Google’s head of search on the company’s AI transition

Reed Albergotti: When it comes to the existential question around the future of search, which still makes up the majority of the revenue for the company, how do you navigate that? Does Google need to disrupt itself at some point and say, ‘Okay, we’re in the AI search era now’?

Liz Reid: We are in the AI search era, and have been for a little bit. At some level, Google has been doing AI in search for a while now. We did BERT, we did MUM. Now, we brought it more to the forefront with things like AI Overviews. AI mode is giving that opportunity. At the high level, people sometimes view it as: it’s AI or search, or it’s AI or the web. We don’t see it as that split. I see it as AI enables search to do more of the things it always wanted to do. It always wanted to organize all the world’s information, but information was fragmented in ways that were difficult to join.

There’s just really this opportunity to answer many more of the questions that people have in their head. If you make it easier to ask questions, people just ask more questions.

We really think AI is a moment where the space of people asking their questions, the space of people getting help with their information needs, is really growing, and also a really exciting opportunity to help people with more of their needs, to really provide a level of reasoning. We’ll talk about going from information to intelligence. If you really want the information to be useful, it can’t just be information. It has to actually be smart, and we think this is an era to do so.

Some people say, ‘I’m off Google, I’m on Perplexity,’ or some other AI, ChatGPT, or whatever. Is that just like a Silicon Valley bubble? And how do you stop that, how do you keep them on Google?

We’ve always had a bunch of competition, that’s been true for years, and I think it’ll continue to be. What we really try to focus on is helping users with their needs and on trying to be the best place to get information. And what we’ve seen with things like AI Overviews is that it’s growing meaningful traffic for the types of queries that you think of as AI Overviews. We’re seeing a 10% increase in those types of queries in the US and in India. If you launch a sports score feature, people ask for a lot more sports scores. They don’t necessarily ask a lot more weather queries. We’re really seeing that ability to blend the AI and search together, the ability to get started and then to dig deeper, is really resonating with people.

And you should ask longer queries because you want to ask more content, not because you feel obligated to put in a bunch of filler words. We shouldn’t force you to type more words and spend more time putting in your query.

On the other hand, if you have a query that has five parts, go for it. Put the five parts in, and don’t feel like you have to break it down. We always think about this question: How do you make search truly effortless? What is the first thing that pops to your mind? If the thing that pops into your mind is a keyword, great. If it’s a long, rambling thing, or very detailed… you should just ask that, and we should organize the information in the way that’s useful to you. Instead of you having to do the work to come to the information, we should bring the information to you. We should transform it in a way that you can understand, you can consume, you can comprehend.

Google did organize the internet, the world’s information, and a lot of these AI search engines are not really doing that themselves. They’re piggybacking off Google. In a way, you served it up to people. I don’t know if maybe you’re putting a stop to that now. And it’s not so much organizing information, it’s also analyzing the information, or processing, right?

What we’ll continue to see with AI is that that will get more and more personalized, not just in helping you with recommendations, but in adapting to the way that you learn? Are you someone who’s going to learn about a topic best with a 30-minute audio because you want to go in depth, but you only have time [to listen]? I have three kids and a job. There’s a lot of stuff I get to learn by podcast, but not by reading because I don’t have that time, but I have a commute. [We’re] really getting to that point where it’s reasoned, it’s tailored, and it’s easy for you to work with. It’s just, frankly, really exciting.

What we are announcing is Search Live, which is bringing the technology behind Gemini interactive. You can go back and forth from voice into search. It gives you short-voice answers and it brings up a bunch of web pages so you can go in and dig in more. It’s a new form of modality that should be orthogonal to the type of information you want to consume. It’s just true that people have interacted with each other for years in all sorts of different ways, and we should make tech that is adaptable, [provide] that ability to handle any environment, and not have your form of input and your form of output be so deeply interconnected.

We’re talking about search, but I’m starting to think of it as just Gemini, and search is part of it. Google should be this platform that I go to do anything having to do with information. Do you see it that way?

Users think about what they expect from search in new ways, and it’s possible that over time, those things really blend, people prefer different versions. If I take a non-Gemini example, people have asked this question about Search and Maps. Should you use Maps to find a restaurant, or Search? It doesn’t really matter. We’re just gonna help you find a restaurant. The way I think of it now is that the user is always right. They should come to whatever tool they find most useful and get a really good information response. And the two have slightly different formats about how they respond, their interactive style, how they weigh creative or factuality. As a user, you should just pick whatever you think works, and we should make it work for you.

Is it a challenge that among billions of users, some of them probably don’t even know what an LLM is, and they just want to see their blue links?

It’s both a challenge and opportunity, most of all it’s a responsibility. We’ve got a lot of users, and they want different things, and we need to address both of them. That’s why we have both AI Overviews and AI mode. If you really want the cutting edge, you’re up for change, and having things change on you every other day because you want to try it, great. Go use AI mode. If you are like, ‘I want that thing I’m more used to most of the time,’ then use Search. And when you ask a more AI question, then you can ask AI Overviews. And if you decide you want to follow up more, you can go into AI mode. We designed this with the idea that probably everyone in the world is not going to be ready for exactly the same thing at exactly the same time.

Are you the early adopter power user, cutting-edge person? Let’s make sure we have a tool that works amazingly for you. For the rest, we’ll figure out how to polish it more, we’ll figure out what really matters, and then we’ll figure out how to graduate into a product that you’re ready to use.

Do you have any metrics on the accuracy of AI Overviews?

The quality has definitely gotten a lot better. It’s still the case about, what are you trying to do with the information? Lots of times, the answer that’s generally very accurate, but not perfect, meets your need.

On the other hand, if you ask a very sensitive medical question, we may try and help, but we’ll put a message at the end that’s like, ‘You should really go talk to a doctor.’ There was a time when people were like, ‘Wikipedia is imperfect. You should never use Wikipedia.’ This was a real thing. And now people are like, ‘I’m going to use Wikipedia for some classes of things, I’m not going to use it for another classes of things. I know how to use the tool, the tool has gotten much better.’

We’re seeing the same thing with LLMs. They’re getting much more accurate, but they’re not perfect. The web, like search, is not perfect. Even outside of LLMs, the web is not perfect. Information can be stale, but it both gets better and better, and people learn how to use it. And we try and make sure in search that it’s not just there, that it is easy to go look up sources that can help you. It’s easy to dive deep. I don’t view the ‘easy to dive deep’ as a weakness of the product. I view it as a key asset of the product. We should make it possible for you to dig in deep, and so we’ll continue to prioritize that. It is an incredibly exciting time for the web.

SEO is a whole industry now. Are you finding people doing that for the AI answers? Are they trying to game the system? And how do you contend with that?

People are always trying to game systems. Forget the internet. There was locksmith spam in the white pages before there was the internet. Where there are financial incentives, people will try and do that. But that’s been an area of focus for search, and a place where we’ve built expertise for years. SEO is neither a strictly bad or good word. It’s a great thing to make sure your content is findable and it’s not indexed, and it’s understandable. But you can have great SEO, and then you can have [people] trying to gain the system and spam. That’s something we’re going to be ever vigilant on because it’s a cat-and-mouse game. But we’ve been able to keep spam under 1% for years because we just stay very focused on it.

What does ad revenue look like in the AI mode versus the blue links? Is it going to be less revenue?

We’re about to start rolling out ads in AI mode. At this point, we don’t know. I think it will look different over time. The modes are different. The types of ads that we did in image search were different from the types of ads we did on main web search. We don’t think we should just show ads no matter what. We should show them when they’re high quality and relevant. And the need for commercial information continues, whether it’s an AI mode or main search. If you’re searching for where to go eat or what clothing to buy, you’re going to be involved in choice, and that’s an opportunity for ads. If people get more specific about what they want, they do more follow ups, they get more specific in the original query, then they’re sharing more of the intent, which means our ability to come up with ads that are of higher relevance and higher quality goes up.

It’s certainly something to work through, but there really is an opportunity. What we’ll see with AI mode is that we see that people are already issuing queries that are 2x to 3x longer than they do on main search, which means we can do better targeted, higher quality ads.

What are the ideas that people throw around for alternatives to the ad model? Maybe there’s some other way to keep the product free and still make money in the AI world that’s not possible today?

I don’t think we know exactly where agentic is going to go, and I think agentic potentially opens up the space for different ways to think about monetization. Traditionally, the way Google has often done this is, first, build something that really has high user value, and then figure out what feels like a natural way to monetize it. That’s what we’ll see in the coming years as the products evolve themselves. There will be more exploration. But I think we forget how much people actually want to choose. To the extent that people are going to continue to choose among commercial things, there continue to be opportunities for ads. There’s both new opportunities, and I don’t think it means ads are about to fall apart.

People at Google like to think about almost everything. When I first came to Google, there was this thing about the tether to the moon. That’s not my solution. That is like just drawing. It’s fun to brainstorm. We’ll see how it goes.

I want to know the tether to the moon equivalent in monetizing search. But I guess maybe we’ll find out one day.

We’ll find out one day.