Google's Robby Stein on AI Mode reaching 200 countries and why great SEO still matters in the AI era

from 10 to 10,000 people. interview some timeless bunch of great interviews and uh yeah, you should go pick it up. I need to I need to relisten to it. Well, without further ado, we have Robbie Stein from Google in the reream waiting room. Welcome to the show, Robbie. How you doing? Welcome. Good to see you. Hey. Hi.

Can you guys hear? We can hear you. We can hear you. Um, I I know people are going to be confused, so why don't we just get out in front of it and explain your role at Google relative to Gemini, relative to AI. Uh, take us on a little, uh, Game of Thrones HBO intro of the map of Google and where you fit in. Sure. Yeah.

So, uh, I work on the Google search team. So, it's the search bar. You type things in. Yep. Yeah. I was going to ask, could you could you What is Google? Yeah, it's a thing. You could type a question in. Sometimes you could, uh, use a camera. I don't know if you've used Google Lens. You can also Google that way too.

There's a there's this app. It's also called Google. All that kind of stuff is is where I focus my time consistent. And the don't forget about the I'm feeling lucky button. It's still on there, right? Yeah. Yes. So So, so how do you think about integration with the Deep Mind team, integration with the Gemini team?

Um how do you think about bringing AI to search? Yeah. So we work very closely with the Google DeepMind team, Demis, Corey, that whole group. And the way you think about it is we want to have frontier models right in search.

So you can really ask anything and have the the ability to use all of Search's incredible knowledge, real-time information systems and context of the web to help give people this incredible information. And that's really where the two come together. Yeah.

there there was this uh question this like you know chattering class being like oh well Google's uh you know it's such a different paradigm chat versus a search 10 blue links but have you drawn on the fact that uh Google's search started with two modalities the I'm feeling lucky button was a different way to interact with Google search uh the 10 blue links was one way uh are you how are you thinking about how you like kind of level up or educate the consumer consumer to use the all the different tools since now there's not just you know search results and and and AI overviews but there's so many different things how do you think about ramping those up yeah so there's always been many ways to use search actually Google lens is a good example you can take a picture of something you can ask what is going on with this plant that seems to be dying and you can get information from that right from the camera and that happens in the app so there's always been different ways that you could access and tap into the search knowledge base um but I think increasingly it feels like you want to go to search ask whatever you have in mind, you only want to think about where you're going to ask that question.

And then if you have something where AI is really helpful, you get this little AI preview that's starting to show up now around AI overviews. And if you click into that, you're in this AI driven experience.

And we've now through a new new project we launched called AI mode allow you to follow up, go deeper, and have a full generative end-to-end chat like experience right within Google search. Yeah. Uh walk me through the international expansion. I'd love to know if legal or engineering is more rate limiting there.

It feels like it's it's got to be incredibly complex to uh check all the boxes when you want to expand. You've obviously expand. So, what's the secret to uh taking over the entire world so quickly?

Well, our newest our newest product um AI mode, which is the thing that lets you really ask anything and within Google search using frontier state-of-the-art models. you know, we we launched in the US and then quickly in India, a couple countries, UK, um, over the summer, I think May, June, July timeline.

And then just a couple months later, now we're at over 200 countries, 40 languages. We moved really fast, uh, to ship that. And so, we're now basically everywhere except for a couple countries. And there's certainly a bunch of considerations, policy, uh, and infrastructure. Honestly, they both affect the timeline.

Oh, yeah. Because there's actual inference going on. So you need local compute basically more than there's also there's there's we want to make sure do quality checks because in every language you have different permutations or you have multilingual with people moving between English and other languages.

Just want to make sure everything is doing what you expect it to do. You know this is also a multi-turn conversational experience. So when you evaluate it you want to make sure that it's dialed before you go.

Um and so each of those just takes time because I know I know everyone's always frustrated like why can't I use it today? It's the main thing I see on X. Um but hopefully now almost everyone can. How are GPUs allocated at Google? You have or TPUs. You have TPUs in the cloud. You're selling them now.

Obviously, DeepMind wants for research. Gemin equally, but I imagine that, you know, is is there a spreadsheet like what's the process to actually uh to figure out where to allocate GPUs or TPUs? GPU. I mean, there's a there's a there's an allocation process. I mean, everyone needs TPUs.

It's just comput is by far the most important thing in driving um infrastructure right now and there's lots of needs but they're they're important things we want search the search is one of the largest ways people interact with AI so that's like a it's very important one obviously we're doing frontier modeling work we've got a bunch of things happening in cloud so that that's there's a process by which you do those funny I did an event and a Google team actually gave me like a old Ironwood like gen prior gen single TPU in this little case and I was joking if somehow I could like crack it open and like refurb it and like get it up and running.

I can somehow like improve improve my standing in the on the team. But uh I don't think it's going to happen. How how often do you guys come back to Google's mission when when thinking about product decisions? We've talked about this on the show before.

There's always, you know, you I'm sure you wake up every morning, there's a new headline. What is Google doing in a uh in AI? Why haven't they released this faster? Etc. , etc. But I, you know, we were never uh we never really stressed about it too much because when you look back, I'm I have a AI overview here.

I said, "What is Google's mission? " My AI overview says, "Google's official mission is to organize the world's information and make it universally accessible and useful and it feels like LLMs are just so aligned. It's the perfect technology to to carry out the mission. And so it's like relax everyone.

I think this is uh this is like a natural evolution for the company. Yeah. I mean we talk a lot internally actually about how the mission has never felt more relevant and it is really incredible to work somewhere. I worked at Google in 2007 for a while did some startups and you worked in some other companies too.

I'm back now and it's it feels that same level of entrepreneurialism in 2007 where it's like you're building all these new things for the first time except you're kind of building them again because AI allows you to say what does search look like if you could really ask literally any question and have created an AI that's the most knowledgeable AI out there that could understand all of Google's information the context of the web and you could talk to it and by the way you could talk to it live like we just announced search live that's available so if you're driving you could just talk to Google literally like on in your car you could take a picture and have this multimodal conversation back and forth now.

Like this is all stuff that talked about a long time ago but was limited because of technology. So that is incredibly exciting. It's it's very motivating and I do feel like is one of the reasons people on the team are fired up.

How do you think I mean it it's it's so interesting that uh as LLM's boomed AI search overviews uh made a ton of sense were baked in adopted loved uh but now we're already in the next phase which feels like agentic purchasing agentic checkout.

uh how do you think evolving how do you think about evolving the product even further to go from knowledge retrieval to taking action? Yeah. Um we have a bunch actually active there and we recently launched in the US an agentic experience where you can book um restaurants um and local services through Agentic.

So it's actually really neat like you through AI mode now um you can just have a conversation. Hey, what's a good date night place? It'll do all the thing. It'll tap into the knowledge of Google. It'll look up Google places. It'll do research. It'll do whatever.

But then if you want to start a task, it'll also go bring back availability across talk and opent like right in the experience and you can just book it and it's awesome. I've been using that. It's just a small example of what's possible. You know, people come to Google not just for information but to get things done.

It just materializes itself as a query. But you're not like I don't like want to just know restaurant reservation availability and like oh sweet. I'm glad to know there's a table available. I'm going to go back to my day. Like you're trying to do something.

We think about that a lot and that's just the you know the surface that there's so much people are trying to do but they're just kind of getting started with those journeys on search. What could we do to really help you? Shopping is a big one.

Um so we have an we just launched a new visual uh way to do AI which like is one of the first ways where AI can be helpful with inspirational tasks. So now in AI I mode if you ask to design a bedroom or look up landscape lighting it actually finds you beautiful inspirational imagery and products in a grid.

You could click on it. You can imagine getting much more help finalizing those purchases, doing, you know, being reminded of price changes and it's much more of this interactive version of search that can do things for you and you can you can really connect with versus just a pureformational experience.

Uh right now my my mental model for is like google.

com the search bar then AI mode almost as a vertical product like flights like uh images uh like shopping where it's sort of a portal or subproduct that I get uh you know I I go down a funnel and I wind up in um do you see that holding or do you think that AI mode acts as a wrapper on top of all the different sub search products.

Yeah, I think what's going to happen is you have this AI mode which is going to hopefully be this most knowledgeable AI possible. It knows everything in Google, billions of products, million hundreds of millions of locations, all of the web and it has access to all of it.

Knows how to use Google as a tool and it's super powerful, but not necessarily the best thing for all things. Like if you just need a specific phone number, you probably get that in like 50 milliseconds right at the top of the page. You just want to know a sports score.

what's what's literally the sports score right now just works you just you know you put it in Google or if you're just looking up a musician's name for the first time you're trying to get a browsier experience like you actually want to kind of see images you want to see what's going on on what are people saying on X which shows up in the search page so I think what happens is you have this incredibly knowledgeable system that we feel like is designed for more complex tasks more of this like how do I do this what's my advice for this I'm doing this trip I need this restaurant I'm trying to buy some jeans how do I get started with that and that if you have knowledge baked into that that's really powerful but then people need how do you how do you bring that to people and there's basically two ways one is you just search and through AI overviews we will show AI where we think it's useful and for many queries it's not useful actually which is why it doesn't show up the system learns that um but for these longer queries where you have a specific question typically shows up there um then the other way is for power users we feel like they kind of have this mental model of like oh like this I'm doing this planning thing or like oh I'm like I'm really curious to know like um what a stock price difference between these three stocks are over some period of time.

You wish you could just type that in natural language and have the thing generate a chart and look at use Google Finance as a tool which it will do and for that you can go right to AI mode. So if you can do that through the mode you can do it on mobile through those direct um kind of buttons to go right to AI mode.

You can go through Chrome now we announced like a way to just type and go right to the eye mode in Chrome and um you could also just go to google. comai now which is kind of fun. What anomalies are you seeing? Uh have you seen any of these screenshots of uh of Google trend data?

People are are posting, you know, people forever have been posting screenshots of effectively search data and uh using them to infer what's happening in in the world, right? And so over the past few months, we've seen people posting uh search queries like help with my mortgage.

and it's just like a crazy uh ramp uh up and to the right.

Uh my uh my intuition is that it's potentially uh a bunch of different keywords are seeing these crazy uh ramps because there's potentially agents sort of like leveraging Google search to uh to drive a higher volume of searches on potentially a search that's happening maybe in an LLM. Have you seen any anomalies there?

or is that something you're uh aware of? Um haven't personally seen any anomalies there. I I don't I haven't really heard of that before. Um so wouldn't look I wouldn't buy too much into that. Um we do have protections for those kinds of things.

Um I think in general what we're seeing is people who are using Google very differently and at a at a very fast growth. So people are asking very specific questions of Google. They're they're using Google Lens and asking multimodal questions. They're asking follow-ups.

I mean, those are kinds of things that we're just seeing we're seeing in such a broad way. Um, I think that's the thing that we mo mostly focus on. Um, but yeah, I can't speak more to some of the trends things that you're mentioning. Yeah. How do you think about uh advice for brands?

Uh, I've I've run brands and uh, you know, grinded my way up the SEO rankings. I never did a ton of uh, optimization.

Mo most of my strategy was just uh try and make something eventually that people talk about and it gets written about and those websites have you know ranking power and then you kind of rise to the top of that keyword. Is any of that changing in terms of uh in the AI shift?

What advice are you giving to uh brands that want to perform on Google these days organically? Yeah. But what's really interesting is I think the core Google search ranking is more relevant than ever because it turns out that one of the best things that every AI model does now is they kind of like search the web.

It's like I don't know like what are you trying to do? Oh, I don't know. I'm trying to figure out if like I should go to this hotel. All right. Well, like parametric knowledge. Do you like know the rates of those hotels? Like can't okay I'm going to search the web, right?

And and even AI mode is is special in that we're Google and so it uses Google really effectively and it creates these query fan outs. It does dozens of queries but it effectively is googling stuff, right? And everyone's googling stuff all the time.

So for a given question, what are the things that show up as highly relevant to a given question, those end up getting absorbed into the context window and have a high probability of being displayed to the user.

And so if you are just thinking how do I build great original content that's trusted and author and and authoritative for specific kind of query. Turns out that's still going to likely be very valuable. Um and you can go read the Google guidelines on content and um human raider guidelines.

I think they're super interesting. There's lots of detail in there around how Google evaluates trustful high quality content and scoring systems that are used. Turns out that's a really good investment because every other system is is kind of proxying for the info that's most useful for a question. Yeah.

Um so that's my main advice is to kind of like dig in more on understanding how those systems work because it's largely going to apply later. Yeah. And then the second one is what are people using AI for? Those are the growing kind of needs. Oh, sure.

And there's a disproportionate amount of use case in different types of domains now with AI because it allows complex needs and it allows for things like advice. It allows for things like really nuance troubleshooting with stuff. It allows for these emotional needs to be satisfied differently. Mental health. Y.

So those are areas that are probably growing markets of use cases. And I would be I would be a student of those as well. What about Sorry, can I have one more on that? So uh what about uh content that was previously buried to Google that is now available uh for search rankings because of artificial intelligence?

I'm thinking about like literally this show it will be a three-hour video on YouTube that you know a a buried mention previously probably wouldn't be perfectly translated indexed etc etc but uh I would imagine that the the strength of like video content and audio content gets relatively better over time in the AI era.

I mean honestly you guys have been doing this for probably a decade but uh take me through a little bit of that. Is that is that a reasonable thesis? I actually we are it is a reasonable thesis.

We have been seeing an increased diversity of the kinds of pages and sites that show up within AI because people are asking these nuanced questions. A lot of times there isn't even like a single web page that has this information. Yep.

But if someone I don't know sometime in the future this this uh video gets segmented and scanned and understood by Google and someone asks for like I don't know uh an interesting conversation on Google's AI journey recreating search the AI era. Oh sure.

Maybe it's like here's a cool conversation you could check out and this this like clip will show up now in a way that would be very difficult for that to happen unless you search for like TVPN like interview of whatever which was like very specific keyword based on or or we'd have to go manually create a transcript and then get that into like the text SEO world.

So yeah, very different uh era. Sorry. uh what are what are the plans around giving uh if there are any giving publishers control over whether or not their content is used in uh various AI functionality. Yeah, we have a bunch of publisher controls we we have.

So there's an over there's an overall opt out and training side, you know, on search you can opt out of crawling, you can opt out of this other thing called snippets where you can you can show up in Google but you won't show up in these rich these rich experiences if you want.

Um and so there's there's a bunch of things that you know publishers can do and um can look at there but you know I think ultimately the belief I have is that to the prior conversation you know AI should be this massive um discovery engine over time because if you think about it these complex needs we're getting growth like we're we're seeing 10% growth for instance in large markets like like India and uh the US for these really specific questions which at Google scale is like enormous enormous number um where you have if you have a really specific question people are doing that more and more and more and more because you can get AI to go deeper multimodal right you're asking taking a photo of something and you want to shop it seeing 70% year-over-year increases in those kinds of questions and these are billions and billions of queries like so this is these are huge numbers wait how many queries billions are billions billions so good the sneaky It's always fun.

Sorry. No, that was totally worth it. I'm glad you did that. Thank you. Um, so like these are big numbers and each of those generate AI experiences with links to go deeper on stuff. Every single one of them.

And so theoretically you should have this like unique opportunity to say, "Oh, did you take a picture of your bookshelf? Go, what book should I read? " Oh, well, here are cool reviews of other books that are like things you've liked. It's like, I could never Google that before.

And so theoretically, you should have because it's such an expansionary moment, we're seeing that this is an expansion more than anything. Like the old like the way people, they're googling and they're using it in all of these new ways. Um, and so hopefully over the long term that's that produces growth.

I have a lot more questions, but we'll have to follow up time. Yeah. Yeah. We could go way deeper here. I would love to even even just dig into how TBPN shows up on Google. It's such a fascinating content because we create so much content all over the web. Uh but thank you so much for taking the time to hop on the show.

This is fantastic. Uh and congratulations on the progress. We'll talk to you soon. Cheers. Have a great rest of your day. Uh before we bring in our next guest, let me tell you about Finn. ai. The number