This document summarizes a presentation on whether links are still important ranking factors for Google search. It discusses both sides of the argument. On one hand, Google says links are still important, but correlation studies on the relationship between links and rankings have been inconclusive. The document also presents data showing branded search volume may explain rankings more than domain authority from links. It concludes that while links may still help sites reach the top competitive tier, other factors like content, user experience, and brand awareness are increasingly important for determining search rankings. The key things SEOs should focus on next are optimizing for users through testing and improving content marketing to build their brand.
33. @THCapper
“And I can tell you what they are.
It is content. And it’s links pointing to your site.”
Andrey Lipattsev, Search Quality Senior Strategist, Google
https://youtu.be/l8VnZCcl9J4
37. Classic examples:
●HTTPS migrations pre-2016
●302s are as good as 301s
●Subdomains are as good as sub-folders
●CCTLDs are as good as .com
@THCapper
51. Potential Mechanisms
1. Complete coincidence - Nicholas Cage and drownings are in fact unrelated
(!)
2. Linearity - both cheese consumption and bedsheet-related deaths are
trending linearly, and thus loosely correlated
@THCapper
52. Potential Mechanisms
1. Complete coincidence - Nicholas Cage and drownings are in fact unrelated
(!)
2. Linearity - both cheese consumption and bedsheet-related deaths are
trending linearly , and thus loosely correlated
@THCapper
53. Potential Mechanisms
1. Complete coincidence - Nicholas Cage and drownings are in fact unrelated
(!)
2. Linearity - both cheese consumption and bedsheet-related deaths are
trending linearly, and thus loosely correlated
3. Reverse causation - it is in fact drownings that cause Nicholas Cage films,
not vice versa
@THCapper
54. Potential Mechanisms
1. Complete coincidence - Nicholas Cage and drownings are in fact unrelated
(!)
2. Linearity - both cheese consumption and bedsheet-related deaths are
trending linearly, and thus loosely correlated
3. Reverse causation - it is in fact drownings that cause Nicholas Cage films,
not vice versa
4. Joint causation - both cheese consumption and deaths in bedsheets are
related to increasing affluence (& effluence)
@THCapper
61. @THCapper
Moz Study My Study
17,600 queries from KWP 4,900 queries from STAT
Top 50 results Top 10 results
62. @THCapper
Moz Study My Study
17,600 queries from KWP 4,900 queries from STAT
Top 50 results Top 10 results
Desktop only (?) Desktop & Smartphone
63. @THCapper
Moz Study My Study
17,600 queries from KWP 4,900 queries from STAT
Top 50 results Top 10 results
Desktop only (?) Desktop & Smartphone
Mean Spearman correlations Mean Spearman correlations
119. @THCapper
1.At the competitive, data-rich top end,
links mean increasingly little
2.But, for now, links might be a big part
of what gets you into that shortlist.
120. Has it already happened?
What could replace links?
What should you do next?
124. User testing for SEO: Places to start
1. Panda surveys
@THCapper
https://youtu.be/At51X-aZ4Y4
125. User testing for SEO: Places to start
1. Panda surveys
2. Click-through rate experiments
@THCapper
126. User testing for SEO: Places to start
1. Panda surveys
2. Click-through rate experiments
3. Plain old CRO - especially focusing on initial bounce
@THCapper
127. User testing for SEO: Places to start
1. Panda surveys
2. Click-through rate experiments
3. Plain old CRO - especially focusing on initial bounce
4. All of the above: Mobile first
@THCapper
128. User testing for SEO: Places to start
1. Panda surveys
2. Click-through rate experiments
3. Plain old CRO - especially focusing on initial bounce
4. All of the above: Mobile first
None of this is new!
@THCapper
So, back in September, I was in Seattle for MozCon
And this guy, who you may recognise as an up and coming face in our industry, gave a talk on link building
During his introduction, Rand said something that really got me thinking -
about how Google has gone from being a link analysis company, with a conceptual understanding of how users surf the web,
To being a machine learning company that uses real world data to inform its results.
In 1998, people navigated the web using links.
Pages with lots of links in and out of them were the equivalent of the day of search engines
and being a small number of links away from one of these key nodes was a big deal.
It meant you were popular
For a whole bunch of reasons, Google no longer needs a proxy for popularity
For example, Google is a browser
- imagine the data collection opportunities if you ran the world’s biggest browser and you wanted to figure out which websites are most popular
A lot of research platforms in our space buy data from ISPs - Moz, Similarweb, Hitwise, for example
It’s expensive and limited.
But Google IS an ISP.
And of course, they’re a dominant search engine - they know what people search for, and what they do next
And this to name just a few of the ways that pagerank could be rendered redundant as a way of figuring out whether something is popular.
And of course, compared to those methods, links are a pretty dirty signal.
That’s our fault - we, as an industry, have polluted this data
Rand’s conclusion that was we should understand what Google is trying to understand via links, and we should optimise for that
In other words, we should build links that might genuinely drive high quality traffic
Today, I want to take this a step further, and ask whether Google needs to use links at all
Before we any further: At no point may you Tweet this
If you do want to tweet something during this deck, tweet this.
I’m advocating a little skepticism, not a complete paradigm shift
Here are the three questions I aim to answer -
we actually could have easily asked these 3 questions 5 years ago, and some people were doing
But today I want to talk through their 2017 answers
Right, question 1.
If Google abandons links as its primary off site ranking factor, what next?
This is actually something that, at Distilled, we often ask candidates in interviews. We say, if you had to build a search engine today, with access to Google’s resources, but only with your design, how would it work?And people whose background is in SEO often talk about links. People without an SEO background NEVER say links.The better answers of either kind, talk about machine learning.
The cleverest answers are around machine learning, of course
This is a slide from Larry Kim’s presentation at SearchLove London.What this shows, is that over time, the higher ranking positions are gaining click through rate. Despite 4 ads, despite answer boxes, and so on.And I think Larry is right when he says that one persuasive explanation of this is that it is a signature of machine learning - of Google procedurally optimising its own results for click through rate.I’ll tell you a story about what this looks like in the real world later
Another option that you might not have thought about in this context is brand
Quantifying brand is of course something that the marketing industry has struggled with for perhaps a century or more now
But this is elementary for Google
Branded search, click through rates, maybe even signals from gmail, appstore and so on.
There’s no reason for Google to stick to one signal, of course
In fact, Google has been working on this problem for years
although this article says Last updated April 2015, it was actually first published in April 2014.
And basically this concerns the use of popularity, which porn sites do have, as opposed to authority, which they don’t have
But the difficult bit is picking all this apart, because of this.
It’s very difficult to figure out which factor or factors are actually moving the needle, when they’re also so closely related to one another
Way back in March last year, Google had come out and said that rankbrain was its third most important ranking factor, and someone asked in this Q&A what 1 & 2 were.
And this search quality strategist at Google said, without explicitly giving a hierarchy, that they were content and links.
So, pretty unequivocal from Google there. Links are still the big thing.
Google has come out routinely over the years with bad SEO advice
Some classic examples.And these cases, in my view, tend to be examples of Google getting more complicated than Google itself understands
Just because links are an input, doesn’t mean they’re interacting with rankings in a way that’s easy to understand (more on this from Will)
And, to complicate matters further, they also had this statement about domain authority - i.e., it’s not a thing, which rather confounds our traditional understanding of links as a ranking factor.
Nonetheless: Google’s answer to our question is still no. Links are important, Google says.
Now this is slightly unfair, because these aren’t exactly analogous - but this is what happens when we compare Moz’s domain authority correlation with those two we just looked atThe maximum, by the way, is 1, which the drownings come impressively close to
I’m going to talk to you about four potential mechanisms.
You don’t have to remember what they’re called, I just want you to consider each possibility when you’re looking at, for example, a traffic drop on the same day as a Google update.
Some mechanisms to think about before you claim that, for example, your work created an uplift
So these two lines are perfectly correlated, because they’re both straight - and this can be a big problem when looking at trends over time, like cheese consumption
Or in our case that whenever a site ranks first, it gets a load of links
This last point is the badger
And that would explain, in one fell swoop, all of these correlation studies
So we have to say that, having failed to control for this, all those correlation studies are actually pretty inconclusive
I just talked to you about this Moz study, and I actually decided to replicate it, and look at how brand awareness and links compare
I talked to you earlier about this Moz study
I talked to you earlier about this Moz study
I talked to you earlier about this Moz study
I talked to you earlier about this Moz study
I talked earlier about how quantifying brand awareness is actually pretty difficult
I’m using this as a proxy for brand awareness(If you want to know how to pull branded search volume for around 20,000 domains, come talk to me afterwards)
This graph shows a data point for every site in my 10,000 pages of search results, with domain authority on one axis and branded search volume on the otherIt turns out that Domain Authority and Branded Search Volume are, as I alluded to earlier, already pretty closely related, but which is the better predictor of rankings?
So the first thing to note is that using Moz’s methodology, branded search volume outperforms their best domain level link-based ranking factorA second and slightly less useful point is that domain authority is far more weakly correlated in my study than theirs - perhaps because I’m only using the top 10
Nonetheless: If you would be concerned if I told you that your DA had dropped, then you should be looking at this metric AT LEAST as closely
So when I looked at this data using regression analysis, I found that both of these variables were incredibly powerful -
we’re talking about getting to 99.99999…. And 56 more 9s, % significant
You might be familiar with this metric from CRO, where 95% is considered the benchmarkThat’s 56 9s
And that’s 95
But when I looked at this for client data rather than the big sample, typically only taking 500-1000 keywords, I noticed this
But when I looked at this for client data rather than the big sample, typically only taking 500-1000 keywords, I noticed this
And I’m not the only one finding thisThis graph is from Marcus Tober’s presentation at SearchLove London, and it shows rankings across the bottom vs. the average number of referring domains on the vertical axis - and it turns out that in the health vertical, the correlation is actually the opposite to what you’d expectNow neither Marcus nor I are saying that more links makes your rankings drop, but the point is that this is nuanced and that under certain circumstances links are fairly irrelevant
This graph shows baserank over time for two sites - baserank being the ranking if we ignore SERP features like answer boxesSo, what do we notice about this graph?
Three things
Both sites are all over the place on a week by week basis
Three things
This graph shows baserank over time for two sites - baserank being the ranking if we ignore SERP features like answer boxesSo, what do we notice about this graph?
Three things
Our traditional understanding of what could cause this falls into these three buckets.
We don’t have time to get stuck into wayback machine now, but you’ll have to take my word for it: neither of these sites had substantive changes in this period
This is Moz’s record of every known algorithm change, which is available publicly on their website if you haven’t seen it.Obviously the algorithm is continuously evolving, but the the only named update during this period is related primarily to local packs.
So, these graphs are from Majestic. They show newly discovered links over time.
Neither site was rapidly gaining links in the run up to that drop, which is the blue area on the left
In fact the biggest spikes on FlyingFlowers’ graph are comparable to the smallest weeks on Interflora’s.Incidentally it’s also not a question of link velocity - the above charts for August and September do not represent a significant departure from their historical norms
So that leaves us without a sensible explanation for this
And it’s not just those two sites by the way - when we add in the other competitors, this whole SERP is all over the place.
My suggestions as to what Google is looking for in these iterations is this - aesthetics and price
This is the Interflora page, which dropped away, and what I notice on this page is how busy it is, and the prices that are stuck up there.
This is Flying Flowers - It’s a much lighter site, and the average price of a product on the homepage is about 30% less.
That’s the biggest difference between these two sites - the actual content is nearly identical. But it’s all about that first impression.
Fleximize is a company that sells small business loans, and worked with Distilled on a content campaign
Okay, so this is a graph from STAT of their commercial keywords.
You can see that in the middle of this graph, they go from having 4% of their keywords ranking between 4 and 10, to having the same number ranking between 1 and 3, where they’d previously had none
- so this is still early days for their SEO success -
And they also gain a bunch of extra keywords in the 11-20 bracket
But if we look at how this happened - first there was a content piece which gained 168 referring domains, and did nothing to their commercial rankings
Then another smaller one gained 22 referring domains, still nothing
Then a piece gains 191, and suddenly everything moves
Will and Rand did a Whiteboard Friday on this way back in 2012
But Panda was arguably the first big jump in Google looking to understand what users were looking for, so you could do a lot worse than replicating their research with your own site
This is one of the types of thing that falls under SEO split testing - changing half your category pages, for example, with some alternative meta descriptions, and playing around with it to see what makes you more trafficI’m sure you’ll hear more about SEO split testing over the next couple of days
Doing what you can to avoid that jump back to the search results when they do actually click through
And of course, in all of these activities, do them first for mobile then figure out how it applies to desktop, not the other way round
None of this is new!
It’s just more important than ever that you get it right.
(Vicke stealing this slide)
This looks a lot like what has gone before
My deck should be seen as part of a two part series with Vicke’s which you’ll see later today
What it all comes down to is this
So rather than trying to keep up, aim for what they’re aiming at, and optimize for people
What could replace links: I’ve talked about brand and user signals as two prime candidatesHas it already happened: Depending on your particular circumstance, quite possibly
What should you do next: Focus on optimising for the things that Google is trying to optimise for