Podcast: Google navigates the AI advertising era (with Dan Taylor)

On this week’s episode of the podcast, I am joined by Dan Taylor, the Vice President of Global Ads at Google. We discuss the profound impact of generative AI on search behavior and the technological evolution required to support the next era of digital advertising. Among other things, we discuss:

• How the shift from keyword-based search to conversational intent will redefine the core mechanics of performance marketing attribution
• Whether the increased monetization of long-tail conversational queries can offset the potential traffic loss for traditional web publishers
• What role LLM-based ad ranking plays in reducing irrelevant placements and improving the overall efficiency of digital campaigns
• If the consolidation of advertising tools into automated systems like Performance Max represents the final end of manual control
• Why the emergence of universal commerce protocols might eventually allow AI agents to handle the entire checkout process autonomously
• How the return to marketing mix modeling and first-party data signals a broader renaissance in rigorous measurement techniques

Thanks to the sponsors of this week’s episode of the Mobile Dev Memo podcast:

• ⁠⁠⁠⁠INCRMNTAL⁠⁠⁠. True attribution measures incrementality, always on.
• Xsolla⁠. With the Xsolla Web Shop, you can create a direct storefront, cut fees down to as low as 5%, and keep players engaged with bundles, rewards, and analytics.
• ⁠Branch⁠. Branch is an AI-powered MMP, connecting every paid, owned, and organic touchpoint so growth teams can see exactly where to put their dollars to bring users in the door and keep them coming back

Interested in sponsoring the Mobile Dev Memo podcast? Contact Mobile Dev Memo advertising.

The Mobile Dev Memo podcast is available on:

• YouTube
• Apple Podcasts
• Spotify

Transcript

Eric Seufert: Welcome to the Mobile Dev Memo podcast. I am your host, Eric Seufert, and I am joined today by Dan Taylor. Dan, welcome to the podcast.

Dan Taylor: Great to be here, Eric.

ES: We have lots to talk about. Before we dive in, please introduce yourself to the audience.

DT: I am Dan Taylor, and I am a Vice President in our global advertising business. I have been with Google for about 20 years, always on the business and advertising side. Today, I focus on ads commercialization, focused on bringing our AI-powered advertising products to market, with a particular focus on measurement and our advertising platforms today.

ES: I often have senior people from the bigger platforms on the podcast who have been with the company for a long tenure. I am always curious to ask what the evolution of Google has been like over 20 years. That is a remarkable amount of time to spend with a company, but the internet has fundamentally changed in that time. I would be curious to hear what you have seen change in the internet landscape, because Google had a front-row seat to that.

DT: I joined Google from traditional broadcast media. I was brought in as part of a crop of advertising executives to convince large brands and agencies that search engine marketing was important. That was a moment in time.

About a year into Google, we acquired YouTube and wanted to convince brand advertisers that online video was going to be important. That was a moment where we were relatively early in the industry. Consumers were spending a lot of time on digital, but the advertising landscape was still firmly rooted in linear television and broadcast media. We were bringing people along on the consumer shift and the technology shifts.

Fast forward a few years to mobile. I think that supercharged a lot of people’s attention on how rapidly and quickly things were changing and how the tools and technology needed to adapt. No one knows that better than you in terms of how you got started with Mobile Dev Memo. Consumer time being spent necessitated that publishers, advertisers, and agencies adapt to those tools and technologies. That has been how we have been spending our time the last several years, following consumer trends, adapting our technologies, and getting advertisers and businesses to come along on that journey with us.

In terms of the overall shifts, it has been about how consumers are spending time, how marketers are investing their dollars, and measuring accordingly. Most interestingly and more recently, I have been thinking about this shift from traditional media not being able to measure well at all to digital where we could measure something, but we were largely measuring proxy metrics. We moved toward measuring something that actually matters, whether that is profit or conversions. Now, with a lot of dollars spent in digital, I need to understand which of those platforms and players are driving the most incremental return for my investment. It is moving more and more toward business outcomes and not just proxy metrics today.

ES: That is fascinating. When you survey your career at Google, you have the consumer adoption of the internet epoch, which seems quaint, but that was a dramatic shift in behavior at that time. Then the mobile epoch stands out as a distinct period. Now, I would imagine you see it as the AI epoch and how consumer behavior and advertising are aligning around that. Would that be an accurate characterization?

DT: I think so. I tend to look back to look forward, and I see a lot of parallels to these big consumer and technology shifts. There are a lot of parallels to the shift to consumers online, the shift to broadband and online video, the shift to mobile, and now today’s shift with AI. The shift with AI feels faster and bigger than the ones that came before it.

ES: I spoke at I/O three years ago. One thing I was struck by was a policy track I was on. This was in the early innings of this AI consumer adoption cycle. The speakers from Google’s side highlighted the application of LLMs across the consumer surface area within Google’s portfolio historically. I remember someone making a point that I had never really thought about: What do you think autocomplete is? That is what it is. We were applying an LLM, though it was not comparable to the scale of the models we use today. This technology is new with respect to a chatbot, but the attention is all you need was 2017. Being inside the company, it would have seemed a slower and more incremental evolution. Now the consumers have become aware of it, but you have been using it internally for some time.

DT: In some ways, and you are right, going back a really long way, thinking about how even when you went into Google.com many years ago and you would start to type a search query, you would get this autocomplete. A lot of that would be driven by what other people are searching for and making that easier for you. We saw smart reply in Gmail or autocorrect for misspellings. You are starting to see that in Workspace and things like that.

We have started to see some incremental changes over time. It was this moment where consumer attention started to realize generative AI can really be seen and felt. For probably a good decade, we had been using predictive and analytical AI in our advertising and consumer tools. When you look at GDPR and ATT and we talked about advertiser metrics and measurement and optimization, we had built this promise up of being able to have a really strong way to measure advertising, and then that got fragmented really quickly as you went from desktop to mobile to connected TVs to privacy and browser changes.

We had to build a lot of technology to project and estimate conversions or audience behaviors where we could not actually see it. That was all AI technology in the background. As soon as you started to bring that stuff to the foreground, consumers started to say this technology is real. It has been real for a while. To answer your question in an end-around way, I think the generative AI technology really just came to the foreground in the last couple of years in ways that consumers noticed.

ES: It also benefited from the rebrand from ML to AI.

DT: For sure. I think we were using the phrase automation and we realized no, it is AI.

ES: One thing the prolonged ATT waiting period taught me was how ill-equipped a lot of advertisers are to adopt these technologies. Maybe that has changed in five years, but I remember in that waiting period I was contacted by large advertisers needing help to adapt to these conversion values. They needed to build some sort of predictive mechanism for LTV against these conversion values. The very largest advertisers had trouble. Some could do it, and some could not. You did see this dependence on the biggest platforms because that is the only option for this happening, depending on the big platforms who have the tech, the expertise, the ML engineers, and the infrastructure to do this. These tools are available and have been democratized in a lot of ways by open-source packages like TensorFlow or PyTorch. But even then, it is hard. That is why the largest platforms accelerated, and PMax probably accelerated as a result of ATT. It shows that capability divergence between the largest platforms who have these massive armies of people and advertisers. That gap probably gets wider by the day.

DT: I would love to touch on that because it brings something to mind for me. There are actually two polarities that I have seen on that front. The first one is on the track that you mentioned, which is as there has been fragmentation in the identity and tracking space, we have built a lot of these AI-powered systems to help advertisers smooth out the bumps to help reach audiences at scale and deliver on their advertising goals.

The other thing that has happened, though, is that advertisers have really gotten smart about taking control of their own data assets, their own first-party data. They are getting their own tagging infrastructure and analytics set up in-house. They are starting to build their CRM system connectors. If you are a lead-gen advertiser, you really understand what kind of data you can send to the platforms you are working with to tell them which conversions you are sending them actually turned into a profitable sale versus which ones were useless to them in the end. Yes, there are these AI-powered tools that help them drive scale faster, but also they are learning that the data I send into those platforms to drive results for me actually matters a heck of a lot more than just turning it all over to the platforms to figure it out for me. It was an interesting polarity that both of those things happened at the same time.

ES: That is a very astute point. I have called this the measurement renaissance. All of a sudden MMMs are in vogue again. Back to the future. I had Carl Mela on the podcast a while back, who is a legend in the brand marketing space. We were talking about how funny it is that MMMs are at the forefront of measurement now. If you told someone that in 2018, they would say that is preposterous. But it forced the discipline of relying on holistic measurement and weaning off deterministic signals because you lost access to them. I do think in a lot of ways it enabled more rigorous measurement, which MMMs and incrementality testing provide. Google has led the way here for a while. There was a skunkworks project within Google that made lightweight MMMs before it was an official product. Just a couple of people thought people need to have this.

DT: That was a big investment for us in productizing that with Meridian last year. Not only open-sourcing that but working with third-party partners to put those models to work inside of their MMMs to make sure that our media and organic search priors and things like that are being baked in. It has been a really interesting back-to-the-future moment, but also critical in this AI-powered moment because the data that you bring in to steer the AI matters a ton and it is a differentiation for marketers to get it right.

ES: Q4 earnings revealed AI overviews queries doubled since launch. Talk to me about what is behind that. Are users adapting search habits to AI overviews? Are they writing queries that are more likely to be answerable by AI overviews, or has the AI overviews coverage increased?

DT: People’s expectations and behavior changed as we brought AI to search. They are seeing AI overviews and they realize they can ask new and different types of questions, things they would not have come to Google to ask before. They are longer, more complex, even visual. We talk about Google Lens and that is 25 billion searches a month today. That is leading to growth in overall queries, including commercial ones.

You no longer have to think about the right way to search on Google, you just ask the question. People are moving away from keywords to more conversational and intuitive experiences, and that is what is triggering more AI overviews and driving more usage of AI mode as part of that expansion. Along with that is coming Gemini’s better understanding of these longer and more complex queries. It is more about how users’ behavior is changing as opposed to how we are tweaking the experiences.

ES: Historically search had been a skill because you needed to know how to construct the query to be concise, to coax it out of the search engine. You had to know how to structure it such that you were putting the focus on the keyword that you cared about, but there was enough detail there to parse out different variations of the data. Now you get a more forgiving interface when it is conversational. Is that right?

DT: That is absolutely right. I was on a car drive last night and I used AI mode and just had a conversation. It was just easy. It turns out when you make search easier, people search more. You just get more shots on goal.

ES: Can you give any updates on AI overviews ads monetization relative to legacy search? The last data point was that AI overviews is monetizing at parity. Is that still the case?

DT: That is right. AI overviews are still monetizing at parity, which is a strong base. What has been really interesting on the ad side there with customers is how AI overviews is becoming a more powerful engine for brand discovery, creating new opportunities to get in front of customers earlier in the journey. It is part of that expansionary moment that you hear us talking about. AI overviews and AI mode are matching ads not just to what the consumer is searching for, but also to the context of the answer.

For example, when an AI overview is an AI-generated snapshot in the results, an ad can appear not just based on what the person searched for, but what shows up in the AI overview. I travel a lot for business and my wife is now sometimes coming with me now that our son is off at college. We started talking about bringing our cat, as the cat gets lonely. We search how to bring a cat on a flight. That type of search triggers an AI overview and our search models break down that question into smaller subtopics. Liz Reid calls that the query fan-out technique. The AI overview shows airline policies, vaccine info, how to keep them calm, and that you need a compliant pet carrier. That all adds up to that broader question. In the old days of search engine marketing, most companies are not going to bid on how do I bring a cat on a flight. But that pet carrier information in the AI overview triggers a new ad opportunity for companies like Petco, Walmart, or a company called Cat-in-the-Bag that sell pet carriers. Now that is a new ad opportunity that did not exist before for advertisers. It creates new monetization opportunities.

ES: You have the less distilled query from a keyword perspective, and it is almost like you are parsing out the relevant keywords to then have bids be submitted against. Is that right?

DT: The piece that was a little underappreciated by me at the beginning and many of our advertisers is that not only is it hard for me to predict what people are searching for and my keyword strategy is not as nimble to these new ways that people are searching, but I also have these opportunities with AI overviews where what is in the response can be a commercial opportunity that would not be immediate. It brings people down the funnel as part of their discovery journey, and that has been part of the expansionary opportunity.

ES: AI mode queries were revealed to be 3x longer than legacy search. That makes sense as a conversational interface. How does that impact ad ranking? There is a lot of additional context there. Historically I needed to know how to search such that I was making a dense nexus of keyword information that Google could parse out to rank the relevant links. Now with AI mode, I am giving you a lot more stuff. How is this contributing to ad ranking and giving more information to ranking those ads?

DT: There are two pieces to that. First, these longer, more conversational queries give us more to work with. There is more context to better understand the commercial intent and a better signal for our ad systems to work with. Historically, serving ads on these longer and more complex searches was challenging. Gemini has dramatically improved our ability to better understand those longer and more conversational queries, particularly in non-English language where we have seen a huge improvement. That has helped us capture interest in new forms of search.

The other half is our ability to predict which ads are going to perform best, which gets into the ad ranking question. For every ad, we generate a prediction of how well that ad is going to perform against a given query. Is the user going to click on it? Are they going to convert? In addition to better understanding the intent behind someone’s query, we also have dramatically improved predicting how well an ad is going to perform against that query. We have been making these improvements to search query understanding at a rate of about one launch per month for the last two years. This has led to a 40 percent reduction in irrelevant ads. For me, I was impressed that our ad system for 25 years has been working quite well, but it turns out with Gemini, we found a whole bunch of new headroom on ads quality.

ES: LLM-based ranking is my favorite topic at the moment. It is fascinating because this is a domain that has had the benefit of 20 years of the best data scientists and ML researchers. You are seeing these significant improvements on that baseline. You just get so much more data to bring to bear in determining if the user will click this and if it matches what they want. You get this semantic understanding to apply to whatever the input is. It is very shocking to me how big the improvements are here.

DT: It turns out that prediction is the perfect problem to put AI against.

ES: There was an infamous statistic that Google shared that 80 percent of Google search queries were not monetizable. There was just no commercial intent. How does that change in the AI overviews and AI mode interaction models? Anything can really be commercially tenable because you can parse out a lot more information in this conversational format.

DT: I do not think we have updated that stat, so I do not know if it is current, but our core philosophy remains the same. We show ads when there is commercial intent. What has changed is our ability to spot it. It is true that the vast majority of searches never involve an ad because ads are meant to be helpful to your search. If you are looking for today’s weather, the most helpful thing is going to be a direct answer, probably not an ad. We are seeing more than 5 trillion searches on Google a year, but the overall number of queries, including commercial ones, are going up largely because of our investments in AI like AI overviews.

As we think about these new experiences in AI mode, for example, we are just testing ads in the U.S. today, mostly focused on getting the user experience right. AI mode is something between a chatbot and a traditional search engine. It is a conversational experience but grounded in the full information of the web. In our tests, what is really interesting to touch on here is we are being thoughtful about where and how an ad might show in a person’s journey. This isn’t enter a query, get a result, move on. It is about when ads are useful and relevant, they are appreciated and do not disturb the experience. If it is a conversation and a multi-query journey or session, putting an ad too early is not going to be a good experience.

About a year ago I got into running. I am ready to move to 10K, but I do not run that far on a regular basis, so I am going to use AI mode for tips on how to train for longer distances. What is a good heart rate, variable weather this time of year on the East Coast, nutrition, things like that. It is a longer and more complex query, and I will probably do a couple of follow-up questions on informational needs. If you throw me an ad for new shoes right up front, you have lost my trust and it is not a great experience. But if I dig in and ask follow-up questions, I could learn that collagen could help my joints. There are opportunities where ads might be welcomed and the right answer to my need. We are really figuring out where in the monetization or where in the experience monetization makes the most sense in things like AI mode. That is what we are looking at right now.

ES: It sounds like you are talking about stacking the conversation based on the level of insight delivered. If I am asking high-level conceptual questions, an ad would not make sense. But as you deliver more and more information to the person such that now they are more primed to buy, then that is the right time to show an ad. Are you thinking about this within a single conversation or across multiple conversations over time?

DT: I break that into a few things. With AI overviews, that is an extension of more or less traditional search. It is a representation of the traditional search result that has more information. That is an expansionary opportunity and we are finding new opportunities to monetize there by understanding the longer queries better and the context of the page. In AI mode, we are finding opportunities to monetize in those longer, more conversational back-and-forths and we are trying to find the right user experience and the right ad formats to introduce there. I am mainly talking about it at this moment within a single conversation, but you could see how that might be something you would introduce over time. One of the things that consumers expect from their AI-powered experiences is context. They want to make sure that they understand experiences over time.

ES: I wrote a multi-part series called Google’s Gambit. The idea of Google’s Gambit was that there is this absorption of engagement that used to be directed externally as a result of the links into AI overviews and AI mode. I was making an observation of incentives that makes sense, especially given the consumer shift into AI interaction. What do you think are the long-term consequences of this on Google’s advertising business? Where does this take search?

DT: We do see this as an expansionary moment overall. AI is helping people ask new questions and enable businesses to grow in new ways, but we also see it creating new opportunities for creators on the web. This is a big technology shift that reminds me of that pivot to mobile, changing how consumers are engaging with content and platforms, including our own. As we are enabling more and different types of questions with search and AI, our ads focus is creating opportunities for businesses, more relevant ads, but also providing improved tools for content creation, whether that is Asset Studio on ads or all the tools that we are giving to YouTube content creators as well.

As consumer behavior and time spent on these new experiences evolves, we will experiment with formats and with measurement to adapt to those business needs. I also see AI enabling greater discovery on the web, helping people go deeper in their research, discover sites, and new brands and authoritative sources they might not have otherwise found. That is one of the things I personally have enjoyed and one of the things we are focused on in the Google search experience in particular. It is grounded in the information around the web. One of the things that is really interesting about AI overviews is it very much focuses on here are the authoritative sources around the web, here are click-to-link and learn more. Perhaps more than any other company, we are committed to making sure that we are sending quality traffic out to the web. I think that remains central to our approach.

ES: Can you talk to me about direct offers? It launched a few months ago. Tell me how it is going.

DT: We announced it in January at NRF. It is a good example of how we are looking to reinvent ads for the new era of search. It was a new type of ad format that we are piloting built specifically for AI mode. It introduces an offer in the AI mode experience based on where the user is in the journey. It talks a little about how where someone is in the journey is important as opposed to just showing them an ad immediately.

I recently got a new laptop that is bigger than my old one, so I need a new bag. I used AI mode to research stylish neoprene backpacks suitable for men in business settings that are under 200 bucks and can fit a 14-inch laptop. I asked Google a couple of refining questions about how many pockets it has, and then Google elevates the most relevant products to meet my needs. Often you are only ready to buy if you are getting a great deal. With direct offers, relevant retailers give a special discount like 20 percent off or free shipping, which is what I am looking for. A sponsored offer can be helpful at just the right moment, helping you get better value while assisting the retailer in closing the sale.

It is still early, but we are seeing some good product-market fit and really strong interest from advertisers. We are testing with Petco, ELF Cosmetics, Samsonite, Rugs USA, Chewy, L’Oreal, and Shopify merchants. We will have more to share in the coming weeks. It is a good example of the types of things that we are looking to experiment with where in a traditional search engine marketing campaign, you would just show that offer to everyone, but in an AI mode experience, we want to be able to understand where someone is in their journey and find that moment where someone is ready to buy and this could be the thing that gets them over the edge.

ES: Also announced was UCP, Universal Commerce Protocol. Give us the same rundown. How is that being received and how is it working?

DT: In a high-level, agentic or AI-assisted commerce is starting to become a reality. We have had a lot of interest and response to that Universal Commerce Protocol. The reason for that is it is a common language to better enable businesses to connect with AI agents as part of the shopping journey. We have been focused on making sure that we get the building blocks in place to support the shopping journey in this new era from discovery to checkout. We built this in partnership with retailers, but it is an open-source protocol so anyone can use that.

We are using it for new shopping experiences on Google. We started by rolling out a new UCP-powered checkout. In the U.S., shoppers can buy items from companies like Etsy and from Wayfair right in AI mode in search and also in the Gemini app without ever leaving the conversation. At ShopTalk in March, we announced new capabilities like adding multiple items to the cart, catalog features, and supporting loyalty programs to make shopping more connected across the web. Companies like CommerceIQ, Salesforce, and Stripe also announced they are going to implement UCP on their platforms in the new future. The industry is still in pretty early stages, but they are flocking to this notion of commerce protocols for the technology to scale so that buying agents and retail agents can talk to each other in a way that is frictionless and secure and that retailers can still own that transaction and that relationship with the end user.

ES: One of the questions I had was what is the tie-in with ads? Is this specifically for ad interactions or is this for anything?

DT: There is not a tie-in with ads directly. It is just really important for us to make sure that we enable retailers and consumers to have a frictionless experience. Consumers love the fun parts of shopping, like being inspired by a new brand or virtual try-on and things like that. They do not love the friction-filled parts like finding it in this size or finding it at this price or filling out my CVV code and remembering my loyalty program. These sorts of protocols take the friction out of the shopping and that is the piece that we are focused on.

ES: I get that it is maybe uncomfortable for advertisers, but the reality is you look at the investments being made by these huge platforms and you cannot match them. You do not have this kind of talent internally. It is not possible. You are getting access to firepower that you would not have access to otherwise. If the outcomes meet your expectations, which is the entire point of these systems because that is what you provide it with, there is nothing really to complain about.

DT: Twelve of my 20 years at Google have been in the product go-to-market side of the house and I spent a lot of it on audiences and inventory. The last few years I have been really focused on data and measurement because in an AI-powered world, if that is steering a lot of the decision-making on an individual ad placement, where the leverage and where the fulcrum is is what outcome you are telling it to optimize for. Where I see most marketing organizations pivoting their energy these days is there. I think that is the right move.

ES: That whole discipline is signal engineering, recognizing that the onus is on the advertiser to determine how good this user is.

DT: AI is only as good as the fuel you give it.

ES: I am cognizant of the time. What do people most commonly get wrong about Google’s advertising business? What is the most common misconception you could disabuse somebody of?

DT: A common misperception is that search is purely a lower-funnel tool. Google Search does not just close the loop for brands, it often opens the loop. The reality is that search is a massive engine for brand discovery. Over 70 percent of shoppers come to Google Search open to trying new brands or products, and that is actually expanding with AI.

YouTube offers incredible opportunities for brands to tap into moments of discovery and engaging with trusted creators. The connection between the two is also interesting. People come to Google for everything from quick questions to high-stakes decisions and engaging with trusted creators. We are helping bring more of these insights to advertisers with tools like attributed branded searches where we can report to advertisers when consumers search for a brand after they see a video ad on YouTube and ways to engage differently with them based on that. We are really focused in on helping advertisers understand how they can capture opportunities with consumers when they are earlier in their decision journey because there is a tremendous amount of opportunity right within Google Search. It is not a place where you just capture demand, it is where discovery starts and decisions are made.

ES: Dan Taylor, I appreciate your time. Thank you very much for chatting with me and for making this happen.

DT: Eric, it was a pleasure. Thanks so much.