Bummer! This is just a preview. You need to be signed in with a Pro account to view the entire video.
A Secret Algo Project Launch65:39 with Pete Meyers
Tired of chasing the Algorithm? Dr. Pete has been watching the Google weather and will help you bring your storm-chasing to the next level.
[MozCon 2012 - Seattle] [SEOmoz] I was watching Marty from the speaker room. 0:00 [A Secret Algo Project Launch - Dr. Peter J. Meyers, Rogue Scientist - SEOmoz - @dr_pete] 0:02 I was joking with him last night and I said, 0:04 "Marty, I've heard you're really awesome on stage." 0:05 "What am I going to have to pay for you to bomb?" 0:07 And I think there was a little bit of a miscommunication. 0:09 I think he heard "atomic bomb" because that was some seriously weaponized knowledge. 0:11 I was a little worried for you guys. 0:16 So I'm going to tone it down for a few minutes 0:17 and we'll come back up and give you a break. 0:20 So A Secret Algo Project Launch. Sounds exciting. 0:23 I have a confession to make. 0:27 I originally pitched a completely different idea for this talk, 0:29 and it was a much more sane idea. 0:33 It was one of those talks where I might make slides for 2 or 3 weeks 0:35 and I'd get to sleep through the night 0:38 and I wouldn't have migraines or anything like that. 0:40 And then back in November, before we were even putting the program together, 0:43 I went to Rand and I said, "Rand, I have an idea." 0:47 And what I thought I said was, "Rand, I have an idea and it's kind of weird 0:51 "and it's kind of complicated and I don't really think it's going to work 0:57 and it's going to take a lot of time and work and maybe you should talk me out of it." 1:01 And I think what Rand heard in his head was, "Rand, I have an idea 1:05 "and it's going to be awesome and it's not going to cost you any extra money 1:09 and it's not going to take any SEOmoz dev resources." 1:12 And Rand said, "Yay! Do it." 1:14 And so then I came back, I built a proof of concept and it worked a little bit, 1:18 and I went back to Rand and I said, "Rand, I think we can step it up a level." 1:21 And again, in my head I thought, "Well, this is going to take a lot more effort 1:25 and I'm not sure it's going to work and I'm not sure we're going to get any data." 1:29 And what Rand heard was, "Yay! Do it." [laughter] 1:32 And so then I got my MozCon topic and I read it 1:37 and it was A Secret Algo Project Launch 1:40 and it was not what I had pitched. [exhaling deeply] 1:43 And I kind of went, "Holy shit." 1:47 It's sort of like getting published before you actually have done any of the research. 1:50 I had no data at this point. This was barely working. 1:53 And this is what I thought in my head, but apparently what I said out loud was, 1:57 "Yeah! Let's go for it." And, "I think we can build it in time for MozCon." 2:00 [laughter] So I can only assume that I was drunk for the first half of 2012, 2:06 and I learned a few things. 2:11 So it's been a wild ride; I'm going to take you along for it. 2:13 This is a little different. 2:14 There's going to be a lot of data. 2:16 But of course one of the top things I learned was next time before I talk to Rand, 2:18 ask this guy because the Magic 8 Ball might say no. 2:22 Rand will never say no. [laughter] 2:28 Rand is, "Go, go, go." 2:32 It's been a while since I've been up here. 2:35 Some of you know I used to work in the trade show industry. 2:37 I actually spoke pretty often, and I kind of got a little burned out. 2:39 And so it's been 6 or 7 years. I'm a little shaky. It's feeling good, starting to feel good. 2:43 But I just want you to remember before you fill out your evaluations, 2:49 before you judge me too harshly, Dr. Pete loves the kitties. [laughter] 2:52 And I love the babies and the puppies too. [laughter] So go easy on me. 2:59 So how did this all get started? 3:07 Back, I guess, in early 2011 I had another idea that I went to Rand with. 3:09 I really shouldn't have Rand's email at all anymore. 3:14 I said, "We have this rich history of the algorithm, 3:18 and there's some great resources out there about all the Google updates," 3:21 but as I dug into them I was kind of disappointed. 3:25 I felt like either they were really early in the process, 3:28 2003 back the bus and update, and they went on for a couple years and they fell off 3:32 or they were really recent and they did a couple years back. 3:36 But every time somebody built something like this, it just kind of died. 3:39 And so I really wanted to dig in and look at the whole history and put out a resource. 3:43 And we published this last year, and it's been very well received. 3:48 It's been great but what I learned along the way 3:52 is as much data is in this and as much as we put into this, 3:57 I think what struck me the most was the tremendous gaps we have 4:00 in our knowledge about the algorithm. 4:04 We've been tracking the algorithm for 10 years now, 4:06 and there is so much we don't know. 4:08 And so a lot of what I want to talk about is what do we know, 4:11 where do we get that knowledge from, and what are the gaps? 4:13 And I said to Paddy yesterday—he did such an actionable talk—I said, 4:16 "Your talk was all actionable; my talk is totally inactionable," 4:21 because my lesson here in a way is as bad as you think it is with Google, 4:25 it's 100 times worse. 4:29 The algorithm is changing on a level that is going to be shocking to you. 4:31 So I hope there's a happy takeaway, but I'm not—[laughing] I'm not sure. 4:37 I may damage you more than Marty. I don't know. We'll see. 4:41 So what do we know? 4:47 Ten years, 2003 to 2012, from Boston to Penguin 4:49 we have 20 named updates. 4:52 And we get very impressed with ourselves and we go, "Oh, Panda, Caffeine, Penguin." 4:55 "We know so much. We're so smart. Panda 3.6, blah, blah, blah." 4:59 And we think that because we gave it a name we know something. 5:02 But look at this time course. Ten years, 20 named updates. That's 2 a year. 5:06 On the scale of a company that makes $40 billion, 5:11 you think they're only doing 2 things a year in organic search? 5:15 So what does this number really look like? 5:19 Well, unfortunately, in 2010 we found out. 5:21 This is Eric Schmidt testifying before Congress. 5:25 It's about as close to the truth as you're going to get. 5:28 Google was in some hot water. This was a very prepared statement. 5:31 Eric Schmidt told us that there were 516 changes to the algorithm in 2010. 5:35 516. We have named 20 of them in 10 years. This was 1 year. 5:41 But it gets worse because Eric also said this. 5:47 There were 8157 side-by-side experiments, 5:51 probably like a quality rater task—people saw an A and B screen— 5:56 but something that they saw that was being tested offline. 6:00 Of those, we don't know if it came from that or if this was a different set, 6:04 there were 2800 click evaluations. 6:07 And what that means is that 2800 things got tested live on Google.com. 6:10 Maybe only 1% of people saw it, but there were 2800 changes 6:15 that were rolled out to some degree in 1 year. 6:19 And of those, 516 stuck permanently. 6:23 So what do we know? 6:29 This is the 2800 things that got tested. 6:31 The purple up there is the 516 that got permanently changed. 6:36 That tiny, tiny red sliver up top, that's the 8 things we know. That sucks. 6:40 And that's not the named things; that's actually my broad list of things 6:47 where we got a little liberal with it. 6:51 This isn't good. There's so much of that 516 we don't know. 6:53 So where do we get our information? [laughter] 7:02 We get our information from 2 sources. 7:08 The one on the left is what Google says. And what does Google say? 7:12 Matt says, "Look, a panda! He's your fluffy nap time pal." 7:15 "Don't you love him?" 7:19 And then over here on the right is the webmaster saying, 7:21 "Holy shit! Giant rampaging penguins!" 7:24 And if you're smart, you're that guy. 7:28 If you're smart, you're that guy hiding behind the desk writing your will, 7:31 finishing your Sudoku, whatever he's doing. I don't know. [laughter] 7:34 But he's not up there smiling. 7:38 This isn't an anti-Google conspiracy talk, although it's going to sound like it later. 7:41 [laughter] A little bit. 7:47 I think we have to listen to what Google says. 7:50 Google is saying a lot more, and I think we both have to listen and learn how to listen 7:52 because they're also saying what they want us to hear. 7:56 And there's more information out there, and that's both good and bad. 8:00 So I'm going to look at these 2 pieces of information— 8:03 what Google says and what we observe ourselves— 8:05 because this is kind of the core of what we've learned. 8:08 So what does Google tell us? 8:12 First of all, we have clues from patents. 8:14 This is an excellent series from Bill Slawski, SEO by the Sea, 8:17 10 Most Important SEO Patents. 8:21 I think it's like a 12 or 13 part post now. 8:23 I am not a patent guy. Bill does tremendous work. You should definitely read it. 8:26 These are clues. These are clues about Google's capabilities. 8:31 But it's tricky because there is a competitive aspect to this. 8:36 Google may file a patent that is to protect something, 8:40 they may file a patent that's purposely misleading in some cases, 8:44 but more often than not what they do is file patents that either haven't come into play 8:47 or won't come into play until they have the technology to make it happen. 8:52 I don't remember who was talking about Caffeine earlier 8:57 where the Caffeine rollout came and all of a sudden 8:59 these things Google had been sitting on for months all started to come into play 9:01 because they had the architecture and the speed to do it now. 9:05 So the trick is there's a lot of clues in there and it's interesting stuff, 9:08 and some people dig through it really well, 9:12 but we don't know just because they patented something 9:14 that it ever became part of the algorithm. 9:18 And if we do, we don't know when. 9:20 It could take years for them to be able to do something 9:22 that they could do theoretically, 9:26 because on the scale of Google globally, just because you can do the math 9:28 doesn't mean you can crunch the numbers on that massive scale. 9:32 We also have now the direct statements. 9:38 Google is talking to us more and that's great, 9:41 but they tell us what they want us to hear. 9:44 This is a tweet from Google, one of the Panda updates. 9:48 Okay, they're talking to us, but 1.6% of queries noticeably affected. 9:53 1.6% of which queries? 9:57 1.6% of all queries, 1.6% of query volume, 1.6% of unique queries? 10:00 And noticeably affected. Did I notice? Did Matt notice? Did my cat notice? 10:06 I don't know who's noticing. 10:12 And this is Matt talking about a mistake they made, which was actually very transparent. 10:16 So I think we do have to follow this. It's interesting. 10:20 One of the staff accounts, Pierre Far, is on Google+ a lot, 10:23 and Pierre actually comes out of the webmaster world 10:27 and I think kind of has our backs. He's a good guy. 10:28 So there's a couple people worth following. 10:31 This is the official Google blog, Inside Search—one of them. 10:35 This is about the Penguin update. 10:39 Again, these things are worth reading, they're good information to dig through, 10:41 but they're Google's words. 10:44 Back in November we started getting some interesting data. 10:49 Google started publishing the search quality highlights. 10:52 And basically it was a list every month of what they did the previous month. 10:56 And this is great data. I mean, I love to dig into this. 11:00 I was talking to AJ last night and he was saying, 11:04 "You know that this one quality update a couple months ago is Italian 11:08 and it translates to this?" And I'm like, "Holy shit. That's amazing." 11:11 And there's probably, like, 3 people in the audience 11:15 who would just think that was awesome. 11:17 But for most people reading through this, it's really dense 11:19 but I'm afraid it's also kind of a distraction for us in SEO. 11:24 Google is feeding us all this information, making it look transparent, 11:28 but they're piling a lot of jargon on us and we're all going, 11:33 "Oh, well, look. They told us all this stuff. Now we can go and do our thing." 11:38 "Now we don't have to scrutinize or pay attention to what they do anymore." 11:42 And the reality is this is advantageous to them. 11:47 I'm not anti-Google. I use Google every day. I love it. 11:53 But I think this is a little bit of a game for what can they feed us 11:56 to keep us happy without really telling us anything? And so we have to be careful. 12:04 But I did want to dig in. Again, 516 changes in 2010. 12:08 Ahmet Singal had a quote somewhere that there were about 525 in 2011, 12:13 so this was not an anomaly. 12:18 So I counted. What do we have so far this year? 12:20 We have not gotten the June search highlights yet. That's overdue. 12:22 Google hasn't said why. 12:25 But 5 months so far we have 198 changes reported. 12:28 If we roll that out to the year, we can project 475. 12:32 I actually think that late 2011 and early 2012 they weren't quite as transparent 12:35 as they have gotten, so I have a feeling that's a bit of an undercount. 12:40 So I think we're in the same ballpark. The last 3 years about 500 changes a year. 12:44 I did a little breakdown, quick and dirty here, 12:51 what categories, what buckets do these changes fall into. 12:54 You look at the top. The red is core major. 12:57 These aren't all named updates, but these are things that I read and thought, 13:00 "That's pretty big." That's quite a few so far. 13:02 We got a little bit of local, a little bit of image. 13:07 Vertical and international are actually pretty big, 13:09 but vertical is sort of everything that's not organic and local and paid. 13:11 And then we got a lot of core minor, and core minor comes down to things like spell check, 13:17 speed improvements, snippet changes, sitelink changes, 13:21 things that they are algo changes but they're often display or feature sort of things 13:24 that aren't impacting us in a big way. 13:29 And then we have another problem. 13:36 This is Panda. 13:38 And I didn't add 3.9. 13:40 Panda so far, 17 updates in 17 months, 3 major and 14 minor. 13:43 This is the reality of an algo update right now. 13:50 This is not Florida where it hits hard and we move on. 13:53 This has been going on for the last year and a half, 13:56 and every month we're getting data updates 13:59 and every once in a while we're getting algo updates. 14:01 And we've kind of settled into this bad pattern 14:03 where we hear there was a Panda update and we sort of go, "Okay." 14:06 3.7, 3.8. 3.9 just hit a couple days ago. I do have a slide on that. 14:10 And I think even the updates we name are getting more complicated 14:15 and more difficult to parse. 14:21 We don't really know what those small updates mean. 14:23 And actually Panda is odd because Panda is happening offline, 14:25 Panda is being processed and fed back into the algorithm. 14:28 We think Penguin is doing the same thing. 14:30 So now we have 2 of these things out over here getting fed in 14:33 and the main thing is changing, and so it's a mess. 14:37 Another problem with these updates we have 14:42 is we don't really know what's big and what's small. 14:44 So these are 2. This is my wordiest slide, I promise, Rand. 14:48 It's still bullet-free but it's wordy. 14:51 These are 2 of the updates from search quality highlights the last few months. 14:53 So number 1, we have tweaked the handling of anchor text. 14:57 "This month we turned off a classifier related to anchor text." 15:01 "Our data suggests that the other methods of anchor text processing had greater success, 15:04 so turning this off made the component cleaner and more robust." 15:08 I clean my table with Lemon Pledge. It gets cleaner and more robust. 15:12 I don't really know what that means. 15:16 On the second, keyword stuffing classifier improvement, 15:19 "We have classifiers designed to detect when a website is keyword stuffing." 15:21 "This change made it better." 15:25 All right. Great. 1% better? 2% better? 50% better? 15:27 Number 2 was part of the Penguin update, we're pretty sure, 15:32 1 of 2 or 3 things they revealed, 15:35 and number 1 we have no idea. 15:37 If you read number 1, it was a change to how they value anchor text. 15:39 That could be monumental and it could be nothing. 15:43 And so I think we really have to be careful. 15:46 Google is feeding us a lot of information, 15:48 but in a lot of ways they're not really telling us anything. 15:51 And so we have to be careful and we have to not stop digging ourselves and get lazy, 15:54 because this information is coming to us. 15:59 All right. So what do we observe ourselves? 16:06 One of our big sources of information, unfortunately, is chatter. 16:11 It's just what are webmasters talking about? 16:16 And it's like the CIA with al-Qaeda when they talk about chatter. 16:19 We're hearing thousands of voices and trying to figure out if something happened. 16:22 And the thousands of voices are saying things like, "Google sucks." 16:27 "I dropped my rankings today. I'm in so much trouble. Was it Panda, was it Penguin?" 16:30 And there's hundreds and hundreds of these, and we're trying to interpret what it means. 16:34 And there are some great people out there, the people at WebmasterWorld. 16:37 For years, if you go back to 2002-2003, that's the only place that still archives information. 16:40 That's the only place that's served webmasters this long. 16:47 Meg and Ted and some of the people over there are amazing. 16:50 But we're trying to take all this noise of all the things people are complaining about and say, 16:52 "What does that mean? What really happened?" 16:58 And it's gotten worse because now we have localization and personalization 17:02 and now everybody is seeing different things too. 17:06 So now the chatter, there's not even a unified experience of search 17:09 that we're all seeing every day where if 10 people said something changed 17:13 we could go, "Yeah, okay. Something changed." 17:16 Now all 10 people are seeing 10 different things. 17:18 This is a search a couple months ago for coffee in Chicago. 17:21 Even my core organic is localized. It's my local Starbucks chain. 17:24 It's personalized. My images are personalized. 17:29 I always make fun of John. I guess John loves French press. 17:31 I've got to ask him about that. 17:34 So what does it mean when all these people are talking about something that happened 17:36 on any given day? 17:40 Now I want to show you how stupid 17:42 the whole process of tracking algo updates has really gotten. 17:45 This is a post by Barry Schwartz. 17:49 "Google Mocks Me For Missing Panda 3.5." 17:52 I am not making fun of Barry. 17:54 Barry has actually been a tremendous asset during building this project, 17:56 and Barry is doing a really hard job for us and I think a really thankless job. 17:59 But the process is idiotic. And here's how the process works. 18:05 A lot of people complain on any given day. 18:09 A lot of people wake up and go, "Oh, crap. My rankings dropped. Was it Panda?" 18:11 And Barry reads it and Barry says, "You know what? Enough people are complaining." 18:15 "I think something happened." 18:20 So he goes to Matt and Barry says, "Matt, did something happen?" 18:22 And Matt says yes or no. 18:25 And Matt says yes or no in the special way that Matt hears the question. 18:27 And I'm not trying to attack Matt, but when Matt says, "No update," 18:31 no update does not mean that none of the 500 things that happened happened today 18:37 because that's 1.4 things per day. 18:41 Every day Matt should say, "Yes, something happened." 18:43 And almost 9 out of 10 days Matt says, "No, nothing happened." 18:46 So what does nothing mean? 18:50 Nothing means no Penguin, no Panda, no update we care to name today. 18:51 That's what that means. 18:56 This is how we get our updates. 18:57 Barry asks Matt. 19:00 And I love Barry and he's awesome and I'm not making fun of him, but this sucks 19:02 because this is a game and this is Google's game. 19:06 I stole this from a meme. This is Gene Wilder. These is Gene calmer. 19:14 [laughter] If you haven't seen Young Frankenstein— 19:20 there are some 20-year-olds in the audience—you should go do it. 19:23 This is the fundamental question I'm trying to answer. What's normal? 19:25 What's a normal day for Google? 19:31 Because we have all these things happening. 19:34 We're doing SEOs, we're trying to change the rankings every day. 19:35 So how do we know when something abnormal happened 19:40 if we don't understand what a normal day is? 19:43 And what I'm going to show you is that we have no fucking clue what a normal day is. 19:46 [laughter] And a normal day is pretty wild. 19:51 A normal day is like Marty on the stage. [laughter] 19:52 And so you don't want to see what an abnormal day looks like. 19:56 So I built something. 20:02 It was originally called Project: Algo Alert. 20:05 It evolved. It was about 50 keywords when we started. 20:07 But here's what it breaks down to right now. 20:10 We tracked the top 10 for 1000 keywords every 24 hours. 20:12 You might think, "Why 1000? That doesn't sound like a huge number." And it's not. 20:17 A couple reasons. 20:21 We built our way up and we found that from about 100 to 500 to 1000 20:22 the noise didn't go down that much statistically. 20:26 But I also wanted to do something a little different. 20:29 There's some tools out there like certain metrics now 20:31 that are taking their aggregate client data, very large sets, 20:35 and trying to see how the algorithm moves. 20:39 And the problem is that their data sets are changing every day. 20:42 Every time a client does new rank tracking, they're adding more and more data. 20:45 And so there's a certain after-effect to that noise. 20:48 So I wanted to try something different and take a very tightly controlled set of data. 20:52 So we have 1000 keywords now. 20:55 They're sampled across 5 volume bins from the AdWords data. 20:58 They're delocalized, so if it's something like New York restaurants, that got thrown out. 21:01 They're detemporalized—I don't know if that's a word. Sure. 21:05 If it's something like 2011 Academy Awards, we throw that out. 21:08 We depersonalize and delocalize the crawl. 21:13 So we're trying to get at the core algo as much as the core algo still exists. 21:16 We take those keywords and we track them every 24 hours at roughly the same time 21:22 from roughly the same location. 21:26 And then we measure something called the deltas, and this is just the rate of change. 21:31 But what we're doing is not looking at any given keyword; 21:34 we're looking at the top 10 as a whole for that keyword. 21:37 So we have 2 metrics we use. This looks worse than it is. I think you'll get it. 21:41 One is called Delta100. Basically it works like this. 21:44 I have 2 top 10s and I want to compare them. They're the same keyword. 21:48 If a ranking in the top 10 didn't move, it's +0. 21:52 If a ranking moved, let's say it went from 3 to 5, it's plus the move, so it's +2. 21:56 If it went from 5 to 3, still +2. I'll explain that in a minute. 22:01 And if it falls out of the rankings completely, it's +10. 22:04 So Delta100 for any top 10 can go from 0 to 100. 22:07 Delta10 we just take the square root. 22:12 That's a little bit of mathematical trickery. 22:16 Delta100 is a power curve, and we want to normalize it 22:17 because if we normalize it, we can do things like take the mean and standard deviation. 22:21 There is a set of processes called Box-Cox transformations, 22:25 which is also fun to say, and it turned out that the square root works. 22:28 I know I'm going to get some questions later, 22:34 and I would love to talk to anyone afterwards— 22:36 I'll be at the party tonight and be around till Saturday— 22:38 about, "Okay, Dr. Pete, +10 if it falls out." 22:40 "Well, what if it falls out from number 1?" 22:43 "That's a lot worse than if it falls out from number 10." 22:44 "And what about this, what about this?" 22:47 I want you to understand that these are not just where we started. 22:49 We've also had Delta10S, Delta55, Delta55X30, Delta55X7, which I call DeltaX, 22:53 and DeltaL, and here's the basic story. 23:00 Delta55X7 is the most complicated thing I measure, 23:04 and I'll try to explain it really fast, but it's really boring. 23:09 Basically I do exactly that. 23:13 If it falls from the number 1 spot, it's a 10. 23:16 If it falls from the number 10 spot, it's a 1. 23:18 So it goes from 1 to 55. 23:20 Then we take that and we take the Delta55s for the last 7 days, 23:22 we take the average of that and we compare it to today and we get a multiplier. 23:27 And then we take the cube root of the multiplier 23:30 because the cube root is the best Box-Cox transformation. 23:32 Sounds awesome. DeltaX! 23:35 But it looks exactly the same as Delta10, and Delta10 is a lot easier to explain, so Delta 10. 23:38 Actually, my favorite Matt, Dr. Matt—I don't know if he's still around— 23:44 actually, Dr. Peters. We like to be Dr. Pete and Dr. Peters. That's cool. 23:47 He has been a huge help during all of this. 23:52 He came up with DeltaL, which was, "Let's change the top 10 rankings into a word." 23:54 "So on day 1 it's ABCDEFG and then on day 2 if it changes, 24:00 "like if the number 2 and 3 spot change, it will be ACBDEFG, 24:04 and then we'll take the Levenshtein distance between the 2 words," 24:07 which sounds really awesome, except it doesn't work at all. [laughter] 24:11 But it was cool. Matt and I had a lot of fun. We sent Rand a lot of emails. 24:16 So what did we find out, now that you have no idea how this works. 24:21 A lot of data. 24:27 I just want to start showing you—we'll mostly be showing Delta10s, 24:29 but these are a couple Delta100 curves. 24:33 This is 1000 keywords across the bottom, 24:36 and this is what a normal day looks like. 24:39 We have some high deltas—40, 50, 60—drops off pretty rapidly. 24:41 We have a plateau at 10 because if anything falls out it's a 10. 24:46 So that's kind of an anomaly of how we measure. 24:49 And then we have another plateau at 2 because if anything switches it's a +2. 24:52 What does a busy day look like? 24:58 This is Penguin April 24th. 25:01 So what you see is not some kind of weird spike, 25:05 not some kind of weird anomaly. 25:07 For a major data update everything moves, everything shifts. 25:09 All the keywords spike, and more keywords show change. 25:13 So this was good news for us. Something big happens, you measure it. 25:17 All right. So what did we measure? 25:21 Our current round of data collection started on April 4th. 25:27 The project goes back a bit there. Our 1000 keyword set goes back to April 4th. 25:30 So the first thing we heard after April 4th was Matt's announcement 25:34 from that G+ post before that Google goofed. 25:38 They classified a bunch of domains as parked domains and they screwed them up for a day. 25:42 And on the day of that glitch we saw a little bump and we said, "All right." 25:47 "Maybe we're doing something right." 25:53 But not a big one. We held out till about a week later when the Penguin update came. 25:55 One day hit hard. 26:05 I was probably the happiest person about the Penguin update [laughter] 26:08 and I felt really bad because I woke up that morning and I was like, "Yes!" 26:12 "Spike!" And then I was like, "Oh, shit. My poor clients." [laughter] 26:16 They weren't so happy. 26:21 But it seemed like things were working, and this was the highest Delta10 we had. 26:23 I am cheating. These Ys are all expanded. This isn't exact science. 26:27 We're trying to track something that's very noisy, and we are amplifying it 26:33 because I want you to see the differences, I want you to see what happened. 26:37 I'm not going to make wild claims about this data. 26:40 The error bars, I didn't put them on there because they're ugly. 26:42 But when the big things happened, we knew we were on the right track. 26:46 And this is the same scale. This is Penguin compared to the glitch. 26:52 Penguin was big. 26:55 Kind of a unique opportunity, one of the things we do 27:01 is we capture all those top 10 URLs and we store them every day. 27:03 So when something big happened, 27:07 we theoretically got to go back and see what it was. 27:09 And so Rand said, "Who were the winners and who were the losers?" 27:12 That's what everybody wants to know all the time. 27:14 It sounded like a good idea because whenever there's an algo update 27:19 we get a big list of the winners and losers. 27:22 But what we get in that list is individual sites who won and lost based on the update. 27:25 And I was looking across the whole data set, 27:31 and so I ran a bunch of these metrics and it kind of dawned on me that, 27:34 "Wait a minute. What goes up must come down." 27:37 In the top 10 if 2 people fall out, then 2 people must come back in. 27:41 For every winner there's a loser. 27:45 And so we're really just looking at the rate of change. 27:47 And so I ran these metrics and all the bars kind of looked the same, 27:49 and I kind of dropped it and I said, "This winners-losers thing, forget about it." 27:52 And then for some reason, I don't remember why, 27:57 I went back and I said, "You know what?" 27:59 "I'm going to do that winners-losers thing again, 28:00 but I'm going to look at the Penguin day." 28:02 And so all these bars are about the same, the blue and red, 28:05 except look at Penguin April 24th. 28:08 We had way more winners than losers. 28:11 And there were 2 big problems with this. 28:17 One was that I just explained to you why there can't be more winners than losers, 28:20 so I was pretty sure this was wrong. 28:25 The other thing was this was Penguin. Penguin sucked. Penguin was bad. 28:30 How could we have more winners than losers on a day that was bad? 28:34 Well, we did, and I'll tell you how. 28:38 This is a typical—we'll call it a typical Penguin SERP. 28:42 1 through 10 up top we have the day before Penguin, 28:47 so our rankings A through J. What happened the day after Penguin? 28:51 In this case, this typical SERP, C and D, the third and fourth ranked positions, 28:56 they fell out, they fell out of the top 10 completely. 29:00 What happens when number 3 and number 4 fall out? 29:03 Number 5 through number 10 all move up. 29:07 And suddenly there's 2 spots to fill, so 2 more move up. 29:10 So we had 8 winners for every 2 losers in this kind of scenario. 29:14 We had 4 to 1. And you know what? This is just the top 10. 29:18 What if we went to the top 100? 29:22 Maybe there were 98 winners for every 2 losers. 29:24 And this is not an atypical Penguin profile. 29:28 Some sites at the top got slapped hard. But you know what? 29:31 For every site that got slapped hard at the top, everybody else moved up. 29:35 And because we deal with a lot of sites at the top, 29:39 we only heard about the people who complained for the fall. 29:42 When you fall out of number 1, you complain pretty hard. 29:46 When you go from number 10 to number 9 you go, "Hooray!" 29:48 But a lot of people won that day. 29:52 I want to talk about a few events, kind of the evolution of this 29:58 and how we measured. 30:01 That's Ninja Penguin. Ninja Penguin guards the Penguin line. 30:03 [laughter] So we will talk about everything from the Penguin line. 30:08 On June 4th something happened, 30:14 and I was pretty sure something happened except that nobody was talking about it. 30:17 So I thought, "Well, I know there's a lot of error. Maybe I'm crazy." 30:21 And then I saw some chatter and then I talked to the guys at SERPmetrics 30:26 who have been very nice during this process and they said, 30:30 "We saw something too." 30:33 I think it was Greg Boser actually who jokingly— 30:35 it was somebody from BlueGlass; I can't remember—called it the Bigfoot update, 30:39 which ended up being very appropriate. 30:42 And so I wrote a big post about this, but I got to dig in 30:45 and I found a couple things that were interesting. 30:48 This is Bigfoot's footprint. 30:50 Again, this is 1 through 10, but this is a little different than the other visualization. 30:53 The day before Bigfoot here, these are the unique domains across the top 10. 30:57 So the letters and the colors are unique domains. 31:02 The first domain got the top 2 spots, 3 through 10 were 8 unique domains. 31:04 So we had 9 unique domains in the top 10. 31:09 The day after Bigfoot, the top domain got 3 spots up from 2, 31:13 the second domain got 4 spots up from 1. 31:18 So 5 domains in the top 10 now overnight from 9. 31:21 And I started to see this pattern repeat across half a dozen rankings or so. 31:26 But I said, "I've got 1000 rankings, I've got 10,000 URLs." 31:31 "What does it look like across the entire data set?" 31:35 Here's what I saw. 31:40 Prior to Bigfoot, we were running about 5800 total sub-domains across the 10,000 URLs. 31:42 So again, 10 rankings times 1000 keywords. 31:47 After Bigfoot a consistent drop, 2.6% drop in unique sub-domains. 31:51 Something had happened to that domain diversity that seemed to be permanent. 31:57 I didn't graph it out to the right, but it goes on for quite a while. 32:02 And so I kind of thought, "Well, what's normal?" 32:05 5800, 5600? I had no idea what this number should be. 32:08 So if we stretch back to what we had, what happened? 32:13 Something weird happened. 32:18 This is Bigfoot over on the right. 32:20 This is Penguin back on 4/24. 32:23 Prior to Penguin, we had over 6000 total sub-domains in that 10,000. 32:26 There was a massive drop on Penguin, and there was a massive drop later. 32:30 And I'm worried about something. 32:37 A lot happened in June. 32:40 And what I'm concerned about is that you have to remember Penguin was punitive. 32:44 Penguin was not a search quality update. Penguin was a punishment. 32:48 Penguin was Google saying, "Stop doing that," exactly as Greg said today. 32:52 And when you do that, sometimes bad things happen. 32:57 And I think that there were after-effects of Penguin that Google didn't expect. 33:00 And I don't think they did this on purpose. 33:04 I don't think they were trying to crowd domains and reduce diversity. 33:06 Google actually had something in March they called the hoard update— 33:09 it was buried in the quality highlights— 33:11 where they basically said they wanted to reduce crowding, 33:15 they wanted to get more diversity into SERP. 33:17 So they pulled some levers. 33:20 But if you look at the data, it wasn't happening. 33:22 Whatever they keep doing to change something in the algo 33:24 keeps making this go down. 33:26 And I think the lesson here is that there are a lot of unintended consequences 33:28 of the complexity of the algorithm right now. 33:32 When Google does something, 3 other things happen. 33:34 Speaking of diversity, 5 domains dominate the rankings. 33:41 This is data for June. 33:45 Wikipedia, Amazon, YouTube, Facebook, and Twitter, 33:47 10.7% of the total URLs across the 1000 keywords. 33:49 Almost 11% of those 5800. Realize this is 5 of 5800. 33:54 Those 5 dominate. 34:01 A couple other updates I want to talk about. 34:06 We've gotten in a bad habit of looking at the Panda updates as all the same, 3.5, 3.6. 34:08 I have data back to 3.5 now. 34:14 This is Panda 3.7. Crushed the Penguin line. 34:16 Even Ninja Penguin is a little surprised. 34:21 We actually had a weekend rollout, which is extremely rare for Google— 34:24 weekends are usually pretty quiet— 34:27 and what looks like a 5-day rollout. 34:29 Three days had more flux than Penguin. And this was minor. 34:33 This is Panda 3.7. It's supposed to be minor. 34:36 Panda 3.7 was an algo update I will bank anything on. 34:39 And I think it probably should have been Panda 4. 34:44 This was not a data update. Something big happened. 34:46 And I don't know if it was just 3.7 or if Google rolled out more than 1 thing 34:50 or if Google rolled out something and something went very wrong, 34:56 but I think the lesson here— 34:59 Again, I'm afraid. In some ways this isn't very actionable. 35:01 But what I want you to realize is when you hear these things, 35:04 "3.7, oh, that must be smaller than 3." We don't know that. 35:06 We're naming these things and Google is saying, "Okay." 35:10 We say, "Matt, should we call it Panda 3.7?" "Okay. Sure." 35:14 Then he comes back and says, "Well, you know, that one was pretty big, 35:18 that one was small." He never says that, but 10 years later over beers we learn that. 35:23 June 18th something happened. 35:30 Again, it topped Penguin. It took 4 days. 35:33 No one talked about it. No idea. 35:37 Haven't had a chance, unfortunately, to dig in. 35:40 But there are things happening all the time that don't get names 35:43 that we don't know about. And this was a big one. 35:45 This is actually our biggest event on record so far. 35:47 Oh, and Google said that day, "No update." 35:50 Barry did ask. I ask Barry now, so Barry will ask, and Matt said, "Nope. Nothing." 35:54 A couple questions. 36:03 A natural thing to think when you see these multi-day patterns is, 36:05 "Is this just bounce?" 36:07 "Is this something changing and then something changing back?" 36:09 So let's say it's Tuesday. We have a big flux. 36:13 We have a big Delta10 because something changed from Monday. 36:15 Then on Wednesday everything changes back. 36:18 We'd see 2 days of very, very high activity, 36:20 but if we looked at Wednesday versus Monday, nothing changed. 36:23 So I decided to take a look at that. 36:27 This is April 8th, pretty quiet, pre-Penguin. 36:30 And instead of just looking at 1 day Delta10 I stretched it out across the 10 days. 36:35 And what you're seeing here is the algo is not bouncing and bouncing back. 36:39 The algo is permanently changing. 36:44 You look at day 10 and you're up to a day 10 of over 4, 36:47 which would be higher than any event we've recorded. 36:51 It does plateau. It doesn't keep shooting up to infinity. 36:54 But this is not a pattern of your rankings bouncing and then coming back 36:57 and bouncing and coming back like they might have done 5 years ago 37:03 where you just relax and go, "Eh, that's today but tomorrow they'll be fine." 37:05 This is the long-term change and it's rapid long-term change. 37:10 So I'm going to show you a few. 37:14 This is a visualization I made up. It took 4 hours the first time. 37:17 And then I decided to code it, and coding it took 4 hours. 37:22 So 3 slides took me 9 hours, 37:25 and at that point I was like, "God damn it, I don't care if these make any sense at all." 37:27 "They're going in the deck." [laughter] 37:31 So you will appreciate the next 3 slides. [applause and laughter] 37:33 It's much less weird than it sounds. 37:39 Over here—I'm getting too much in one place— 37:41 we have top 10, 1 to 10. 37:44 This was right after Penguin because that 10 days was pretty quiet. 37:46 This is a quiet keyword, biodiesel, for some reason. 37:51 Ten URLs over 10 days. 37:54 So the blue to red is just unique URLs. 37:56 So we move across the 10 days and not much happens. 37:59 The number 3 and number 4 spot kind of swap. That's about it. 38:02 All right. Okay. 38:05 This is a medium. This is kind of an average keyword, car calculator. 38:07 Thirteen URLs over 10. 38:12 So what you see on that fifth day, that's a new URL popping into place in the number 2 spot, 38:14 then it goes away and number 2 comes back. 38:19 You see some shuffle in the 5 to 10, 38:21 but you're not moving a lot. It's not crazy. 38:24 This is a volatile keyword, stocks to buy. 38:29 Look at day 1. It all looks blue. 38:33 That's actually 10 different colors because there are 37 URLs in play over just 10 days. 38:36 And what you're seeing is permanent change. Look at day 10. 38:41 Eight of those URLs weren't even around on day 1. 38:46 In 10 days the profile of this search has changed completely. 38:50 And it's not coming back, it's not bouncing like the slide before. 38:55 If you were here at number 8, you're gone. 39:00 And so if you're in this competitive landscape, 39:02 I want to say this isn't an algo update; 39:05 this is just the day-to-day normal reality of that keyword. 39:07 And what we see here actually is interesting. 39:11 I dug into it a bit. This is a QDF phenomenon. 39:13 This is basically a whole bunch of very large sites with a lot of power 39:16 who write a new article every couple of days. 39:21 And so every couple of days a new URL comes out, 39:24 and they're powerful enough to kick right into the top 10. 39:27 So if you don't have that power, you're not there. 39:30 But if you're tracking those URLs, by the time a week has passed they're gone. 39:34 Everything has changed. 39:38 So what's normal? 39:41 Well, it's not good. 39:43 Over May and June 79.7% of SERPs changed every 24 hours. 39:46 I want to explain that. 39:52 This is not over the course of May and June, 39:54 this is every single fucking day. 39:56 Eighty percent of SERPs have changed. 39:59 This isn't Penguin. This is normal. 40:02 Some changed a lot and some changed a little. 40:05 Maybe you had 2 swap place. 40:08 But that's massive to me, and that's our normal 40:11 when we're trying to figure out what's abnormal. 40:16 This is SEO in 2012. 40:18 If you're not prepared for that rate of change, 40:21 if you're tracking rankings the way you did 5 years ago, 40:23 if you're sitting on 10-year-old content and strategies from 10 years ago 40:27 and just riding your coattails and hoping everything will stay okay, it won't, 40:32 because even if the algo doesn't change, 80% changes every day. 40:35 A couple graphs just for fun. These aren't real important. 40:42 Google busiest day? There's a reason we don't like Mondays. 40:45 Google is pretty busy on Mondays. 40:49 The weekends are quiet. 40:51 The funny thing is, the weekends were really, really bad in June, 40:53 so I ran this analysis last month and the weekends looked really quiet 40:55 and then it changed. 40:59 I was going to tell you to take it easy on the weekend but, no, don't. 41:01 Even that good news, gone. 41:05 And then this is an analysis of the keywords by volume. 41:08 I told you before they're split across these 5 volume bins. 41:11 The highest flux is the medium low keywords 41:14 and the lowest flux is the low keywords. 41:17 If you have any what that means, come up to me after the show and tell me 41:19 because I just made the graph. There you go. 41:23 This is the fun part to me. 41:28 One of the things that's been interesting is to compare these events now. 41:30 These events we all treat as the same. 41:33 So this is kind of a hierarchy of what's happened the last few months. 41:35 Again, Panda 3.7, unnamed event at the top. 41:39 Panda 3.5, 3.6 are way down here. Penguin 1.1 pretty quiet. 41:43 Penguin 1.0 really important. 41:48 3.8 and 3.9 didn't make it on this slide but they're actually right on par with 3.6. 41:51 So we have all these events that we're kind of treating as the same 41:56 and they're not, and we need to be aware of that. 42:00 The one nice thing—the gray is the average— 42:03 all the named events have been above average. 42:05 I wouldn't have put this slide in the deck if that didn't happen 42:08 because that would have been bad. 42:11 And then I had to throw it in. Panda 3.9 hit on the 24th, just a couple days ago. 42:15 How bad was it? 42:21 Well, not very bad. It was around 3.6. 42:23 It was probably a data update. 42:26 What's funny—I didn't get a chance to run it out one more day— 42:28 the day before Panda 3.9 and the day after Panda 3.9 42:31 were actually worse than the day of Panda 3.9. 42:35 And so you start to track these long enough and you start to think, 42:39 "Is Google just fucking with me now or—?" [laughter] 42:41 "Are my numbers that far off?" I don't know. 42:45 But we've had 3 days of flux now and Panda 3.9 was the small one, 42:49 so when they say there's an update coming tonight, 42:52 let's just say I wouldn't completely trust it. 42:58 All right. A Secret Algo Project Launch. 43:02 You might have thought, "Is he going to launch something?" 43:05 Not yet. [laughter] I have a little story to tell first. 43:07 I know this is a question some of you are thinking, and I want you to think it. 43:12 And it's this: Should we be chasing the algorithm? 43:18 We've had some great speakers up here talk about we can't rely on Google. 43:20 We can't have all our eggs in one basket. 43:25 We should be doing more content marketing. 43:28 We should be doing RCS like Wil talked about. 43:30 Greg's talk, same thing today. I absolutely agree with that. 43:32 I don't want you just to chase the algorithm. 43:37 I don't want you to be completely reliant on Google for your business. 43:40 I believe in content marketing. I want you to do good work. 43:44 I want you to build better products. I want you to talk about things that matter to you. 43:48 So why am I doing this? 43:52 Am I just doing this because I'm a data nerd? Well, yeah, sure. 43:53 Am I just doing this because Rand pays me? Yeah. Why not? It's fun. 43:56 But there's 2 other reasons. 44:01 One is that all this great stuff we've talked about, 44:02 I'm sorry to say most of you aren't doing it. 44:07 And when these updates hit, they hit most of you hard. 44:10 And I've seen it this year with clients. 44:14 And I'm not saying they lost a couple keywords. 44:16 I had a client where it was too little, too late. 44:20 I came in very late. Their business went under. 44:22 They were going to lose their house. 44:26 These were nice people. They hadn't done anything shady. 44:28 They had let some things slide. 44:32 They hadn't done the kind of work in the last couple years they should have. 44:36 They used a business model that was getting outdated. 44:40 But they weren't stealing anything, they weren't buying links, 44:43 they weren't doing anything bad. 44:46 And their lives have fallen apart. 44:48 And that's a story I've seen, sadly, more than once. 44:51 And I have a lot of people come to me now and say basically, 44:54 "I'm screwed. Can you help us?" 44:57 And part of it is that. I want you to be prepared. I want to defend you. 45:00 I want you to know what's going on because until you turn things around, 45:05 until you're not completely reliant on Google, you need to know. 45:09 But there's another reason, and this one is a little crazier but I believe this. 45:15 The algorithm is our portal to all of human knowledge right now. 45:21 And you might say, "Well, wait a minute. We have Facebook and Twitter." 45:27 "We have all these ways we can talk to each other and blah, blah, blah." 45:31 But what is the only one collective resource of all of our knowledge so far? 45:34 The biggest thing we have is the Internet as a whole. 45:40 It dwarfs the Library of Congress. It dwarfs anything we have. 45:42 The only engines that really take all that information 45:47 and index it in a way we can view it are the search engines, the major search engines, 45:51 trying to crawl as much of the Web as we can. 45:55 And the Web is the closest thing we have to a repository of all of our knowledge. 45:58 And Google controls 60% of that market. 46:02 Google controls 60% of our access to our own knowledge. 46:06 And this isn't just, "Where do I go get a haircut? What's the best pizza?" 46:11 kind of knowledge; this is everything we know. 46:15 It's how we access information, it's how we shape our opinions. 46:18 We see it with politics. We see more polarizing. 46:23 It's how we understand everything, and it's shaping our culture and it's shaping us. 46:26 And 60% of this is controlled by Google. 46:32 And the algorithm is a filter, and we have to have a filter. 46:35 Please understand I'm not saying Google is bad. 46:39 You can't look at the whole Internet. We're not Neo in The Matrix and, "Oh, hey." 46:41 "I see people in there." We have to have a filter. 46:47 You can't just put it in alphabetical order. The algorithm is essential. 46:50 But the algorithm is controlling how we see all of our information. 46:55 And you have a company that, good or evil, controls most of that. 46:59 And it doesn't matter what their motive is, it doesn't matter if they're good. 47:04 I think that should scare us a little bit. 47:06 And I don't think we think of it that way as SEOs. 47:10 So should we chase the algorithm? Yeah, you're damn right. I think we should chase it. 47:13 I think we should hunt the algorithm. 47:17 I would like the algorithm to wake up screaming in the middle of the night 47:20 in a cold sweat wondering if we're right behind it [applause] 47:23 and saying, "Holy crap! I hope Dr. Pete doesn't get me." [applause continues] 47:27 Because the algorithm, if it gets away from us we're in trouble. 47:32 And I sincerely believe that. 47:37 End of crazy sci-fi rant. 47:41 I'll settle down. 47:46 I was going to launch something an hour ago or so. 47:47 Let's do that. 47:51 [applause and cheering] 47:56 This morning MozCast.com went live. 48:02 MozCast is the Google weather report. 48:06 I decided I shouldn't have all the fun. 48:08 This is a direct translation of our Delta10 metric. We're trying to be very transparent. 48:10 So actually, the weather on any given day, the temperature is just Delta10 times 28. 48:15 An average day is about 70 degrees. 48:21 And just for fun we have these little weather states. 48:24 It's a little weird because it's actually directly dependent on the temperature, 48:26 but sunny, partly cloudy, cloudy, rainy, stormy. 48:29 The stormier it is, the more change. The higher the temperature, the more change. 48:33 So if it's 100 degrees and rainy, stay indoors that day. 48:38 You can check it out later. 48:44 We've got a couple of things. We've got a 5-day history on the left. 48:46 You can't see it. I should have done another graph. 48:49 There's actually a 30-day history down at the bottom. 48:52 You can scroll over anything in 30 days and see down to 1 decimal point, I think. 48:55 The Events page shows all of the major named updates for the last few months 49:01 and gives the severity. 49:05 I want to be able to give you that data where you can see 49:06 that Panda 3.7 was 101 degrees, Panda 3.5 was a pretty cool 68— 49:09 I don't remember what they were— 49:14 because I think that relative view is really important. 49:16 So this is in beta but it's working now. 49:20 This data has been collecting since April 4th, so I feel pretty comfortable. 49:24 I'll be around till Saturday, so talk to me about it. 49:28 We're also launching a new Twitter account, @MozCast, 49:31 so please follow that. I will tweet out the weather in the morning. 49:34 I'm hoping now that things have settled down I can do more analysis when things happen 49:39 and kind of get into some new techniques. 49:43 So it should be fun. And check it out. 49:45 [applause and cheering] 49:48 [female speaker] Was that today's weather that was up there? 49:59 It's always the day before. It's always yesterday. >>That's really cool. 50:03 That slide I did on Monday. If you go to it now, it will be yesterday. 50:07 You know what it reminds me of? 50:10 In California we have a lot of earthquakes, and we have the earthquake graph 50:11 so when it happens you get this chart and it shows where they were 50:15 and how big they were, and everyone can go in and talk about what happened. 50:20 My wife is from Northern California, from the Bay Area, 50:23 and I like how they have to make up weather in California 50:26 because it's like, "It's still 72 and sunny today." 50:29 So it's like, "The wind fluffiness index is 308," 50:33 because otherwise there would be no weather people. 50:38 They'd have nothing to do. 50:40 I'm sure you guys have a lot of questions. Woo! Hands everywhere. 50:42 All right. Let's start back there. 50:46 [male speaker] Hello. 50:50 Hello? I think I speak for everyone, or at least myself, when I say that was amazing. 50:52 I think it was really awesome. [applause] 50:59 One thing I was wondering was for individual sites 51:07 I think it would be really interesting for them to analyze this kind of shift and movement 51:10 over their own set of keywords. 51:14 Do you have any plans to open source the kind of analysis tools 51:16 and certainly the kind of visualization stuff 51:20 to help people analyze their own set of keywords in this manner? 51:23 If you're interested in that, this is very much an aggregate view. 51:28 We're really looking at Google as a whole. 51:32 It might help to dive a little deeper later more into niches and individual sites. 51:35 But if you're interested in that, please come 51:40 for Martin Macdonald's talk tomorrow morning. 51:42 I'll try not to spoil too much of it, 51:44 but a couple months ago Erica sent Martin and I our 2 descriptions 51:46 and we kind of read the first line and we went, "Oh, shit." 51:50 "Are we giving the exact same talk?" 51:53 And then I saw his slides and went, "No, no, no. This is good." 51:55 So Martin is going to show you a very different angle on a similar idea, 51:58 and if you're in a competitive niche or if you're operating a big site 52:02 and you want to collect this data for yourself, stick around for him tomorrow. 52:05 You're going to like it. 52:09 [male speaker] Hi, just a general question too with this is, one, just for clarification, 52:18 is this being run once a day? And then a little tech geek question. 52:25 Are you running this as well the same server 52:31 or are these going across different servers and so the data might fluctuate in that regard? 52:33 [Peter Meyers] It is run once a day. We try to keep it consistent right now. 52:37 We might look at changing that up a bit. 52:41 I don't want Google to catch me. [laughing] 52:43 So we are borrowing the results from them. 52:49 Let me say this: We do use proxies, so we are trying to keep it off of— 52:56 It's on one main server. We try to keep them consistent. 53:01 So each keyword has the same proxy if we can at roughly the same time. 53:04 So within a keyword we keep it consistent, 53:08 but across keywords we use a variety of locations. 53:11 Actually, a couple months ago it dawned on me that this entire thing, 53:15 this weather station, is just a chunk of code and data. 53:20 And so I split it and I actually have a Station A and a Station B now. 53:24 And so I actually run 2 sets of data every day 53:28 and you do see a lot of noise and the numbers are different on any given day, 53:30 but what we see, B runs from a different location. 53:35 I mean, it really is like a weather station in Kansas and a weather station in Los Angeles. 53:38 And what we see is that the pattern of change from day to day 53:43 is pretty consistent across the 2. 53:46 So yeah, there was definitely a lot during the early days of, 53:49 what do we gut check this against? 53:53 Google doesn't tell us anything, so it's this number versus a black box. 53:56 So we are running 2 now concurrently and they're pretty consistent. 54:00 We're trying to improve that as we go. 54:03 And actually, I'm going to use the second one more for fun now, 54:07 shift it 4 hours, see what happens kind of thing, 54:11 because the time shift matters. 54:14 If you go a few hours out, it's a big jump. 54:15 This isn't just 24-hour change. Things are changing every hour. 54:19 Mike is dying over there. I don't have the mic. I'm sorry, man. 54:25 [male speaker] First of all, thank you. It was great. 54:30 We're definitely seeing the Bigfoot update that you've brought up 54:33 in terms of domain diversification. Two questions. 54:36 Why do you think they haven't talked about it, 54:40 or do you think that it maybe hasn't been as volatile as maybe you had seen? 54:42 And then we also have a theory—interested to hear what you think— 54:46 about trying to increase PPC for them by lowering domain. 54:49 You now don't show up, so your only option is to buy into that spot. 54:53 Just your thoughts on the update itself and why. 55:00 It just seems like the user experience is lacking. 55:03 [Peter Meyers] I agree with that. 55:06 I agree that a drop in domain diversity is bad for search quality overall. 55:07 Do I think that Google is doing it on purpose and do I think they're doing it for PPC? 55:12 I don't really think they're doing it for PPC. 55:16 I'm not even sure they're doing it on purpose. 55:19 They seem to be trying to react to it. I think it's a side effect of some other changes. 55:21 And what I suspect this is, we've had a lot of buzz about brand advantage, 55:26 like are brands getting all this advantage 55:29 and is Google purposely giving them an advantage? Big brands. 55:31 I think what's happening is—I wrote about this a few months ago— 55:35 the ranking factors as they evolve are bringing in more offline signals. 55:38 So what did we have originally? We had on-page. 55:43 And with on-page, any Ma and Pa site, if they got the formula right, could rank. 55:45 Then we add links. Links are a little tougher. 55:50 And if you're a big brand who doesn't really do SEO actively, 55:54 you might still attract links. 55:56 And then we had social. 55:59 You could be a big brand who doesn't do any SEO 56:01 and you're still getting mentioned. 56:03 People are talking about you. 56:05 I think Google has to do that to some extent. 56:07 They have to bring these offline factors into the algo 56:09 so that it reflects the real world. 56:12 And the real world is unfair. Brands have an advantage. 56:13 So they have an advantage online. 56:16 So I think we're seeing some of that naturally bleed into domain diversity. 56:18 I don't know. I don't know how much of it is on purpose. 56:23 I'm concerned that it seems to keep happening, 56:25 and I only have 3 months of data, but I'm going to keep tracking that as a variable 56:27 because it does seem to be getting worse. 56:30 I do agree that I think it's a bad user. 56:32 What happens when we're down to 3000 or 4000? 56:35 Should 5 domains have 1 in 10 rankings? That's crazy to me. 56:38 And sure, they own YouTube, and Wikipedia, we all see it in number 1 way too often. 56:44 But yeah, I think it's getting worse. 56:49 I don't know why, and I don't know what to do about it. 56:51 I wish I could turn that into something actionable. 56:52 That's kind of the next step. But what can you do about it? 56:55 Well, you can be a big brand. Be bigger. Be richer. Great. Thanks. 56:57 [male speaker] Thank you, Peter. Over here. 57:05 With your scores, is it a linear scale or is more logarithmic? 57:09 Is a 3.5 considerably double a 3.0 or is it a bit more? 57:15 [Peter Meyers] No. It's essentially linear. 57:20 It's kind of hard to interpret the exact number, 57:22 but yeah, it's not like a page rank or a DA/PA kind of score. >>[male speaker] Got it. 57:25 And in terms of domain diversity, are you looking at the domain name 57:31 or are you also including sub-domains, 57:36 because it seems like a lot of the spammier sites 57:37 are going to have a much wider diversity of sub-domains 57:39 and we're kind of seeing a shift away from that perhaps. 57:42 Those graphs, the 5800 were unique sub-domains. 57:44 So yeah, like an Amazon.com versus, I don't know. 57:49 If it was www versus something else, ShopDot, 57:52 I would count that as 2 in that. 57:55 I'm going to try and break that out going forward, 57:57 but right now that's just sub-domains. >>[male speaker] Thank you. 57:59 [male speaker] Right here. 58:04 Two questions. One is about the domain diversity. 58:06 I wondered what you had to think about with regards to localization of terms. 58:09 And the second part would be just before Bigfoot 58:15 was when Google initiated Google Places to Google Local+. 58:18 [Peter Meyers] Hmm. Yeah. 58:24 These are basically straight organic, 58:26 and we're trying to delocalize them. 58:30 So we basically are looking at Google.com US results delocalized. 58:32 So I'm not really measuring changes in local results. 58:36 [inaudible audience member question] 58:40 Yeah. No. That is possible. Definitely. Yeah, yeah, yeah. 58:46 I've tried to pull out anything that was obviously local, 58:48 but yeah, we definitely did see a shift where Google started to treat things as local 58:50 that weren't necessarily local. I don't have a good answer to that. 58:55 I can't tease that out right now. 58:58 But yeah, we've seen some things like that that could— 59:02 We saw Knowledge Graph rolled out too. 59:04 It shouldn't affect what I'm seeing, but— 59:07 [inaudible audience member comment] >>Oh, oh. I'm sorry. 59:11 The question was basically that before Bigfoot there was the change 59:16 in Google Places and whether this had a tie-in to that, 59:20 but also Google is starting to treat more results as local intent 59:24 that were not local intent, and am I seeing any of that in the results? 59:28 The short answer is it's very possible but I don't know. 59:32 [male speaker] I have a question for you. Back here. 59:37 Since you're tracking all of these different keyword segments and domains 59:40 and you're aware of what's shifting up and down, 59:45 could you take it one step further and monitor the commonalities 59:48 and the differences amongst those shifting domains 59:52 to reverse engineer the algorithm? 59:55 There is a lot of noise and it is kind of hard. 59:59 As we get into a smaller set and say, "Let's look at just 10% or 5% of this," 1:00:01 it's going to get trickier and trickier. 1:00:07 Yeah. I don't know what the next steps are yet. 1:00:11 We're trying to reduce the noise. We're trying to make the system a big bigger. 1:00:14 I think the big question for me right now is when something happens 1:00:18 to kind of build a system of metrics 1:00:21 where I can easily go back and say what happened. 1:00:24 So this domain diversity is one number. 1:00:27 There are probably 5 or 6 numbers like that where if I see a big change 1:00:29 I want to be able to quickly look back a week and say, 1:00:33 "Okay, as we break down in these metrics, what changed?" 1:00:36 I don't have enough data at this point to really dig into, say, a niche, 1:00:39 an industry category or a specific set of sites, 1:00:44 but that could come down the road. We're not there yet. 1:00:47 This actual site has existed for 2 weeks, more or less. [laughing] 1:00:52 The data has been out there a while, and Derek and Casey I have to thank 1:00:56 because they pulled it together quick. 1:01:00 [male speaker] Over here, Pete. 1:01:03 Hi. I had a question about— 1:01:06 You're kind of assuming causation from Google's algo or changes to that. 1:01:09 Is there any plan to include actual site data 1:01:18 that could be impacting their own SERPs rankings, 1:01:23 like maybe using the data that SEOmoz contains from crawling the sites as well? 1:01:27 [Peter Meyers] In terms of a site causing a change themselves? 1:01:35 Yeah. Maybe they gathered more inbound links and those— 1:01:39 [Peter Meyers] Yeah. No. That's a very good point. 1:01:43 That's when we see that 80% shift a day. 1:01:45 That's not just the algo; that's people doing SEO, 1:01:48 that's competitive landscape changing. Absolutely. 1:01:52 Right now that's kind of all in the mix, and that's why we try to take the day-to-day 1:01:56 and try to figure out what that normal number is 1:02:01 because that normal number reflects what's everybody doing. But you're right. 1:02:03 As you dig down into any given site—and this is why your job is so tough. 1:02:07 If you look at my rankings change, just one, well, what did you do? [laughing] 1:02:11 We all want to blame it on the algo. 1:02:17 "Oh, I get 3 calls a week. My rankings tanked. It was Panda." 1:02:19 Well, first of all, you've been tanked since 2009. 1:02:25 I literally had a call like that the other day. 1:02:30 I'm pretty sure that wasn't Panda. 1:02:32 You have 5000 spammy links and you dig in in 5 minutes 1:02:35 and you see, "I don't think the algo is at fault here." 1:02:40 So yeah. No. That's a tough question. 1:02:43 It's an interesting thought that we can tie in some of the site data we have site-to-site 1:02:45 but that's going to be a while. 1:02:49 [female speaker] Question over here. 1:02:52 [female speaker] I guess I kind of wanted to take a little bit of a different approach 1:03:00 to a question. 1:03:04 You talked a bit about the fact that Google is telling us what they want us to hear, 1:03:06 and I kind of see what's happening here 1:03:11 and that it could be maybe 1 of 2 things that's creating a problem for us. 1:03:14 One is that it could be, yes, Google is telling us what they want us to hear, 1:03:19 and for them maybe it's very convenient for everyone to be very, very focused 1:03:25 on Pandas and Penguins 1:03:30 and not really taking any notice of the other stuff that's happening. 1:03:32 But the other side of that is it could just be that we're really being ignorant 1:03:35 and not focusing on the other stuff and kind of zeroing in on things. 1:03:39 So do you have any sense of what that means for us, 1:03:43 how we should look at that? 1:03:49 And also just a little second question, which is maybe not for you, 1:03:51 but did all of this completely freak out the guys at SEOmoz 1:03:54 who are doing the ranking stuff? 1:03:58 [Peter Meyers] I'm sorry. I missed the last part. 1:04:02 Did this all freak out the guys who are actually engineering the rankings tool in PRO 1:04:04 for SEOmoz because it kind of— >>[Peter Meyers] Oh, no, no, no. 1:04:10 I should say actually we don't use any client data. 1:04:12 This is engineered all on its own. 1:04:14 So they kind of let me run by myself. 1:04:17 [female speaker] No, I didn't mean that. 1:04:19 What I meant was that we're getting weekly reports, 1:04:20 and this is kind of saying that a whole lot of stuff is happening, 1:04:24 and does that raise issues for the way that the reporting is being done? 1:04:28 [Peter Meyers] Oh. Well, I don't know yet. [laughing] 1:04:36 I'll find out when I get the calls tomorrow. 1:04:39 But no, your first question, I think that's valid. 1:04:42 Again, I'm not a Google conspiracy guy. 1:04:44 Part of it is our fault, 1:04:47 even if the search quality highlights, for example, 1:04:50 were completely a real effort to be transparent. 1:04:52 And I will say this—that they put a lot of effort into those. 1:04:55 Those do take time to write and do, and they didn't have to do that. 1:04:57 I think it's making us lazy. 1:05:01 And so what should we do? We should stop being lazy. 1:05:05 Look at the media. He said, she said because real reporting costs time and money. 1:05:08 Well, we have to get off our asses and dig into this information ourselves 1:05:13 and not just take this free gift and go, "Thank you," 1:05:17 because I don't know if it's a Trojan horse or if it's not, 1:05:20 but we're not doing our jobs. 1:05:22 I think we have to be more proactive. 1:05:26 [applause and cheering] 1:05:31 Yay! Stand up. [laughing] 1:05:35
You need to sign up for Treehouse in order to download course files.Sign up