At Moz, we have been following up on our 2013 Search Engine Ranking Factors study by continuing to analyze interesting aspects of the data. One of our most frequently asked questions is, "Do you see any systematic differences in Google's search results across search volume or topic category?" By design, our main study used a broad keyword set across all search volumes and industries to capture Google's overall search algorithm. As a result, we weren't able to answer this question since it requires segmenting the data into different buckets. In this post, I'll do just that and dig into the data in an attempt to answer this question.
Our approach
We used a subset of the data from our 2013 Ranking Factors study, focusing on a few of the most important factors. In the main study, we collected the top 50 search results for about 15,000 keywords from Google, along with more then 100 different factors. These included links, anchor text, on-page factors, and social signals, among others. Then, for each factor we computed the mean Spearman correlation between the factor and search position. Here's a great graphic from Rand that helps illustrate how to interpret the correlations:
In general, a higher correlation means that the factor is more closely related to a higher ranking than a lower correlation. It doesn't necessarily mean that there is causation!
In addition to search results and factors, we collected the categories from AdWords (e.g. "Home and Garden") and the monthly US (local) search volume. This allows us to examine correlations across these different segments.
Search volume
First up is search volume. We segmented each keyword into one of three buckets depending on the average local (US) monthly search volume from AdWords: less than 5,000 searches per month, 5,000-15,000 searches per month, and more than 15,000 searches per month.
To begin exploring the data, here is the median page and domain authority in each bucket, along with the total percentage of results with a domain name exactly matching the keyword:
Not too surprisingly, we see the overall page authority, domain authority and the exact match domain (EMD) percentage all increase with search volume. This is presumably because higher-volume queries are targeted by larger, more authoritative sites.
Now, an overall higher page authority for high-volume queries doesn't necessarily mean that the correlation with search position will be larger. The correlation measures the extent to which page authority (or any other factor) can predict the ordering. As a example, consider two three-result SERPs, one with page authorities of 90, 92, and 88 for the first three positions; and another with values of 30, 20, and 10. The first SERP has higher values overall, but a lower correlation. To examine how these impact search ordering, we can compute the mean Spearman correlation in each bucket:
And for those who prefer a chart:
From left to right, the table lists link-related factors (page authority, domain authority, and exact match anchor text); a brand-related factor (number of domain mentions in the last 30 days from Fresh Web Explorer); social factors (number of Google +1s, Facebook shares, and tweets); and keyword-related factors (keyword usage on the page, in the title, and EMD).
Looking at the data, we can see a few interesting things:
- The correlations increase noticeably with search volume for link, brand, and social media factors.
- The correlations are mostly constant for keyword-related factors (keyword usage on the page or in the domain name).
One implication of point #2 is that Google's keyword-document relevance algorithm is the same for high- and low-volume queries. That is, their method for determining what a page is about doesn't depend the query popularity.
We can make this more concrete by considering two different queries and SERPs: one high volume ("cheap flights" with more than 1 million searches per month), and one low-volume ("home goods online" with less than 500 searches per month). For reference, here are the top results for each search, with the page and domain authority from the MozBar:
Above: Google SERP for "cheap flights"
Above: Google SERP for "home goods online"
When a user enters a query, Google first determines which of the many pages in its index are relevant to the query, then ranks the results. A popular query will likely have several relevant pages (or more) with many links, since they are targeted by marketers. In this case, Google should have plenty of signals to determine ranking. A relevant page with high page authority? Check, put it in the top 10. On the other hand, pages in the dark corners of the internet with relatively few links are likely most relevant to low-volume queries. In the low-volume case, since the link signals aren't as clear, Google is forced to rely more heavily on other signals to determine ranking, and the correlations decrease. This example oversimplifies the complexity of the algorithm, but provides some intuitive understanding of the data.
Site category
We can repeat the analysis for the different AdWords categories. First, the median page and domain authority and EMD percentage:
And the mean Spearman correlations:
Overall, the trends are similar to search volume, with significant differences in the link correlations, and smaller differences in the keyword-related correlations. The explanation for these results is similar to the one above for search volume. The industries with the largest link and social correlations — "Health" and "Travel & Tourism" — tend to have broad-based queries targeted by lots of sites. On the other hand, the industries near the bottom of the table — "Apparel," "Dining & Nightlife," and "Retailers & General Merchandise" — all tend to have specific or local intent queries that are likely to be relevant to specific product pages or smaller sites.
Takeaways
In this post, we have explored how a few individual ranking factors vary across search volume and keyword category. Correlations of link- and social-related metrics increase with search volume, but correlations of keyword-related factors (usage on page and in the domain name) are constant across search volume. Taken together, this suggests that Google is using the same query document relevance algorithm for both head and tail queries, but that link metrics predict SERPs from popular queries better then tail queries. We see something similar across site categories with the largest differences in link related correlations. Industries like "Health" that have broad, informational queries have higher correlations than industries like "Apparel" that tend to have queries with specific product intent.
Awesome stuff Matt, especially the Site Category analysis, quite something to think about.
One thing I want to ask regarding correlation of PA and DA to the SERPs. Did you factor in the fact that both PA and DA are on a logarithmic scale, and thus a lot more likely to correlate at higher values?
Good question -- I did factor that in. I used the median instead of the mean PA/DA in the two summary tables since the mean might be skewed by the logarithmic nature of these metrics.
This is a great and unique way to look at this, congratulations, that's great research. When it comes to some of the conclusions, though, I am not sure I follow you:
1. Higher EMD because more authoritative sites - why do you think that? I would have expected the opposite. Higher authority often comes with brands, and then the brand is the domain name, but less likely a product or service, so I understand DA, but not EMD through this. I'd rather suspect smaller seo optimized pages behind this, skimming a niche or particular market.
2. I don't see why travel is likely to have less local search aspect to it than apparel for example, it might be just a different location than the one you're in when searching. Also - are retailer brands not quite large, like Amazon? Without seeing the majority of keywords in each section this is hard to tell. Stories as examples are great, but we need to know if they show a good representation.
3.Why the assumption that 'dark corners' of the internet with few links are most relevant to low volume searches? Long tail is VERY important for large sites as well. (I work for Dell, trust me on this one :-)) So, large sites are as relevant - if not more - for low volume searches as well.
Perhaps I am just missing insight into the detailed data which you have in your research? Have you thought to adjust this to some extent with the number of total keywords in each industry, or even better the distribution head/longtail per each?
Related to your EMD doubt, I suggest you to read what Matt wrote in the first post presenting the Ranking Factors 2013.
In that post (as well in its session at MozCon) he was noticing how EMD influence effectively decreased after the EMD Update, but tended to rise again (not at the same level) after few months.
That was due to the fact that the update punished the crap EMDs, but not the concept itself of EMD, if it is sustained by other value, which we can resume in "the brandization of the EMD itself".
For instance, in Tourism, Booking,com is an EMD, but it is still über strong in the SERPs.
I don't understand what your comment means to my comment:
I don't doubt the concept of EMD or of the trends in the relevance of EMD, but I don't see the basis for the statement that EMD domains tend to be among the high authority sites.
Mmm... maybe I wasn't able to explain myself well, and I beg you pardon for that.
What I was meaning is that Google has always given a plus to domain having integrally or partially the targeted keyword in the domain name, and that didn't change with the EMD update, which simply tried to correct the EMD by itself, without any other signals (as can be backlinks from authoritative sites, co-citations, co-occurences and, on secondary aspect, social signals) could outrank more authoritative sites.
Therefore, it is still true that EMD that become the expression of a Brand (from there my example of Booking) weren't affected by that update; so EMD plus good DA are still possible. And cases like Booking are not so rare in many niches.
Hi Sir Andreas,
Please see if the statements below answered your questions:
1. EMD will have a higher recognition if the contents in your site are highly relevant, correlated, and compliments each and every pages in your domain. Higher authority often comes with brands, yes, that's correct. But you can't say that it is less likely a product or service. I have served an online marketing provider with over 1000 clients (SEO resellers, Gurus, and SEO Companies) and your statement is not what I saw in our day to day operations.
In addition, Matt seems to use a non-branded terms, so this research is acceptable. The research is referring to SEO and not branding, sir Andreas.wpv.
2. Probably because social media advertising and promotions dramatically influenced the number of search for travel for the last two years.
3. Its simply because most of the people, based on their objectives, expectations, and business goals, will target and prioritize the most relative and exact (service or product) words rather than long tail keywords. Yes, long tail keywords are very important. For me, a scale of long-tail keywords VS generic to localized keywords is 2:8.
Hi Andreas,
Thanks for your questions and glad you enjoyed the post :-) I'll do my best to answer them.
For #1, I don't necessarily think that EMD is associated with high authority sites, only that both EMD and high authority sites preferentially target high volume keywords. For #2, mintradz above gave a good answer. For #3: Agree completely that long tail keywords are important for large sites, especially since they have a lot of content to target them. My hypothesis for why low volume SERP results have lower DA goes something like this: (and this is just a hypothesis until tested): (1) large sites with high DA are still only a small percentage of the entire internet (2) there are so many rare, long tail queries that they target essentially random content so (3) they will on average return lower DA domains.
Very well exemplified, very eloquent. I appreciate this article a lot.
Thank you for sharing this post with us!
Hi Matt, all of this math is confusing me :(
What are your suggestions for pages and domains targeting long-tail keywords and key phrases, if link metrics predict SERPS from popular queries better than tail queries?
Good
Well.Its super like Thanks Peters. Its nice analysis.
Correlations are important and the linking strategy is also most important.
You can see Wikipedia is strong because they are good with content as well linking.
Regards Denish Verma SEO Manager DnAWebSolution.us
Matt@
Even i was wondering by getting some search results displayed by Google, i assumed that Google is manipulating the search results just to increase their paid services (PPC), i did a research to solve my problem, i analyzed lots of websites, those who were in Top position and who were just below them. I didn't get the answer, as some Facebook Page, twitter accounts and LinkedIn account were on top position on some highly competitive keywords, while those website who deserve the top ranking were suffering. Now after getting your, i am hoping that i will solve my problem and will get the exact solution. Thanks for sharing such type of brilliant information.
The keywords or key phrases the searcher asked for are the only commonality between organic search results and paid search results. AdWords isn't a factor in SEO.
John,
I am not saying that Adwords is a factor in SEO, i am simply sharing my experiences that how ranking is getting manipulations, as we can easily findout that some Facebook Pages, Twitter Accounts and Linked Accounts are on top 10 position on highly competitive keywords. I was very curious to know that How these social accounts are getting rank ? I only got to know the fact that these are the part of manipulated rankings .. thats it.
Thank you for the research, I like a few others who have previously commented would like to know the direct effect of social signals vs. DA vs. PA in other various search terms including long(er) tail keywords.
Your contribution was terrific though :)
Great post, somehow the perfect recipe is still the same, add some exact match or partial match and some fresh content and you´ll be able to beat any site. Hope to see more influence in PA and DA in future Google´s algorithm as it is somehow an unfair ranking.
Great post. Very well done!
Question:
You refer to large-volume queries over low-volume queries yet you don't specifically speak to "general" versus "specific" or long-tail as being a potential factor. Though it is normally true that general queries result in higher search volumes while contrarily long-tails result in lower volumes. Google attempts to provide results based on the level of generality/specificity of the query - which would, in most cases, result in the search volume data that you've referred to; however generality/specificity targeting is different and opens up a whole new can of worms when considered as a signal instead of search volume.
How do your results differ when the generality of the term versus the generality of the result (and specificity of each) are taken into consideration?
Thanks and best
Great post Matt! It always intrigues me when someone starts to reverse engineer the algo. I had a few thoughts regarding the post. Some are just rapping philosophical, some are about application.
It would seem to me that EMD and page level metrics are a bigger deal at the smaller query level simply because there is less links, social media, etc. to go on. If Google was trying to rank some query like "small business accounting memphis tn 38122" there will probably be less links, social shares, etc than "small business accounting software" simply because it's a smaller market. So those on page factors by necessity would have a bigger piece of the pie.
Now a question as to actual application, I realize that you analyzed this at a site category and domain name level. However, seeing as how changing your domain name to chase a smaller portion of the market won't make sense for most businesses who have already invested in serious SEO, what measure do you think the rest of the URL has on the rankings for the content focused smaller queries with no links,social shares, etc.? i.e. bigsite.com/exact-match-query-with-small-search-traffic vs. bigsite.com/some-almost-relevant-title
Hey Matt. I second the guys above asking about how social shares affect a site's DA/PA. If you ever have any information or studies about that, I'd appreciate it if you'll share it. More power to you !
That was a lovely post. Thanks for putting this upa
In a perfect world SEO would never change lol! Look at what has happened the past 5 years though!
There are many interesting articles in this blog, this is one of them, I hope to continue learning more things in moz and investigate to complement some details, excellent post
Nice post but would be better if you pointed out metrics in a column - so i wouldnt have to read whole article to just see metrics and speculations about them. Of course there is a pic but i like ot have things on plate ;)
Good article Guys, Thanks. By the way I found this awesome site for Financial Data and Analysis : https://thinknum.blogspot.com/ You can Search over 7,000,000 datasets. It has a <a href"https://thinknum.blogspot.com/">financial data analysis</a>engine which brings the functionality traditionally found on wall street proprietary trading desks to an open platform. You can visualize, save and share data. This tool is similar to PlotTool at Goldman Sachs or DataQuery at JP Morgan. Check it out, You may like it!
Thanks for the good information.
Must say, in Italy EMD domain continue to be rather strong in serp, even if site is not high quality made. Is it the same in your area?
OK... that seems strange that the algo would somehow work out diff (ranking factors), or is it just that when analysing we concentrated on those? Needs more investigation, I cant believe th algo works differently
Good Research, thanks for sharing with us. Recently came to know about some good points like search engine results page.
5000 queries per month is still a fairly high number (I presume we are talking exact match), and although it is not what I would call a major term, I wouldn't really call it long tail either. It would be interesting to see what happens when you get into the real long tail where search volume is < 100 month. My guess would be that you would get a much higher correlation with EMD's.
That's a good question. Unfortunately we don't have enough data at really low volume to answer it.
Empowering post. I also really like the new page stats link at the bottom of the post. Very cool
thanks a lot for the analysis..
Such a good post, Matt. Thanks for your time and energy.
Thnx! Great post will use it sell my clients on the idea of content creation to boost social sharing and likes and natural link building. For most of the shot tail, root keywords that I target on page SEO has barely moved the ranking needle.
This is superb stuff, Matt. Really interesting data sets.
It's clear that links are still, and will be for some time, the strongest ranking factor - that is if we consider correlation to imply causation! (however, I don't think many would disagree on this one). What I would love to see is the effect that social signals have on rankings in both the short and long-term. I've noticed that with most emerging news, social shares seems to be a key factor for quick rankings and then unless they are supplemented with solid links, the rankings decay over time - if you're doing (or have done) any experiments around this then I would LOVE to take a peak!
...I've noticed that with most emerging news, social shares seems to be a key factor for quick rankings and then unless they are supplemented with solid links, the rankings decay over time...
What you are describing is Freshness, which is biased by the Social Graph a lot.
You've done your damn research Matt! In terms of DA/PA is it affected by social shares (I know they display on the right) but have always been interested and never gotten an answer? Otherwise, it seems the correlation between links and rankings are still super strong! I saw this post about how links are still having the most dominant effect, though it's harder to build links now that'll have a really good influence after penguin etc.. I saw in the Moz ranking factors that links had taken a bit of a dive but social was up, yet I've had some great content that's gotten some really good shares in terms of content yet it doesn't rank nearly as close to the 1st page stuff I get on the links built to posts I do. I also found it slightly strange that the computers section average DA is so much lower than the Health industry, I would of presumed that to be #1 as it'd be technical guys writing about it who'd have SEO experience.
GodofSEO- I second your questions. I'd love to see some real case study data on social shares correlation or causation for rankings vs links, or combined with links.
GREAT post Matt.
I also found it slightly strange that the computers section average DA is so much lower than the Health industry, I would of presumed that to be #1 as it'd be technical guys writing about it who'd have SEO experience.
<sarcasm>Knowing how median Hacker News folks tend to hate SEO, the DA in Computers Section seem correct</sarcasm>
Remember also the computer is not the same than online marketing as topic, and that online marketing is represented by SEO just a possibly minority percentage... hence, we should not consider SEO as the center of the universe, therefore that all the sites are doing what SEOs are preaching; to tell the truth is usually the contrary.
Glad you enjoyed the post! To answer you question about DA/PA: no, we don't currently use social information in DA or PA. These are based solely on in-link metrics.
We hope to spend some time in the future looking at the social shares correlation vs causation question, stay tuned!
Great Post!! Thanks for sharing the data... Awesome site analysis report... really an informative post. Thank you once again.
Hey Matt, great article! Would the algorithyms associated with on page factors which include time on page, bounce rate, etc be included in the PA factors? Once again, thanks for the post!
No we don't use bounce rate or time on site in PA or DA, only in-links to the page or domain.
Great post!
Thanks a lot, its helping me :)
Such a great Post Matt !
Websites which have keywords in starting of Title ranking high. category Analysis is superb by you.
Good but not better , it may be better then this. thanks
He guys great post, could anyone point me in the direction of where to find
**Total National (Australian based web/ Google searches, broken down by state or capital city). I'm basically looking to break down the % of search per city vs the nation. Any information is appreciated.
Interesting that EMD correlates higher with higher-volume search terms. I wonder if this is simply because any given domain name is more likely to use high volume search terms than low volume search terms, so they are naturally going to make up a larger percentage of the results. Obviously not every domain name uses high volume terms, but pretty much any high volume term is going to be in lots of domain names.
Yea, that's my thinking too. If someone is going to create an EMD to target a specific query, then it is more likely to be a high volume one.
Thanks for the research you have made for yourself and published a part of it for the MOZ community. Wherever I can see, PA & DA both do not correlate to the ranking factor as Lower PA & DA websites are ranking high in SERP for HIGH VOLUME SEARCH QUERY like "cheap flights". This is quite surprising as a higher PA & DA should be higher in ranking. Do you count social shares while calculating the PA/DA?
Apart the answer Matt may offer you, maybe you can be interested in the official definition of PA and DA given by Moz here in the site :)
Great insights!
Its Helping for me,for improve my ranking on SERP,
Very interesting!
thanks a lot, it was a great post