Back on June 21st, Matt Cutts replied to a tweet about payday loan spam with an unusual bit of information (reported on Search Engine Roundtable):
The exact timeline was a bit unclear, but Matt seemed to suggest a prolonged algorithm update covering as many as three weeks. Four days later, we tracked our highest temperature ever on MozCast, followed by more record highs:
Seven days during the "multi-week" timeline showed temperature spikes near or above 90°, with six of those days exceeding the severity of the original Penguin update.
Was It A MozCast Glitch?
Let me perfectly honest – Google rankings are a moving target, and tracking day-to-day flux has proven difficult at best. Any given temperature on any given day is prone to error. However, this was a sustained pattern of very high numbers, and we have no evidence to suggest a glitch in the data.
There were some reports that other tools were not showing similar spikes, but some of these reports were based on apples-to-oranges comparisons. For example, if you look at SERPmetrics flux data and isolate just page 1 of Google (which is what MozCast tracks), you'll see this:
Sorry, it's a bit hard to see the dates on the reduced image, but the two spikes equate to roughly June 28th and July 4th, with a smaller bump on June 25th. While they're not an exact match, these two data sets are certainly telling a similar story.
Was It A Large-scale Test?
This is a much harder question to answer. Our beta 10K data set showed similar patterns across multiple C-blocks of IPs, so we have no reason to believe this was specific to one or a very few data centers.
What if Google made a massive change one day, though, and then reverted it? Theoretically, we would see two days of high MozCast temperatures, but if we looked at the two-day flux (instead of two one-day numbers), the temperature would be very low. While this multi-day flux is theoretically interesting, it can be very hard to interpret in practice. Some rankings naturally change, and Google can roll out multiple small updates in any given week.
If we look at the overall flux between the start and end of recorded spikes (June 25 - July 4), we get a MozCast temperature of 120.3°, not much higher than the one-day temperature on June 27th. The average daily temperature for this period was 92.5°. Now, let's look at a similar time period (May 28 - June 6) – the average temperature for that period was 66.8°, and the multi-day temperature across the entire period was 114.7°.
Comparing the two time periods, the overall flux for the period of record temperatures was roughly the same as the peak and about 30% higher than the multi-day average, whereas the overall flux for the quieter period was 72% higher than the average. This is an inexact science at best, and we don't have a good historical sense of multi-day patterns, but my gut feeling is that some of the multi-week update involved changes that Google tested and later rolled back.
What About PMDs & EMDs?
In my post on the June 25th temperature spike, I reported a noticeable single-day drop in partial-match domain (PMD) influence. That post happened very early in the multi-week update, so let's look at the PMD influence data across a 30-day time period that includes all of the high-temperature days:
While there was a lot of movement during this period, you can see that PMDs recovered some of their initial losses around July 4th. The overall trend is downward, but the June 25th drop doesn't appear to have been permanent.
It's interesting to note, even if not directly relevant to this analysis, that the long-term trend for PMD influence in our data is still decidedly downward. Here's a graph back to the beginning of 2013:
So, how have EMDs fared? They seem to show a similar pattern, but in a much tighter range. Scaled to the same Y-axis as the PMD chart above, we get this:
The EMD data is fairly consistent with Dr. Matt Peters' early report on our 2013 Ranking Factors study. Keep in mind that we are measuring two different things – the correlations show how well PMDs/EMDs ranked compared to other domains, whereas MozCast tracks how many PMDs/EMDs ranked across the data set. If the number of total PMDs drops, but they rank roughly as well, the correlations will remain stable, but the "PMD Influence" metric will drop. In other words, the correlations measure how well PMDs rank, whereas MozCast measures how many PMDs rank.
Which PMDs Lost Long-term?
There's one more question we can ask about the drop and subsequent recovery in PMD influence. Did the PMDs that fell out eventually come back, or were they replaced by different PMDs? The metric itself doesn't tell us, but we can dig deeper and see who lost out long-term.
On the initial drop (between June 25-26), 62 PMDs fell out of our public 1K MozCast query set. New PMDs always enter the mix, so the net drop is smaller, but 62 PMDs that were ranking on June 25th weren't ranking on June 26th. So, let's compare that list of 62 to the data on July 5th – after the apparent recovery. On July 5th, 37 of those PMDs (60%) had returned to our data set. This certainly suggests some amount of legitimate recovery.
So, which losing PMDs failed to recover? Here's the complete list (query keywords in parentheses):
- californiacarshows.org (car shows)
- digital-voice-recorder-review.toptenreviews.com (voice recorder)
- fullyramblomatic-yahtzee.blogspot.com (yahtzee)
- virginiamommymakeover.com (mommy makeover)
- www.appliancepartscenter.us (appliance parts)
- www.appliancepartssuppliers.com (appliance parts)
- www.campagnolorestaurant.ca (campagnolo)
- www.campagnolorestaurant.com (campagnolo)
- www.capitalcarshows.com (car shows)
- www.chicagoweddingcandybuffet.com (candy buffet)
- www.dollardrivingschool.com (driving school)
- www.elitedrivingschool.biz (driving school)
- www.etanzanite.com (tanzanite)
- www.firstchoicedrivingschool.net (driving school)
- www.fitzgeraldsdrivingschool.com (driving school)
- www.monogrammedgiftshop.com (monogrammed gifts)
- www.moscatorestaurant.com (moscato)
- www.newjerseyluxuryrealestate.com (luxury real estate)
- www.ocsportscards.com (sports cards)
- www.phoenixbassboats.com (bass boats)
- www.rvsalesofbroward.com (rv sales)
- www.sri-onlineauctions.com (online auctions)
- www.stoltzfusrvs.com (rvs)
- www.vibramdiscgolf.com (vibram)
It's not my goal to pass judgment on the quality of these domains, but simply to provide data for further analysis if anyone is interested. You can see that there are a few examples of multiple PMDs falling out of a single query, suggesting some kind of targeted action.
How Did The Big 10 Do?
In MozCast, we track a metric called the "Big 10" (I did my grad work at U. Iowa, so I should probably have thought twice about that name) – it's just a count of the total percentage of top 10 ranking positions held by the 10 most prominent sites on any given day. Those sites may change day-to-day, but tend to be fairly stable. Looking back to the beginning of 2013, we see a clear upward trend (this graph starts on January 8th, due to a counting issue we had with YouTube results at the beginning of the year):
The "Big 10" gained almost 2-1/2 percentage points in the first half of the year. Some of the gain across the year represents a shuffling of sites in the mix (Twitter falls in and out of the "Big 10", for example, and the root eBay domain struggled earlier this year), and some of this is a symptom of other changes. As Google gets more aggressive about spam, the sites that already dominate naturally tend to take more spots.
I thought it would be interesting to look at these numbers alongside the year-to-date PMD and EMD numbers, but the "Big 10" doesn't seem to tell us much about the multi-week update. As a group, they moved only a fairly small amount between June 25th and July 5th (from 14.97% to 15.17%). Whatever Google tested and rolled out over this period, it didn't dramatically advantage big brands in our data set.
What Happened, Then?
Unfortunately, the patterns just aren't clear, and digging into individual queries that showed the most movement during the multi-week update didn't reveal any general insights. The volatility during this time period seems to have been real, and my best guess is that while some changes stuck, others were made and rolled back. Google may have been doing large-scale testing of algorithm tweaks and refining as they went, but at this point the exact nature of those changes is unclear. Between the multi-week update and Google's announcement of 10-day Panda roll-outs, it appears that we're going to see more prolonged updates. Whether this is to mitigate the impact of one-day updates or make the update process more opaque is anyone's guess.
Generalisation alert:
Regarding the EMD and PMD, with their being a general downwards trend and not anything overly dramatic - I don't think there's been any update surrounding those kind of domains specifically.
PMDs and EMDs, quite generally, are pretty low in site quality (generalisation over). Furthermore, the domains that the doc provided look pretty poor to me, both on-site and off-site.
The fact that those sites were ranking to begin with shows that PMDs and EMDs still carry some significance in my eyes - but the "magic" effect they have can quite quickly vanish if the site isn't up to scratch. It all comes back to building quality, user-friendly and useful sites with some strong links that have been acquired in the right way. If you do that, your PMD and EMD will be fine.
Thanks for the report, doc - really interesting analysis as ever.
Yeah, other than last year's "EMD Update", I think Google has realized they can't just massively devalue all EMDs/PMDs without a lot of collateral damage. So, I suspect they're looking at low-quality signals that typical correlate with low-quality EMDs/PMDs. As they weed out more spam, more PMDs naturally disappear, but the ones that remain still generally rank well. EMDs/PMDs aren't bad - Bad PMDs are bad, basically :)
during these changes, i saw (and still see) a table designed EMD with several onpage errors and redirection problems and duplicate content (exact copy on 3 other EMD for other keywords) jumpin from P5 to P1 and the site still is on P1 (the side has some backlinks, but not that much and not that good).
The P2 now - before that it was on one over 6-8 weeks (not that sure) is a perfect On and Offpage optimized Side was fallen down to 5 and came back to P2 this week. And thats not an PMD/EMD.
I really cant figure out why - ok P2 isnt my site - but both are my competitors - I just went 2 positions up - thats all. And the normal movement from P4 to P9 (it was a real strong mixing all 3 days over weeks) seems to be stoped. I cant figure out what that was/is
Pretty interesting if you start looking at the links for some of the PMDs that didn't recover:
California Car Shows has links from link pages on sites that look like FrontPage-generated 1990s nightmares, poker sites, etc.
VirginiaMommyMakover has links from design directories, CozyNursingBras.com and 'Discreetbreasts.com.' Not making this up.
As you dig lower into the profiles of the linking sites, it starts to look like a river in Elizabeth, NJ: Turgid, toxic and kinda stinky.
So they may have lost some rankings oomph from the PMD update, but they clearly got hit for manipulative/low-quality linking, too. That may be why they're not recovering.
Yeah, I think the PMD part is mostly a coincidence. There are definitely some low-quality patterns behind some of these sites. My gut feeling is that Google tried a few things, found that some of them caused too much collateral damage, and backed off. Some of it stuck, though.
Very interesting data; great post as always, Dr. Pete!
Not sure if the prominence of "big 10" is good or bad - it could be good because it means google is catching more lower-quality sites, but the beauty of google was that you could find good quality content on websites you didn't know about. If google's going to give you the websites you already know, then what's the use of it?
It's really hard to separate cause and effect. Some people will rush to say this shows Google is favoring big brands, and I think Google is certainly sensitive to brands (and making money). On the other hand, if you weed out a lot of spam, then the big sites naturally fill in some of the holes left behind, and diversity suffers. This isn't necessarily Google's intent, but it's a natural consequence of removing low-quality content, even if they always got it right.
Great insights.
I have been doing some digging of my own, and even though I don't have enough data to call it for sure, I am seeing one strange thing. Many of the domains that where reported as "dropping out" in the beginning of this update (according to Searchmetrics for instance) seemed to have come out of nowhere to rank very highly just before the update.
Almost as if Google first removed some filters only to replace them a few days (weeks?) later. So that sites that where filtered out before, first came back and took high rankings for a short period, and then where hit again.
californiacarshows.org for instance seems to be one of those.
Any thoughts on this?
Interesting... it's hard to track that at large-scale, but for that URL I see it appearing and disappearing periodically. For example, it appeared on 6/24 and dropped out on 6/25 (after not ranking since 6/15), but it did rank the first half of June. Part of the problem is that it's hovering between about #6-#8 most days, and we only track the Top 10 (so, if it fell to #11, it would look like a disappearing act). Interestingly, that domain reappeared on 7/13 and ranked solidly around #8 until 7/20. Google is definitely tweaking things.
Saw another major movie in EMDs on Friday, but then it partially recovered on Saturday. I'm gonna go drink more.
Its good to see EMD's are consistently decreasing in influence. I am still seeing poor quality EMDs out ranking high quality websites in local search terms which is frustrating.
I noticed the following EMDs are in the top first 4 results in the SERP for the keyword
Packers and Movers in Gurgaon
packersmoversgurgaon.3th.co.in
packersandmoversingurgaon.8th.co.in/
www.packersandmoversgurgaon.5th.co.in
Wow, that's all kinds of quality. I've been looking for a list of the "Top 3th Companies" for a while :)
I'm sure near the holidays we are going to see the next big post like this from Dr. Pete. If you have been effected then go make your site better now. :)
Pete - thanks for doing all the digging in the data. My issue I have with PMD's is that using words that describe what you do has been a business naming convention for decades, so Google are making it harder to rank if you have keywords in your business name (often the case with PMDs) while promoting big brands/names in their place.
Without sizeable marketing budgets would you agree Google is making it harder for newer sites to rank?
I don't like to prognosticate too much, but I feel fairly confident saying that Google will probably never do an update that penalizes or devalues EMDs simply for being EMDs. Too many brands (big and small) and legitimate sites have perfectly reasonable EMDs that make sense for them to have.
If you're Martin's Pizza, you have a legitimate business that customers like and talk about, and you own martinspizza.com, you will probably be fine for the foreseeable future. If, on the other hand, you register buyapizzaonline.com, no one knows you buy that name, you have no co-citations or brand mentions, you don't have a physical presences, you don't advertise under that name, etc., then I think your odds of ranking well for your exact-match terms are going to continue to decline.
I'm not saying that it never works (even in 2013) - just that using an EMD or PMD as your primary tactic is going to get harder and harder.
This is pretty interesting. Thanks!
Thanks Dr. Peter for the update. I have heard that "payday loan" SERP is controlled by mafia in UK.
The recent updates are far beyond Panda & Penguin , as Google is trying to locate new quality signals to organize search results. They are moving closer to a "Semantic Model" rather than the "Classical One" based on Content, links, etc.
I was wondering what actual changes happened in the algorithms causing such a high temperature spike.
Great analysis as always, Pete! :)
The recent updates are far beyond Panda & Penguin , as Google is trying to locate new quality signals to organize search results.
The development of algorithms makes lass sites go up and down positions quickly, and this will be worse in the future. My website minicreditos online in a week has dropped 30 positions to 25 positions back up to 1 day.
[link removed -km]
Your graphs worry me, at what point in time will sites 1through10 (or 1through7) simply be a rundown of those with the biggest pockets
Very interesting data. This is a really awesome examination into the probable algorithm update, and a latent partial-match domain (PMD) connection.
Thanks for sharing one more informative analysis.
Yes I also feel some up and down for sites ranking from last few weeks, and really god bless to Google these changes are really positive for my site SEO Agra and I really want these kind of more good multi week algorithm. Thanks.
What was wrong on friday? 105° !!
Dr-Pete ??
Emds and pmds are penalized due to keywords stuffing and lot of exact anchors text I have rank few emds lately using relatively less exact anchors vs natural anchors
Agree about EMDs, if EMDs did not work then nobody would rank for there own brand
Great work Dr Pete thanks for sharing this great analytic about EMD and PMD.
Thanks from the analysis Dr. Pete! Really helpful as usual.
If this is the case, it appears that there is a correlation between keyword-agnostic features and rankings
We are constantly getting push down by the big brands
I think EMD/PMD all depends on quality. It depends on quality of backlinks of that particular website.
Vijay & Dr Pete, You can found the same result with keywords: Packers and Movers Noida & Packers and Movers Pune
Keyword: Packers and Movers Noida
packersandmoversnoida6.webnode.in/
packers-and-movers-noida.webnode.in
packersmoversnoida.3th.co.in/
packersmoversnoida.8th.in/
Keyword: Packers and Movers Pune
packersmoverspune.6th.in/
packersandmoverspune.5th.co.in
packersandmoversinpune.8th.co.in/
packersmoverspune.8th.in/
Thank for an insightful post. Good to see all the algorithm updates stuff here. It is really an useful post for all the webmaster in order to optimize their websites in better way using all the latest techniques as per the Google.
I love this kind of data... and the coments - thanks all :-)
Matt Cutts tweets regarding the issue true depict the whole scene in true sense because the facts and figures are also defining the same thing. PMD and EMD influence graphs also explain the patterns and how other factors influence on the graph pattern. Good attempt indeed!
I have done a check this morning for same local keyword in different cities in the UK and out of page 1 of the SERPs, a minimum of 3 have EMDs
Thanks for this post, Dr. Pete. One question for you: We've seen about a 50% drop in our bounce rate in Google Analytics since the first of July. Could these algorithm updates somehow affect the bounce rate? I've not heard anything about Google redefining what constitutes a bounce, so I'm assuming somehow different people are finding our site through Google and they're sticking around longer, but that's pretty much a complete guess. Any thoughts? It's one of the stranger things I've seen in our analytics b/c there wasn't anything gradual about it, it just dropped 50% on July 1 and has stayed around the July 1 number ever since.
The only reason I can think of that an algo update would impact your bounce rate is if you started ranking for a different page. In other words, for "keyword x" your rank-tracking shows you in #3 before and after the update, but it turns out that the position is held by a new page, and that new page is a much less appropriate match. Otherwise, it's more likely a GA glitch or implementation issue. It's really tough to say without digging deep into it.
Cool, thanks for taking the time to reply.
I saw a few posts bashing MozCast but I think its a lovely data set to look at:)
Something did happen during those weeks were I saw twitter jumping to 1st for a large amount brand searchs while brand sites dropped to 2nd.
Gotta love when you get hit with a dozen and two things at once. It makes it so much more fun to figure out what is actually going on. Who doesn't live testing and retesting and not knowing for weeks if it worked or not?
Thanks for sharing yet another great analysis Dr. Pete!
"Payday Loans on co.uk" reminds me of this video by Matt Cutts done in May, on what kind of things are in Google's pipeline where he spoke about cleaning the spam SERPs for this keyword and a lot of other changes on the radar.
Just thought of sharing it here for those who haven't seen yet.
KAS
Hey Dr. Pete,
I think your observation related to PMD and EMD are correct. But if you have initialized then these days cached date of many sites are going back and that ultimately loss in ranking. I have seen this with many of my clients and when Google crawls it back ranking are back but for the time being rankings are down in SERP. So, surely there is some algorithmic updates are going in Google.
One more thing I have seen that Google is now giving more importance to local search queries and local listing sites. As nowadays changes are more seen in location specific search queries. So, I think there is some changes going in the local search algorithm. What you all say about this?
Thanks for sharing.
Stephen :)
Didnt saw that here, but i didnt really watched the values :)
I take a look at these values and than write back in a few days...
Thanks waiting for your reply :)
Most of the locals of my clients are still stable. but a few are struggeling now. so I have to thank for that hint and I have to see whats wrong with them. 4 out of 175 locals have now enorm impression loss
Awesome post Dr Pete! I really feel like the Domain Crowding update is having continued and progressive effect on the SERPs. Across all of my clients website we have seen incredible flux with their total results while their top results have held consistent. What are your thoughts, do you feel like their is any connection and perhaps it was tied to the multi-week roll out indicated by Cutts?
Thanks for this data report. Are you/how are you measuring flux related to the 85+ SERP features you overviewed at MozCon? We know that results with site links or local intent are taking up a lot more screen than the old school 10-link view, but what about carousel, marketplace, knowledge graph, etc? At this point, isn't it a much safer bet to just print your entire website and mail it to your consumers? That way at least you know you got in their face for a split second.
Of course I'm kidding. But for real, how are you adjusting mozcast for SERP evolution?
In terms of core temperature/flux, we're not adjusting it for SERP features. This was essentially the topic of my MozCon talk, and is definitely something I'm very interested in. We're looking to launch a "feature graph" that explores the evolving SERP landscape in more detail (on MozCast.com), and are exploring this reality in other tools as well. We have a secondary set of flux metrics that weights the temperature by ranking positions, but the truth is that the overall patterns we see are usually very similar.
Hi Dr-Pete,
Please can you define the term 'keyword-agnostic features' in your pie chart summary.
Presumably this means keywords not related to the search performed in your study?
If this is the case, it appears that there is a correlation between keyword-agnostic features and rankings, albeit less than the use of targeted keywords?
How can this be, or have I simply misunderstood your term?
Was this a reply to one of Dr. Matt's posts? I think he uses "keyword agnostic features" in the ranking factors survey, but I don't mention it here (and there's no pie chart in this post).
your done keyword agnostic features in the ramking ......
yoga teacher training los angeles
Thanks for sharing one more informative analysis, Very interesting data
Good work.
Hope all these efforts made by Google web spam team will lead to a spam free SERPs. If someone can make an analysis report on how far the goal of spam free SERPs is already achieved by strict updations conducting during last few years.
Looks like the the big brands always win with Google while smaller ones just survive the SEO war.
We are constantly getting push down by the big brands. We are still high in the the serps, but constantly need to improve content just to survive. The big brands virtually make no changes to these landing pages. We also started to see a trend where more long-tail kw's are getting eaten up by them as well.
Great stuff Dr Pete (as always)!! EMD are still on the decline. We will have to stay tuned as changes continue to roll out...
Hello Pete,
Thanks for the much awaited analysis. I think the PMD update is not just regarding authority or spam link structure. For this, I checked the two sites 'www.newjerseyluxuryrealestate.com' & 'virginiamommymakeover.com' for the keywords you mentioned in your list, I found that they rank first for EMD (luxury real estate new jersey & mommy makeover virginia respectively). So, EMD can't be a penalty kind as against common perception. I think Google checks for stronger signal regarding how closely you satisfy the keyword used in the domain or may be they follow the history of users' click play around the keyword. Keywords like 'luxury real estate' & 'mommy makeover' are more general in sense to direct the users for such keywords to location specific domains. Need more analysis on this but advice from you in this regard will be of great help in right direction.
Excellent overview! Thank you for getting into the details, which is something you excel at Dr. Pete! ;-)
We have a client who was hit hard on 6/26, and it appears to be a brand related update for a number of queries. This site's impressions in Google Webmaster Tools, their rankings, and subsequent traffic dropped dramatically for product terms that included that product's respective manufacturer name. The brands are now dominating 2-3 SERP listings for these queries, and they were not that dominant before. Interestingly, when evaluating DA and number of links, the brand pages don't have the same strength. Google is definitely looking at different search quality signals than prior to that date.
*Sigh* Google is making it quite difficult to reverse engineer the influential factors of their algorithm updates. So far, no one on my team has reported seeing any drastic drops during this multi-week algo update. Can't wait to see what the ranking reports reflect next week ...
Interesting to see how EMDs and PMDs are becoming harder and harder everyday.
Thanks for updating this deep analysis for the webmasters. Without these regularly updates we can not survive in SEO world.