly.

This opens up opportunities for a wider range of advertisers to leverage these tools and apply them meaningfully to their campaigns. The resources needed to make such projects work are steadily decreasing, making advanced AI-driven strategies more accessible to all.

Frederick Vallaeys: Interesting. Do you think any of these capabilities could also be deployed in a scenario where you maybe do have the data scientists, but you don’t have enough conversion volume at the start of a campaign? You’re maybe looking higher up in the journey and saying, okay, we got all of these visitors to the site and they’re doing something, but use a predictive model to say of these behaviors, which ones correlate to that end state of the sales qualified lead.

Tetsuo Konno: Exactly. And I think you’re referring to the lead generation example here, but this principle applies broadly. For lead generation advertisers, especially those with long sales cycles, it’s challenging to provide accurate bidding data for that final conversion because the path to purchase can take, say, 90 days after the initial click.

To tackle this, both large and small advertisers often rely on signals or micro-conversions that can serve as good predictors of the eventual conversion. By leveraging a lot of historical data, you can make an initial prediction about the value of that conversion. In real-time, when you feed this prediction into the bidding algorithm, it helps optimize the process even before the final conversion takes place.

Frederick Vallaeys: Yeah, and you made a great point about potentially shifting us toward lead generation, but let’s stick with eCommerce for now, since that was the example you gave, right? So, in the case of sales, let’s say a particular fashion brand has a high return rate, like half of their sales are returned, or even 70%, as I hear happens in Germany. That’s a tough challenge to manage.

When it comes to returns, do you also factor in the profit and margin data, or do you just focus on the basket value? What are some other layers of optimization you could apply here to make sure you’re optimizing for what truly matters to the business, which, at the end of the day, is usually profit?

Tetsuo Konno: Exactly. And I’m glad we’re diving into this topic, because more and more advertisers are starting to realize that just focusing on ROAS (Return on Ad Spend) isn’t always the most meaningful metric. At the end of the day, most businesses care about maximizing profitability, both in the short and long term.

I can reference the Emoda example here, which is a public case that has even been mentioned in the Wall Street Journal. If you’re curious, you might want to check that out. For Emoda, profitability is absolutely key, and they’ve realized that while ROAS is important, returns have an outsized impact on the bottom line.

So, what we did for them was take a closer look at the basket being purchased. We calculated the margin for each transaction, but then we also factored in return predictability. This led us to come up with what we call a “profit value,” which incorporates both the margin and the likelihood of a return. This, in turn, provides a more accurate and actionable metric for profitability that aligns much better with their business goals.

Frederick Vallaeys: Yeah. And that’s when we have terms like value-based bidding, which can be, it’s just generally deployed for e-commerce where you want to communicate the actual value of something, but you can also do it on lead gen, right? So you could say, okay, this lead is worth more than that lead. And so hence the bidding algorithm can become more sophisticated.

Now, when it comes to value-based bidding historically, like when it comes to things like returns, this usually pivots on the GCLID or some transaction ID, some parameter that Google gives to the advertiser, you put that in your system, and later on, you communicate back to Google. For this transaction ID, actually, it wasn’t a transaction, or maybe this much was returned, or for this GCLID, it turned out to be a really good lead or a really bad lead.

But then there are questions about durability, right? Google talks about the durability of these tracking solutions quite a bit in a more privacy-centric world. So talk us a little bit through What’s the state of the art today in terms of the data pieces you need and what are we looking at tomorrow to make sure this will work into the future?

Tetsuo Konno: That’s a valid point, and it’s something we’re discussing with many advertisers. Many are still relying on things like order IDs or Google Click IDs, but we expect that this dependence on single identifiers won’t be sustainable in the long run. Our recommendation is to move toward a more robust measurement approach that’s tag-based, whether client-side or server-side.

This shift could help address challenges where you only know the true value of a lead or product after the fact. One potential solution to this, especially in cases where you need to predict outcomes before they happen, is using AI for prediction.

However, there’s definitely a tricky balance between how quickly you can provide that information and how accurate it is. But given the current state of the measurement ecosystem, moving toward tag-based solutions makes sense, especially for lead generation. We also have solutions, like enhanced conversions for leads, that still rely on HPII data and allow for uploads after the fact in a durable way.

Frederick Vallaeys: One common question I hear about identifiers and uploading data after the fact is about the window that Google allows for reporting conversion value changes. Even though Google provides a fairly long period to make those adjustments, I often hear the message that “the sooner, the better.” Even if the data is less precise, you’d rather know quickly whether something was good or bad than wait two months for an exact figure. From your experience, what timeframe would you recommend for this kind of reporting? And also, consistency is another factor I’ve heard emphasized. Does it matter more to consistently report back after the same number of days, rather than varying the reporting time, like sometimes after 5 days, other times 15 days?

So talk about that.

Tetsuo Konno: To start with the latter part, we definitely recommend consistency in upload times. This helps keep the algorithm stable. If you’re using offline conversion imports or conversion adjustments, try to maintain a consistent timing and frequency for those updates. This will be beneficial.

As for the first part of your question—how long is too long—it’s difficult to answer definitively, and unfortunately, we don’t have specific statistics on that. The general principle is that the sooner we receive that information, the better. Even if the data isn’t 100% accurate, it might still be better for bidding algorithms to work with that rather than constantly making adjustments. Consistency and frequency are key, and it’s also really dependent on the advertiser. So, while it’s a bit of a typical Google answer, the sooner and more consistently we receive data, the better the outcome.

Frederick Vallaeys: And I think it’s a fair answer. It always depends in PPC, but that’s where you get into these complex situations, right? So say that somebody buys a bunch of clothes and returns two pieces after a week.

And then another week later, they’re like, these other bunch of pieces come back. So you end up having these multiple stages when you could communicate back to Google that some amount of value has changed. but I guess your point is don’t make these micro adjustments too frequently. If there’s like a big, meaningful change, then report that back, and try to be consistent about it.

And usually, the returns policy window is the thing that gives you at least some consistency. Like after 30 days, we can say this is now a locked-in conversion value. But if you think about machine learning in general and statistics, if you basically go two months of having assumed one thing, and then after those two months, you say, okay, actually, what’s different, there’s like this bigger window in which things were still going the old way for that 60 days of bad data to be smoothed into the new averages of where it’s better.

It’s going to have, it’s going to take longer for you to see the results that you want. Whereas if you report back to Google within three days, that something was good or bad, then you only deal with that three-day window when it was imperfect. And that gets smoothed out much more quickly. I think from some level, it’s not purely about just what machine learning is good at, but like the results that you as a human look at, are you able to understand that?

Yes. On the 61st day, it started working better. But if you’re looking at two months of data, That one day, which is like one and a half percent, even if it was a fantastic day, you might still look at horrible data. And that’s the thing that we have to clean.

Tetsuo Konno: It’s important to keep in mind how data is being used for decision-making and bidding as well. Even if some data is uploaded 40 to 50 days after a click or conversion, it can still matter. However, recency is also a key factor when it comes to conversion values. Instead of striving for perfection, as you mentioned, the 80/20 rule is a solid approach. Staying consistent in your data updates will help you achieve the best results, both in terms of outcomes and actionability.

Frederick Vallaeys: Great. You’ve done a solid job covering the machine learning and predictive mechanisms. Now, let’s explore generative AI. Where does it fit in? So far, we’ve communicated the right conversion values to the system, allowing it to make better decisions. But what role does generative AI play in making our campaigns better?

Tetsuo Konno: Yeah, I think a lot of this is super relevant. But if it’s okay with you, I think we should look at the bigger picture. As marketers and PPC professionals, we often focus on what we can do within marketing to get better results with generative or predictive AI. But it’s an exciting time, not just for how we work, but for how consumers are experiencing search through generative AI.

When I think back to when I started, search was just a list of blue links on a desktop with yellow backgrounds. Over time, search has become so much more. We now have multiple entry points for how we search for information, and generative AI is playing a massive role in transforming the kinds of questions people ask and the answers they receive.

For example, it’s tough to convey what’s happening with something like a broken record player through just a text prompt. But with tools like Google Lens, you can film the problem and then ask, “Why is this happening?” In this multimodal context, AI is able to provide a clear, concise answer: something we couldn’t do as easily before.

Generative AI is revolutionizing both marketing and how consumers interact with search. One recent development we’ve seen is the AI overviews, which, as we discussed during our earnings call, have rolled out to over 1 billion people in more than 20 countries. What’s interesting is that people are more satisfied with these results. They’re also searching more, especially younger users who are highly engaged with AI overviews. We’ve found that the traffic generated from these results tends to be of a higher quality, with users spending more time on destination sites.

So, in this context, we’re seeing real shifts in consumer behavior, which is opening up new opportunities for advertisers to be more visible, relevant, and aligned with user intent.

Frederick Vallaeys: Yeah, that’s fascinating. And the way I see it is that, for over two decades, we’ve gotten used to search looking a certain way. As you mentioned, it was those blue links, slightly different colors, slightly different layouts of the page. Google kept improving the relevance over time, but the appearance itself remained fairly consistent. But now, it’s fundamentally different. As you said, you use your camera, narrate your question, and the generative AI picks up on that. It pulls together the best sources and summarizes them for you, and suddenly, you’re having a conversation. That draws you in, right? You find something new, and it’s very easy to ask follow-up questions that, in the past, you might’ve just walked away from.

Now, Microsoft shared some data in the past, so I’d love to know if Google said anything similar at the earnings call. Microsoft mentioned that the click-through rate on ads within the Copilot experience was higher because the generative system, with its memory element, fundamentally understands the consumer better and what they’re looking for. So it’s better at matching ads, resulting in higher click-through rates. Did Google say anything about click-through rates and the relevance of ads in this context?

Tetsuo Konno: No, not directly, and not with exact statistics, but as I mentioned, what we do see is that people are engaging more. They’re actually searching more, especially in the context of AI-driven overviews, which is interesting because the answers and queries are usually quite long and complex, already very to the point. What we’re seeing, though, is that people are spending more time on destination sites once they land there. This is, I think, an indication of the relevance that AI overviews are providing, and people seem more satisfied with the results as well, which is a good sign beyond just CTR.

Frederick Vallaeys: So, one of the points I’ve made about the future of landing pages is that they may be less useful. But what you just said is making me rethink that a bit. Here’s why I thought landing pages were on the way out: Generative AI offers such an engaging experience that people can become better informed through tools like ChatGPT or Google search’s generative features.

They get those answers right there. Instead of jumping to a landing page to figure out more, it’s often easier to continue the conversation with the chatbot and get further answers. By the time you land on a page, I thought, you’ve already figured out what you want, and you’re just looking for the buy button. So, I saw landing pages as less important.

But what you’re saying is that people are spending more time on those landing pages, so I’m curious: If the chat is helping them so much, why do they end up spending more time on the landing page?

Tetsuo Konno: Yeah, to be honest, I can’t speculate too much here without oversharing. But what I can say is that a lot of people would likely agree with your initial hypothesis, and it’s something we’re hearing from clients as well. As I mentioned earlier, it very much depends on the type of question you’re asking. If there’s commercial intent, as you pointed out, you still need to buy the product. However, the difference now is that instead of bouncing between 10 or 14 products, the product recommended by generative AI could be so much more relevant that the likelihood of making a purchase increases.

When you reach the landing page, as you said, there’s less research needed because the product feels like a better fit. You dive a little deeper and eventually end up purchasing from that advertiser.

Frederick Vallaeys: No, that makes a ton of sense. So what do advertisers do to be part of this generative experience?

As I understand it right now, you just have Google ads, you place your ads and Google will automatically decide when to show ads along with the generative experience. And that makes sense, right? But then from some other perspective, if I go to the generative AI and say, listen, I need a PPC management software.

It’s going to say some things about Optmyzr and other tools. And you want to make sure that’s correct. And now we’re talking about SEO a little bit, but it’s about content creation, like how do advertisers or companies think about this interplay of SEO

Tetsuo Konno: So, with PPC, I think the question now is: What do advertisers need to do to remain relevant and ensure they show up? As you mentioned, we don’t have a specific recommendation that dictates exactly what to do to appear in search results. However, there’s an interesting perspective that I’ve shared with colleagues, and it’s this: The things that were already important, providing good structural information to ad platforms, whether that’s structured text and product details through feeds, or providing high-quality images and videos are now just amplified and even more critical for search results.

This is still grounded in Google’s core search systems, but with the added capabilities of large language models (LLMs) layered on top. These models need to pull information, and the more structurally and readily available that information is, the better. Having a variety of high-quality visual content, alongside clear and concise text and feed information, is essential. In short, the best thing we can do to ensure relevance is to make sure all this information is crystal clear and easily accessible.

Frederick Vallaeys: Yeah. And that makes a ton of sense. And even from an SEO perspective, I know you’re not going to comment on this, but the AI is very good at summarization and finding the right pieces of data.

So I think the more content that someone produces as a business about what their product does, specific use cases, that just gives the AI more information to pull from and give correct factual answers. So talk to us a little bit then about like when you say feeds and images, right? We already do a lot of that stuff as advertisers.

Can Gen AI also help us be more creative with those things or get 200 percent feed coverage? How would advertisers use these things? And maybe talk a bit about Gemini if you can.

Tetsuo Konno: Yeah, definitely. In PPC, I think there’s a lot to gain from generative AI tools, whether it’s in the creation process or more broadly, like creating ads or better feeds. Honestly, I think as a community, we have a lot to gain here. Personally, when I used to write ad copy, I wasn’t particularly good at it. In fact, I hated it. I wasn’t that creative—I preferred data and analyzing the details to perfection. But at the end of the day, what matters is what people see, right? They see your ad, so it’s crucial to get it right.

I think you’ve discussed this with former podcast guests like Andrew Locke and Jill. A lot of the ads we see in PPC and the ER still follow a simple format like “buy product for price.” Generative AI can help us create better, more appealing ads, and it can do this at unprecedented speed and scale, along with constant iterations. One area where AI shines, especially with Gemini, is in the speed and quality of this process.

I personally don’t like to do feature comparisons between different AI models, especially since the space is evolving so rapidly that comparing models can sometimes be irrelevant. But when it comes to Gemini, from our perspective, we aim to make it easy to use. One way we do this is by embedding Gemini models directly into our ads products, like within Google Ads.

I know this might not solve every use case, especially for enterprise advertisers and agencies, but we are integrating these capabilities directly into the Google Ads system, like through automatically created assets and the recently introduced image generator for PMX and search.

The other way we apply Gemini is through more customized solutions. These aren’t always official Google products, but we have a lot of great solutions available on GitHub. Our G tech teams have developed templatized solutions that solve specific use cases. One example is Feed Gen, which uses the Gemini model and is trained on best practices for feed optimization. By providing just a few high-quality examples, you can create a much higher-quality feed.

So, think of Gemini as a powerful AI that we embed into many of our products, Google Ads being one of the most important. At the same time, we’ve also developed more applied, custom solutions that you can use for specific needs.

Frederick Vallaeys: Yeah, and that makes a lot of sense. With feed generation and telling generative AI the format the feed needs to come out in, that’s a capability a lot of people don’t think about deeply. But you can even apply this to landing pages. A landing page is often a template with certain fields filled out. That’s essentially a variation of markdown, for example, or it could be a JSON file that determines which pieces of text go in which sections of the page.

So, if you give Gemini the structure of that JSON or markdown and then tell it, “Here’s the product we’re trying to advertise,” it can generate the full structured file that you can directly upload to your website. This allows you to have a better landing page without manually editing it or going into the database to sort through the markdown.

It can really help you scale things if you start taking advantage of these nuanced capabilities of generative AI.

Tetsuo Konno: That’s a super interesting point because we’re talking a lot about ads, but in some of the clients I’ve worked with, it’s also being used for campaign ideation. Our YouTube teams have some great examples where Gemini was able to come up with creative and nuanced ideas for campaigns. For instance, it helped localize humor from the typical U.S. style to something more fitting for the Netherlands.

I also work with a retailer who has very high inventory turnover. They often receive just a product image and maybe one sentence describing what the product is about. I think generative AI can really help there, especially since most models by now, including Gemini, are multimodal. It can take the information from the image and turn it into well-written product descriptions, which is very powerful. Moreover, having that quality product information allows you to be better matched in search, for example.

So yeah, good points there—definitely think beyond just ads and also consider how to provide instructions to Gemini models, like through FeedGem, for instance.

Frederick Vallaeys: Correct. Then you get better ads, a higher quality score, which means you pay a lower CPC for those same clicks. So your ROAS goes up, and everybody’s happy.

Tetsuo Konno: Yeah, exactly. And with feed generation, it’s nice to use generative AI, but it can actually have a big impact on feed quality. We’ve seen this with quite a few advertisers. Improved feed quality leads to more impressions, better CTR, and lower CPC, as you said, which then results in better business outcomes, most importantly.

And I also want to plug this in: we have a nice solution called FeedX, available on GitHub. It helps with A/B testing for feeds, which has always been quite a challenging topic. I know there are some companies with solutions for it, but this is an open package you can use to try and test FeedX. So you can actually see that doing those generative AI optimizations results in better outcomes.

Frederick Vallaeys: Okay. I’ll put some of these in the show notes—exactly where on GitHub to find these cool Google projects. Thanks for sharing those.

Tetsuo Konno: And a disclaimer, we always say it’s not an official Google product, but it uses some of the technology and strengths of Gemini in a templatized way. This way, you don’t have to do everything yourself. It’s a templatized solution that you can use for specific use cases, like feed optimization or testing feeds. We also have solutions for generating image assets and linking them to the right ad group, or creating RSAs through AI, and those kinds of things.

Frederick Vallaeys: I’ve used generative AI quite a bit for writing scripts and for its multimodal capabilities. I’ll just go on the whiteboard, sketch out what I would do, and it generates the script. Especially with the latest GPT models, which have better reasoning, they’ve been really good for me in that regard. And I’m sure, since Google is behind ad scripts, Gemini must be pretty good at that as well.

Tetsuo Konno: Yeah. To be honest, I don’t have that much recent experience with writing scripts, but what I’m hearing from advertisers is that generative AI, in general, can obviously help a lot with things like coding. We also have something in our cloud environment called a code assistant. For Google Ads scripts, it can be very powerful as well. It really speeds up the time you can go to market, from ideation to the actual script.

Frederick Vallaeys: Yeah, for myself, I used to have an idea for a script and then I’d spend a couple of days writing it. Now, it’s usually like Friday afternoon at four o’clock when I think, “I feel like I should write one script this week,” and by five o’clock, generative AI has actually produced something that works. I put it on Gist on GitHub, and it’s out there. It’s a useful little thing that people can take advantage of for free.

But let’s shift gears a bit towards reporting and analysis with generative AI. What are you seeing in that space?

Tetsuo Konno: Yeah, I think that’s also a super interesting space. Leveraging generative AI to make better sense of the data we have is a game-changer. Something that we have in Looker as well with generative AI is particularly interesting because it creates a more conversational experience. By asking or prompting questions, you can actually get visualizations from datasets in the form of graphs or tables.

On top of that, you can ask it to explain the data to you. It almost becomes like a commentator or a narrator. You can say, “Give me this data,” and then you’ll see a graph. Then you ask, “Explain the trend here. What are the three key takeaways?”

Those are the kinds of things we have with Gemini in Looker, but we also have a Google Ads report generator, which is also based on Gemini. It can help you make sense of the data and even create slides or summaries. So, I think in that space, it’s definitely super powerful.

Frederick Vallaeys: Nice, I love it. And we have some Optmyzr Labs capabilities, one of which is a slide builder. It allows you to narrate your PPC performance and say, “Look at the last 30 days of data and tell me five things I could have done better” — five optimization suggestions, or whatever it may be. It’s funny because when you go to the generative AI and ask, “Tell me five positive things about the account,” you take those suggestions and talk to your client, saying, “Hey, we’ve done amazing work this week/month,” and they give you a thumbs up, “Keep going.”

Then, you turn around to the same AI and ask, “What are five things I could have done wrong in this account?” and it responds, “Here are five things you could fix to make your client even happier next month.” It’s really cool how it acts like that sideline player, that coach.

It’s also really great for training, I’ve found. If you want to get into statistics but don’t fully understand them, generative AI will spit out a chart, but you can actually say, “Explain this chart to me. How was the data analyzed? What were the look-back windows? What are some potential pitfalls I should watch out for?”

What’s so cool is that the education we typically get comes from reading web pages, watching YouTube videos, and everything is structured the way someone else thinks we should consume it. But with generative AI, you can jump back and forth and customize it to suit your preferred learning style. That’s what makes it so unique and powerful.

Tetsuo Konno: Yeah, and I think what’s also interesting in the reporting space is that I was tagging along with a colleague from the cloud team who did a Looker demo. By asking questions, it automatically generated reports and so on. But the fascinating part is that you no longer have to be good at coding to gain these insights.

When you open up the SQL code, you can see it’s being written in the background to produce the visuals from the database. Even though I work in this space quite a lot, I find that quite mind-boggling. It makes getting insights much more accessible for everyone in the company, as you’re no longer reliant on others to pull and create those data sets. As I mentioned, you don’t need to be a coder anymore. By simply asking a prompt, SQL code is generated in the background, making insights and data more accessible to a wider audience within a company.

So, I think that’s definitely a very powerful part of generative AI.

Frederick Vallaeys: And I think that’s a perfect segue into the idea that you can now do things yourself that in the past you would have had to ask others in the team to handle. What’s your take on the future of the PPC specialist?

How do we future-proof ourselves and ensure we continue to make a living? We may not necessarily need a job, but we’ll still need income. So, how do we maintain that?

Tetsuo Konno: Yeah, exactly. There are a few interesting angles here. First, a thought that might sound a bit cheesy, at least, I think the Dutch folks might call it that. But I still believe there’s truth in a statement we use in some of our decks: As a marketer, you’re not competing against AI, you’re competing against marketers who are using it effectively and making the most of it.

So the first point is, don’t fight it. Instead, ensure you’re getting really good at using both predictive and generative AI in your job. Learn how to work with it and continually improve in that area. This will help you better understand what works and what doesn’t. To be honest, the role of a true single-channel PPC specialist might not be enough anymore. With generative AI and all the tools available, there’s really no excuse not to broaden your skill set. AI can free up a lot of time for more strategic tasks.

I think the future of a specialist role is to be more well-rounded, a bit more T-shaped. There are still areas, at least today, that AI hasn’t fully mastered — particularly complex, multidisciplinary projects. For instance, data activation always requires alignment between the data team and PPC. How do we ensure that alignment? Or, how do we prove to the CFO that our investments are paying off? How can we demonstrate the incrementality of a certain channel or show the role of search within the total media mix and how that search data is fed into the model?

So, I think stepping back and becoming more strategic, more T-shaped, is definitely the way forward. I even asked Gemini, inspired by your book, where you talk about the roles of PPC doctor, pilot, and teacher, to come up with a fourth role based on what I’ve just mentioned. They suggested something like a chef, preacher, or orchestra conductor, someone who oversees multiple moving parts, connects them, and sets the direction for each channel. That’s how I would describe the future role of a PPC specialist.

Frederick Vallaeys: Nice. Yeah, I once wrote a post where I compared it to being a restaurateur. The analogy was that when you’re using chat assistants, you can ask them to “cook me a great dish,” and it’ll create something amazing, but you might hate it. That’s because the default is that you didn’t specify your dietary restrictions or the ingredients you like.

One interesting thing about these chat systems is that the more restrictions you put on them, the better they actually become at coming up with innovative, cool ideas. So, the trick is to give it more boundaries and tell it who you’re making this dish for. Once you do that, it’ll come up with something great. It’s about knowing how to prompt it and work with it to get the best results.

That also ties back to the localization example you mentioned earlier. One of my favorite examples is exactly what you said — it’s not just about translating an ad from English to Dutch. You need to make it funny to a Dutch person. A Dutch person is different from an American one. We watch different shows, we have different sensitivities.

I actually did this for an ad for a flower delivery service. I asked it to localize the ad for a Parisian audience, and it started introducing the word “élégance” — which is a common term used in Paris. The next day, I saw an ad for Air France, and it had the word “élégance” in it. At first, I thought GPT was just making that up, but it clearly understands that “élégance” is a local sensitivity that makes sense to incorporate into the messaging.

Tetsuo Konno: Yeah, exactly. Another way to think about it, and I believe this analogy has always applied to PPC, especially looking to the future, is that you have to be an “input master.” It’s about putting the right data into the system, having clear goals, and making sure the data is aligned. But in the generative AI space, as you mentioned, being an “input master” goes beyond that, it means having the right prompts and context for those models. So, you’re still an input master, but also an “output evaluator.” Everything in between can likely be handled by AI-powered media buying — the various campaign types we’ve discussed in Google Ads. But evaluating that output, putting it into context within the overall marketing mix, and proving incrementality are still incredibly valuable tasks.

As we move away from campaign management, the focus should shift to becoming an input master and an excellent output evaluator.

Frederick Vallaeys: Yeah, absolutely. It’s been fantastic having you on, Tetsuo, and thanks for sharing those amazing insights into generative AI and how to leverage it within the Google Ads framework. If people want to get in touch or learn more about the things you talked about, what should they do?

Tetsuo Konno: Yeah, so you can follow me on LinkedIn. I don’t have a nice URL here, but maybe you can add that later. But yeah, follow me on LinkedIn, and if you reach out, I will definitely get back to you. And we can also stay in touch through my Google email.

If you’re at events around Amsterdam, definitely reach out. I don’t think I’ll be in the U.S. anytime soon, but I’ll be at Friends of Search in March. So, please don’t hesitate to reach out if you have any questions.

Frederick Vallaeys: Great! And for everyone watching, we’ll include Tetsuo’s LinkedIn in the show notes, along with a bunch of other resources he mentioned. With that, we’ll wrap up another episode. I hope you enjoyed it! If you did, please give it a like and subscribe so you’ll be notified about future episodes. Thanks for watching PPC Town Hall. We’ll see you for the next one.