I’m sitting in the back of Google’s cafeteria, which today has been repurposed for what (to me) looks like a very large amount of journalists. Right now, Elliot Schrage is welcoming folks. He just mentioned that the day is being webcast, so I’m going to assume it’s okay to blog this.
Now they’re doing a video to show what’s new in the last year. Mmm, techno music.
And now Eric Schmidt is up. Eric points out that Google is putting even more effort into core search quality. He also noted that our machine infrastructure scales very well with the growth of the web. “We have more people working on search than ever before.” He talks about the virtuous cycle of having more searchers, more advertisers, and more innovators.
Alan Eustace is up. He’s our Senior VP of Engineering, so he doesn’t have to wear a suit. Cool, Alan’s going to walk through the life of a query! He runs through the need to crawl, index, and then score relevant results. “Speed matters.” With 8 billion pages, it would take 253 years (I think I got that right) to fetch pages if you fetch one page per second. “It’s important that we gently crawl the web at very high speeds.” Alan talks about duplicate pages, which can vary from 30-50% of pages with a naive approach. Alan notes that you have to avoid infinite loops such as calendars. Freshness matters. Size matters, especially with long-tail queries.
Now Alan is walking through indexing. Heh. He’s working a simple example with posting lists. Queries with two words intersect those posting lists. So if heart is on page 5, 9, 25 and attack is on page 7, 9, 22, then the best intersection is on page 9.
Once you’ve intersected posting lists, you have a smaller set of documents to score. Alan mentions anchors and PageRank; lots of pages (and important pages) link to Stanford, so it’s fair to consider them an authority. Then Alan points out the anchortext “Knight fellows” on the Stanford home page and mentions that anchortext can be handy.
Alan debunks the notion that results are generated manually (it’s surprising how many people believe search results are created by hand). He mentions that 20-25% can be unique (and often hit the long-tail), which is why size and automated algorithms matter. Alan also mentions that thousands of machines contribute to the scoring of one query.
Hey, he just mentioned spam, and that it’s important to return quality information! Cool. Heh, he’s showing a spammed guestbook and mentions that that technique doesn’t work anymore. Now he’s mentioning spammers trying to build large link networks to look real. Google rarely discusses spam at all, so it’s a breath of fresh air to hear someone mention spam. For me at least.
Ooh, now he’s talking about eval and how it’s difficult to impossible to make a change that is 100% better on 100% of queries. And that it’s hard to do eval really well. He mentions that we take quality wins in our algorithms when we see the chance. He also touches on how hard international queries are to do well (segmentation in Chinese, for example).
Question: Improving the search experience from mobile?
Alan: We have a number of deals, but there’s less mobile content out there. It’s also important to make the entire web available, which is done via transcoding, which re-optimizes for a small screen.
Q: How does expertise apply to video?
Alan: Video is tough because you don’t always have the same signals or information (enterprise is another example where the challenges are different). Popularity and authority matter more. When you don’t have PageRank, there can be other ways to distinguish reputation.
Q: Non-search products don’t always have
A: We use the same infrastructure for everything (PC-based commodity parts running Linux). We also try to build on similar infrastructure (Google File System, Mapreduce, etc.). Search has to have many 9′s of reliability, but non-search product don’t always guarantee as much uptime. Something like a Labs launch might not be at every data center, for example. Labs is
Q: How does the use of personal information to make search better while respecting privacy?
A: Terms like jaguar or bass are overloaded (fishing vs. guitars). Starting from scratch every query would be a disadvantage. Personalized search (which is opt-in) is important and there’s still room to improve there.
Q: How do you measure the quality of your searches internally?
A: Heh. No way is Alan going to answer that. We put a ton of effort into that. Yup, he regretfully declines.
Alan wraps up and Elliot is back up. Elliot mentions that over half our queries comes from outside the United States, which is something most people don’t realize. Then he introduces Omid Kordestani, Nikesh Arora, Adam Freed, and Sukhinder Singh Cassidy.
Omid talks about how Eric told him to diversify income from other countries. “Get on a plane and don’t come back” is how Omid describes it. Now 40%-ish of revenue is from outside the U.S., so I guess the directive worked. AdWords is available in 43 languages now.
Q: How do you leverage services outside the U.S., and how does localization happen?
A: Products need to be easy-to-localize, no matter where they’re built. It won’t always be English or Mountain View-based.
Q: Delivering clicks to the right people important; how do you handle click fraud?
A: Click fraud is something that Google monitors for carefully. It’s a big focus for us as one of many things that we pay attention to, because Return on Investment
Q: Bambi asks about how much Google is investing in China and closing the gap with Baidu?
A: Omid: It starts with the users, and the investment therefore is often focused on engineering and product management and taking the long-term view. Sukhinder: Mobile in China is 350M users while internet is more like 100M users, so it’s an interesting, different market.
Q: Partnership and competition in Europe?
A: Lots of people partner with Google because they see that Google can send lots of user traffic as well.
Q: Yahoo! is almost “dominant” in Japan. How do you address that?
A: Omid: It varies by country. Local servers or services often help. Yahoo! has a longer history in that market, but those countries are a priority for us. Focussing on the user experience drives innovation, which is a good thing. Sukhinder: global reach provides a lot of opportunities, so someone that wants to sell a product world-wide will often find Google beneficial.
Q: Plans for click-to-call?
A: We will pay attention to that, and all ways to help users and advertisers.
(Wow. They’re doing a lot of Q&A today. Cool.)
Okay, last presentation is Jonathan Rosenberg and Marissa Mayer. Jonathan mentions changes from the last Press Day based on feedback, such as finishing earlier for 5 p.m. Eastern deadlines. Jonathan also mentions that the team is pushing more into the 70% effort on search and ads.
Marissa is talking about the fundamentals of search: comprehensiveness, relevance, speed, and user experience.
Jonathan takes the baton back to talk about innovation in advertising. For large advertising, things like Site Targeting, Content Bidding, and Position Preference. For smaller advertisers, things like AdWords Starter Edition. AdSense is in 23 languages. Gmail is in 38 languages.
Marissa says part of the philosophy is innovation, not instant perfection. Marissa points out that Video went through three major iterations in less than a year. This is a key point. I want to talk more about this in the future, but now they’re introducing new things. Um, http://www.businessweek.com/innovate/content/apr2006/id20060411_490873.htm is a good place to read.
Jonathan is talking about things that people might have missed. He mentions Google Pack and Google Local for Mobile. Someone in a white lab coat is doing a live demo of Google Local for Mobile. The demo was really nice; I’m getting tired of the white lab coat thing though.
Ah, they’re showing the ability to show trends in search. This is new. They’re using Sudoku as an example. It’s like “getting the keys to the Zeitgeist kingdom.” They did [full moon, equinox, solstice], and you can see the different frequencies and periods of queries. The journalist next to me just said “Wow.”
Now they’re showing local searches, so for [surfing] you can see that Honolulu, Brisbane, Perth, and San Diego. [boxers, briefs] says that boxers are more popular queries.
Moving right along, they’re introducing a new version, v4, of Google Desktop. It provide Google Gadgets, with layers, transparency, and animations. Pressing shift shift hides and views the gadgets. There will be an API available so that everyone can make gadgets. You can drag gadgets from the sidebar over to the desktop. Google Desktop can automatically suggest and confiure a personalized iGoogle home page. Damn. Now I’m going to have to upgrade from v1 of Google Desktop.
Moving right along. Marissa is mentioning something else new: Google Co-op.
Pausing to publish. Please excuse typos.
Ah, one example of that is Google Health searches.
Moving right along. Marissa is talking about Google Notebook. You can store urls, and it looks like you can drag stuff onto the notebook. Very mimex-y. Sounds like you can can make a notebook public too. Ah, they’re color-coded. So a private notebook is blue, and a public notebook is yellow-orange so it’s clear what is public.
Q: Gadgets and Notebook is half-browser and half-desktop. What’s the tension between them?
A: Marissa: We’re pragmatic, not religious. If you need richness, a client may be most appropriate.
Q: Some kind of AdWords vs. AdSense question.
A: Jonathan: Both are growing. (I need more caffeine to follow this answer. )
Webcast Q: Will we be able to see weather in Google Earth?
A: Jonathan: I think that would be cool. Remember that satellite photos lag behind for a while. Marissa: remember that weather is available already in Google search and the sidebar that Google Desktop Search provides.
Q: Other types of advertising, such as dMarc?
A: Jonathan: Anything effective with good ROI is interesting.
Webcast Q: Why not use clustering technology to help searchers refine queries?
A: Marissa: Google Co-op comes at this from a different direction by using labels from authoritative . Jonathan: You might think that this iteration would be helpful, but if the page already has relevant results, sometimes it slows the user down to
Q: Any chance of the ability to query for people with cell phone pictures?
A: Pictures aren’t easily searchable, so it’s a harder (but interesting) problem.
Q: Is the definition of search changing, e.g. because websearch is maturing?
Plus webcast Q: Is Google a portal?
A: Marissa: Originally portal meant doorway, and Google is the first place to start for a lot of people. The definition changed in 1999. Our layout and experience are really different though. Marissa also believes that there’s a lot of potential headroom left in search. Jonathan: still room to interpret queries better, still room to personalize. And lots of ways to make the ads more relevant?
Q: What about competition with Microsoft?
A: Jonathan: We’re going to continue to focus on innovating as fast as we can.
Q: Monetization of new products announced today?
A: Jonathan: history is building great products, and then adding ads where it makes sense. We’re not focussed initially on the monetization. Marissa: Advertising follows consumers. If users love
My hand hurts. Geez people, ask and answer questions slower!
Q: Do the participants in Google Co-op pay?
A: Marissa: No, no one pays or is paid. Jonathan: Which fits with our philosophy.
Q: Do you need to have the Google Desktop installed to get Gadgets? Maybe put it on mobile?
A: Marissa: Yes. Still too early to say. We’ve done it with the personalized home page and Google Local, for example.
Okay, now it’s a general Q&A with Eric, Larry, Sergey, Elliot and David Drummond
Q: Talk more about Microsoft?
A: Sergey: Given the Microsoft’s history, it’s important for us to be mindful and on our toes (I’m paraphrasing, of course.)
Q: What do you think of Quaero?
A: Larry: Anything that gets users more information, we’re in support of.
Q: Timeframe on TV ads?
A: We don’t pre-announce our products, and we’re generally excited about measurement and metrics in new spaces like video.
Q: Talk more about the systemization within Google? What changes happen? Is there the risk of bureacracy?
A: Eric essentially says that measurement is important, but we of course want to avoid bureaucracy.
Q: Skype is now better than a cell phone quality. What about IP TV? Will you pursue it?
A: Ideally, we want info to be
Webcast Q: From Philipp! He’s asking if we have negotiations with China about sites that need to be dropped?
A: David: (I believe that he said as with any regulatory authority, sometimes communication takes place. Watch the webcast to get the exact language.)
Q: Missed this one.
Q: Microsoft is spending more on machines. Will that cause Google to spend more?
A: Historically, we have always been aggressive in investing in our business and R&D. I don’t expect us to stop, but we won’t be wasteful.
Q: Wireless WiFi in San Francisco; does that mesh with the need for uptime with a service provider?
A: Larry mentions that Google search has many 9′s of reliability, so we have the capability to be reliable. Excited about working with Earthlink.
Q: Goal to create a larger API for all applications on the web? And a way to monetize that?
A: Sergey: No, but it could happen. No uber-API goal right now. Eric: We’ll do what makes sense based on what users need.
Q: User-generated content and how to get access to more content?
A: Larry: The web is a large fraction of user-generated on the net, so we have a history of incorporating feedback and information. Looking forward to providing new ways (e.g. Google Co-op) that people can provide more intelligence to computers.
Q: What about buying companies that have content?
A: Most people benefit from providing their content to Google. That’s a very healthy ecosystem that is working well?
Q: Do you think “Do no evil” can resist being a business?
A: Sergey: “In short, yes.” We’ve done a good job on aligning our business goals with the goals of our users and doing the best we can. “In general, we’re doing a very good job on staying true to out mission.” But there will always be some people who disagree about some decisions that we make.
Q: Missed this one.
Q: Collecting demographic data or additional information such as Microsoft?
A: Larry: history is that we would fail compared to Yahoo! because Yahoo! has zip codes for lots of people. Larry’s contention is that providing a zip code only take ~2 seconds to provide, and Google can collect helpful info in a way that respects the user when it makes sense. The larger thing isn’t demographics, but having the user’s exact interest right when they’re searching.
Q: How do you see net neutrality playing out, and how are you preparing for worst-case?
A: David: It’s early to say. This is a large public policy issue and isn’t Google-specific. Elliot: the worst case scenario is one that hurts innovation. Google would be less affected than smaller companies.
Q: Google has a large database. You say “we’re not evil” and won’t abuse data. How can you guarantee you’re not evil?
A: Larry: We rely on the trust of our uses, and if we did something bad as defined by our users, it would severely hurt us. The good news is that large companies that have a good brand are aligned with users’ interested: they have a large disincentive that keeps Google trying to do the right thing for users. I would worry more about companies that don’t have a user brand but are gathering a ton of info.
Q: What if Google is bought by another company?
A: Sergey: Email is the most personal for me, and for most people. It happened with e.g. Hotmail going to Microsoft, and confidentiality of email in general has been protected well by the industry. Eric: the shareholder structure makes it less likely.
Q: Why the openness push now?
A: Elliot: Part of it is that the company has gotten bigger, it’s important to pay attention to responsiveness. We do have proprietary stuff, but Google as a whole wants to be more transparent when we can.
Q: How to compete in Japan? Outlook for Japan?
A: Sergey: I’ve been pleased with the performance in Japan and the amount of competition. Eric: first mobile advertising was done in Japan.
Q: Muni wifi is interesting–how big is this initiative?
A: Larry: we’ve said that this is an experiment. Google works by trying lots of different experiments.
Q: Do you worry about dependence on underlying providers?
A: Larry: net neutrality means that few things work as well as the net.
Q: Interest in changing “upfronts” and how TV advertising works?
A: Sergey: I was the first to stop objecting to AdSense (Eric notes that Sergey helped drive AdSense a lot). Sergey: We’re going to continue to try experiments and try to bring something to the table.
Q from Kevin Delaney: Do you want to clarify the
A: Sergey: Already, a lot of people use Gmail to store email and servers on the web, but nothing to announce. We discuss ideas internally (count is up to ~4000 ideas), and many of those bear fruit, but I wouldn’t read too much into things.
Q: Since there is only finite advertising inventory, how long until it dries up for various services?
A: Sergey: from the point of view of providing places to put ads, there’s a lot of places and web pages to put ads. From the advertiser perspective, many advertisers would happily take twice as much inventory. Larry: online spending is still small compared to total advertising.
Q: Will people be able to get by with just cloud ? Comments or color about Microsoft, because let’s face it: that’s what we’re going to go back and write about.
A: Larry: With Gmail, we didn’t think about it as a cloud application–just an improvement on email, e.g. better search. We think about making things better, and better enough that people want to use them–not replacing other stuff. We think about competition as every company should, but we also try really hard not to obsess about what another company is doing. Example is Gmail, which changed how people viewed email. If you use your computer, there’s still all sorts of frustations to be solved. In general, you get there by thinking about what people want and need, not what other companies are doing. Another example is the iPod. Looking at other companies wouldn’t have produced the iPod–Apple was looking forward, not sideways. Eric: We have the luxury of time because of success to look for new problems to solve. Collective belief that there’s room for more than one winner. The broader play is that many companies can be winners.
Q: Driving traffic to Myspace? Bambi says that the people that leave Myspace go to search engines, so News might want to buy a search engine. Thoughts?
A: missed Eric’s answer.
Q: Plans to integrate research on statistical machine translation into other services?
A: Sergey: This won awards last year, but we built it to be the best possible, but not the most productionized.
Q: Mobile marketing? How to get advertising on cell phone?
A: Eric: We have a number of trial forms. If you do the numbers, cell phones are much more prevalent, so over years that may . No interest in doing our own MBNO.
Q: Is social networking a big part of Google’s future? Or do they not scale as well?
A: Eric: Google Co-op will be very algorithmic and provide better ways to improve search, so
Q: What are you doing to attract more branded advertising?
A: Larry: we emphasize usefulness to users, and that’s been successful with a large network.
Q: More transparency to the investment
A: Eric: We don’t want to get into the guidance business, but we do want to keep looking at ways to be more transparency.
Q: Apple talks about iPod sales, while search industry is murky. Have you talked about disclosing more information?
A: Many third parties estimate this.
Q: Are you interested in bidding on wireless? How many data centers and complaints from webmasters?
A: We look to partner in wireless. We’re looking at the feedback. Sergey: overall, this new search index is a definite win, in our opinion. Queries that webmasters may be doing are not as typical as normal users’ queries, but we have a team looking at the feedback now. Good/correct answer from Sergey and Larry if you want to watch the webcast archive.
Q: missed this one
Q: Tradeoff between free-and-not-perfect and a product with great quality?
A: Sergey: We think of them as great already even though they’re not perfect. “We probably abuse the word beta a little bit.” Labs is supposed to be the place where things can fall down. People expect a lot more from Google these days, so we could communicate more about which are in the beginning stages and which are more mature. Elliot: conception of Labs is a great example. We wanted to be able to throw stuff up, but people expect more from Google. PR will try to communicate more clearly about expectations.
Q: Why did you cash in a huge pile of shares last year? Do you wish for the days when you were smaller? No China or government conflicts, no inviting press?
A: Eric: “We are delighted to have you here.” Execs around IPO time were required to enter into 10b5 plans as part of best practices. Sergey: I’m happy holding 80% of my stock all in one company. “The vast majority I intend to keep forever.” Eric: What about simpler life? Larry: I remember when we were 100 people, my argument was that search was too important and too meaningful and too global to the world for a small company to really succeed. We really do believe that we’re accomplishing a lot and making the world a better place, and you have to be larger to do that. Eric: There are so many overwhelming benefits for things like recruiting and resources. Now Eric is telling a story. It would be too long to type.
Okay, The Q&A is over and people are going to grab lunch. So I post an update again.