Note: if you just want to skip to my conclusions, you can find them here.
Recently, I had a bit of a breakthrough.
As we’ve expanded the agency, I was finally able to use our internal resources to build out & rank our own projects. I’ve always had the mindset of “drinking our own Koolaid”, and as we’ve gone down this path, I recently stumbled into a rabbit hole that gave me a huge burst of excitement and an increase in expectations for what we could do in the near future. But it came at a cost: paranoia.
Once the dust settled on the improvements we made, I took a major step back and realized that what we were building was more or less sitting on the fault line of a tectonic plate.
It could all come crashing down in an instant, all because of one critical assumption that I’ve made to date: that links will continue to matter.
I quickly realized that I needed to have a better gauge on the longevity of links beyond the tweets I happened to read that day. I’ve never had much cause for concern over the years regarding this issue (evidence of why is listed later), but if I was going to make a major bet over the next 12-24 months, I needed to know the parameters of what could go wrong, and this was one of the items at the top of the list.
I ended up discussing things over with a few trusted colleagues of mine, as well as reaching out to a few other experts that I trusted the opinion of in regards to the future of SEO. So I wanted to share with you my thinking, and the overall conclusions I’ve drawn based off the information available.
Separating Facts from Opinions
The main source of “facts” that the industry points to as a whole are statements from Google. Yet, there have been numerous instances where what Google is telling us is, at the very least, misleading.
Here are a few recent examples to illustrate in what way they are misleading:
1. In their “Not Provided” announcement post in October 2011, Google stated that “the change will affect only a minority of your traffic.” Not even two years later, Danny Sullivan was told by Google that they had begun work on encrypting ALL searches. The rest is history.
My thoughts: even when we get the truth from Google, it should be labeled with huge, red letters of the date the statement was made, because things can change very, very quickly. In this case, it was probably their intention all along to gradually roll this out to all searches, in order to not anger people too greatly all at once.
2. Google’s John Mueller made this statement a few weeks ago about 302 redirects passing PageRank. It implies that 302 redirects are OK for SEO. As Mike King quickly pointed out on Twitter, that’s very misleading based off most SEO’s prior experiences.
My thoughts: is it difficult to believe that 302 redirects pass at least 0.01% of the PageRank of the page? I don’t think so. So really, this statement isn’t saying much. It’s a non-answer, as it’s framed in comparison to a 404 (no PR passes) instead of a 301 (~90% of PR passes), the direct alternative in this case. So really, it doesn’t answer anything practical.
Take those two examples & realize that things can change quickly, and that you should try to decipher what is actually, concretely being said.
So, with that in mind, here are some recent statements on the topic of this post:
1. March 24, 2016 – Google lists their top 3 ranking factors as: links, content and RankBrain (although they didn’t state the order of the first two; RankBrain is definitely 3rd, though).
My thoughts: this isn’t anything new. This list lines up with what they indicated in the RankBrain initial news article in Bloomberg when they stated RankBrain was #3. All that was left to speculate, until now, was what #1 and #2 were, although it wasn’t too difficult to guess.
2. Feb 2, 2015 – Google confirms that you don’t necessarily need links to rank. John Mueller cites an example of friend of his who launched a local neighborhood website in Zurich as being indexed, ranking, and getting search traffic.
My thoughts: this isn’t very surprising, for two reasons. First, that the queries they’re ranking for are probably very low competition (because: local + international), and because Google has gotten a lot better over the years at looking at other signals in areas where the link graph was lacking.
3. May 5, 2014 – Matt Cutts leads off a video with a disclaimer stating “I think backlinks have many, many years left in them”.
My thoughts: as much of an endorsement as that is, a haunting reminder of how quickly things change is Matt’s comments later in the video talking about authorship markup, a project that was eventually abandoned in the following years.
4. Feb 19, 2014 – Google’s Matt Cutts stated that they tried dropping links altogether from their ranking algorithm, and found it to be “much, much worse”.
My thoughts: interestingly enough, Yandex tried this starting in March 2014 for specific niches, and brought it back a year later after finding it to be unsuccessful. Things change awfully quick, but if there’s any evidence on this list that can add reassurance, the combination of two different search engines trying & failing this is probably best. With that said, our main concern isn’t the complete riddance of links, but rather, its absolute strength as a ranking factor. So, once again, it’s still not all that reassuring.
Opinions of Others
Let’s now transition to the opinions of others in the industry. It could be argued that these can be a much better gauge on the reality of SEO than whatever Google is telling us (and I’d agree!).
The most substantial opinion piece to start off with is Moz’s Bi-Annual Search Ranking Factors study. Half of the study is based around a survey that was given to 150 experts. In the survey, questions were asked about the most important ranking factors, both for today, and for the future. Here are the results of current ranking factors:
And here are the results for predictions of future algorithmic changes (only linked, not embedded, because it’s quite long). For these, note that zero of the “predicted to increase in impact” factors were link-based. Furthermore, the only 2 in the “predicted to decrease in impact” were link-based.
As I mentioned earlier, I decided to touch base with a few specific people in the industry that I place a lot of trust in: AJ Kohn & Justin Briggs. Here’s what their thoughts were when asked about the future of links as a ranking factor:
Links are and will continue to be an important part of SEO for the foreseeable future because they remain a powerful way for Google to measure authority and expertise.
The link graph has been at the heart of Google’s search algorithm from the start. One of the more interesting videos Matt Cutts did related to separating popularity from authority. He makes the point that popular sites might include porn but people don’t often link to porn. On the other hand he says that many government websites aren’t very popular but they do attract a number of links.
In the same video, Cutts also discusses how the anchor text used in those links can help Google to better understand the topic for which it might rank. And there are numerous patents that delve into how much weight to give anchor text and how that might aid in establishing topical relevance.
Now, Google is getting better and better at understanding the meaning of content, but that doesn’t mean that links will suddenly lose value. They might matter slightly less but I generally see these improvements as being synergistic.
But let’s put all of this aside and look at the bigger picture and use some logic. Does Google still police paid links and other manipulative link schemes? Of course they do. And the only reason to do this is because links still matter.
Currently, and within the short-term, links are here to stay (at least in the traditional information retrieval of documents aspect of search, which is shrinking over time). An often undervalued aspect of links, in a very traditional PageRank sense, is that “link equity” is an input for URL discovery, crawl scheduling, crawl budgeting, crawl depth, and likely hundreds of other processes and checks. I see links as the first layer in rank determinations. The net effect is that their “slice of the pie” is getting smaller, but that’s not exactly what’s happening. Results may be put in order based on more traditional ranking processes, then search engines integrate usage data (CTR, bounce, bias), brand affinity, search sessions, query refinement, machine learning, localization, and personalization. The net outcome of these “re-sorts” is that the perceived weight of links goes down, but links are responsible for getting the URLs into the original consideration set for rankings.
The value of links in Universal Search has eroded, because search is about more than retrieving articles. Mobile, voice, entities, structured data, personal search, conversational search, predictive search, and apps have little dependency on links. Some of these technologies never refer to the link graph, with the caveat that many of these rely on the desktop index to run (or at least to “learn”).
When looking at SEO, I’m less concerned about the changing value of links and more focused on the declining importance of traditional, document-based search results in a company’s overall search strategy. However, we think of links in terms of digital PR and promotion. A marketing plan always has room for good promotion.
Additionally, another person I was going to ask to contribute to this section was Will Critchlow, but he did the legwork for me when he published an article titled, “Google to Announce that Links are no Longer a Major Ranking Factor” (disregard the clickbait title).
In it, Will talks about how RankBrain being added to the mix affects the future potential value of links for SEO. I will pull out the most relevant bit:
What this means in practice is that even after whatever change is made to dial up the dependence on RankBrain and dial down the dependence on the human-tweaked algorithm, I believe that we will continue to see link metrics be better correlated to rankings than any other metric we have access to.
In other words, RankBrain will be more important than all the individual signals in the human-tweaked algorithm (including links) but links will remain the dominant signal that RankBrain itself uses.”
My Own Opinions
Let’s take a step back. Have links stopped being an indicator of the quality & relevance of a website? Has a link from TechCrunch or the National Institute of Health stopped being relevant to the assessment of the legitimacy of a website? Has that changed?
I only see two main things that have changed in our understanding of links as a ranking signal:
- That some links do a better job than others at indicating the quality & relevance of a website.
- That there are things beyond links that can also indicate the quality & relevance of a given website.
Google has done a better job of understanding those two things since they first started. For the first item, that’s why you have Penguin. For the second item, that’s why you hear about things like unlinked brand mentions & social signals.
But the idea that links not being a signal in the future altogether is beyond ludicrous.
It would be discounting the foundation of what the algorithm is built upon. And that’s not important because of historical significance, it’s important because it’s based off how the Web fundamentally works. Links are just connections between things, and some of those connections hold more importance than others. Throwing out links altogether as a ranking signal would be the equivalent of disregarding recommendations from people that you trust.
So really, the argument over link-based factors playing a role versus no role at all, is dumb.
Note: so now that we’ve established this, when I talk about links in the context of the rest of this article, I will be talking about the links that Google WANTS to count, not all links on the Web.
If, so far, we’re on the same page, then the real question is how strong of a ranking factor links will be. There are two main things that will influence this.
- The introduction of new factors.
- The relative strength of each factor
The first is simplest to explain, so let’s start there.
The Introduction of New Factors
As new ranking factors are added to the algorithm, inevitably, dilution happens. There is only 100 percentage points that make up the entire decision making process behind an algorithm. It’s a limited amount of space. So the introduction of something new, even if it’s tiny, inevitably takes space away from all others.
And if the factor does its job and holds meaning, then that’s good. That means a smaller reliance on any one, single factor. That doesn’t mean just links. That also means things like content-based or user experience-based factors.
The concept of new factors being introduced into the algorithm represents an unknown. And I could never claim to have an accurate pulse on new things altogether that Google might be introducing into their ranking algorithm.
The Relative Strength of Each Factor
Note: I will usually be using the phrase “the concept of RankBrain” instead of simply the term “RankBrain”. This is because I only know that it’s using machine learning, and will describe it from the standpoint of what machine learning models do, in order to extinguish any confusion about me having any real idea of what RankBrain is & does, which I don’t, because not much is publicly known.
People are talking a lot about the concept of RankBrain, and for very good reason. It, without much doubt, dictates the future of the importance of individual ranking factors. But to illustrate why I think that is, I’ll back up a bit.
After reading all of the wild speculation about RankBrain, I noticed that there are a significant amount of people that still don’t know the basics of what machine learning does, the technology that RankBrain is said to be using. This is how Wikipedia describes it:
Machine learning explores the study and construction of algorithms that can learn from and make predictions on data.
In essence, machine learning is used to make predictions. It can’t help Google magically figure out what is the best result is for a user who is conducting a specific search. That would involve things like definitively knowing exactly what “the best” is, and outside of things like math equations or historical facts, is most likely impossible. So for now, Google can only guess, and get really, really good at guessing.
And that’s why we’re talking about predictions.
So let’s now focus on figuring out how to make an accurate prediction. Predictions are based off a set of factors. The “secret sauce” of machine learning is figuring out which factors are more important than others in determining what you’re predicting. As Will put it in his article, it won’t be humans doing this manually in the future, but rather “the machine tweaking the dials.”
That explanation helps to explain why links and the concept of RankBrain are not at odds with each other. They’re apples and oranges. It’s like saying historical forecast data and a weatherman are at odds with each other over predicting the weather. One is information and the other is an interpreter. They are two separate types of things.
So hypothetically, in the case of links, two potential things could happen when machine learning gains more control over the ranking algorithm:
- The dial is turned back on links, realizing that they’re not as good of an indicator of the quality & relevance of a website than what Google engineers had previously given them credit for.
- The dial is turned up on links, realizing that they’re a better indicator of the quality & relevance of a website than what Google engineers had previously given them credit for.
To date, I’ve never seen the second hypothetical situation talked about. And while its probability can be (justly) questioned, I think it’s an interesting scenario to discuss.
What if the hysteria around links in our industry has caused Google engineers over time to manually & mistakenly give them less weight than they deserve? What if their new machine learning models indicate that quality links (remember: Penguin is changing the game here) are actually a really good indicator, more so than previously given credit for?
I don’t have a clear idea on the likelihood of each of the two hypothetical scenarios listed above, but let’s be clear: links being dialed up as a ranking signal due to new machine learning models is as real of a potential outcome as links being dialed down as a ranking signal.
Now that we understand the potential outcomes that machine learning in a ranking algorithm can have from the standpoint of links, it’s now time to discuss the probability of each outcome happening. This is where the real discussion begins. There are two main ways that a dramatic change could happen in the given value of links as a ranking signal:
- Google engineers, previous to machine learning, had done a poor job in determining the exact importance of links as a ranking signal.
- Google engineers, previous to machine learning, had done a poor job in determining the exact importance of other ranking signals.
And as a result of either, when a highly accurate machine learning model is introduced, the correction is made. The limited space within the algorithm would be re-distributed.
So let’s now discuss each of the above two possibilities separately.
Links As A Ranking Factor
The first option sounds improbable.
Links are the oldest signal in their algorithm. These PhDs have had almost two decades to screw around with the dial. It’s very reasonable to think that a machine learning model is not going to significantly alter their importance as a ranking signal, as it would imply that these engineers had been horribly wrong after all this time, in one direction or the other.
But there is a very real scenario to consider. It involves Penguin.
We’ve all seen numerous examples of link spam, even in shockingly recent times. Those examples, coupled with the insane amount of time Google has taken in releasing the next Penguin update, shows that they’re still scratching their heads & don’t quite have it all figured out.
But in the context of this investigation, its importance here is significant. It’s a wild card. I’ll explain why.
Let’s assume, just for a moment, that Google had been right for placing their immense trust in links as a ranking signal. Let’s pretend that somehow we were able to divinely identify what really was the best indicator of a quality search result, and that at the top of the list of indicators was links. This is important because, soon enough, they’re going to find this out the more that they use machine learning.
So, if Google’s views on the importance of links as a predictor of a quality search result does not change, what will happen when they perfect the art of cutting through the noise & only identify and give weight to links that indicate a true endorsement of a website (a quality link)? If Google has been giving links as much weight as it has in the past, even when they didn’t fully understand which links were good & which ones weren’t, just how much further would the dial potentially be turned up once they’re near-perfect at this?
The conclusion I’m trying to draw here is that, once again, there’s a very legitimate potential outcome that links could INCREASE in importance as a ranking factor as they continue to refine Penguin and their overall analysis of link-based factors.
I think that’s a profound realization, and yet, once again, it’s not even being discussed.
Personally, though, I don’t think that this will happen any time in the near future. Here’s why:
- Overall, Google still seems to be far off in correctly classifying links 10 times of 10 as either spam or not.
- The last Penguin update has taken a while. This could be because they’re not happy with the results, or because they’re putting internal resources elsewhere. Both aren’t good signs for links, although other reasons could exist.
- There are many other new signals that haven’t been tested & used to the extent of links.
Other Ranking Factors
Now let’s discuss the second scenario. Unfortunately, it’s a much more complex discussion than the first because:
- 1 signal is simpler to discuss than hundreds of other individual ones & all their various combinations.
- We’ve gotten information about links publically from Google. For a lot of other signals, we don’t know much.
- As a historically important factor, there have been a lot of studies & opinion articles published about links in the marketing community.
So with that said, here are my main thoughts about this group as a whole.
1. Time. It’s on the side of a lot of new factors Google has been rolling into the algorithm over recent years, at least in comparison to links.
Machine learning aside, even though I’m guessing they’re much more efficient at doing so in 2016 than in 2006, they still haven’t had relatively much time to mess with the dials of each, as opposed to something like links.
Additionally, for a lot of newer signals, it’s doubtful that they’ve cut through all of the noise for each, in the same way that they’re trying to cut through the noise in regards to links via Penguin. I assume that’s what is holding back a lot of UX signals.
Note: for a more concrete set of timelines around specific factors, checkout SEO By The Sea. Bill Slawski has done a great job surfacing Google patents (as they’re granted) that talk about some of these, and they all have a filing date, which is better than nothing.
2. Segmentation of ranking algorithms. The implication of an answer given by Google in an FAQ help doc about the mobile friendliness update is just one piece of evidence signaling a division in SEO, in which the concept of a singular ranking algorithm is dated.
Earlier examples of this concept are found with things like the Payday Loans updates, in which the organic results of certain industries were ranked differently than for other industries.
In most cases, especially with things like mobile, I fail to see much of an opportunity for links to be a beneficiary of these segmentations. I more so see it as links being more or less a “fall back” when they aren’t able to use factors that do a really good job for specific segmentations of searches (i.e. UX factors for a search done on mobile, dwell times for an investigative search, etc.).
With that said, there are a number of very interesting problems that Google has here. A few of them are noted further down in this write up of a recent Googler’s presentation on search.
3. The increasing complexity of the algorithm. Inevitably as more signals have been introduced, and the dials of each have been tweaked and re-tweaked 100s of times, and that each of those dials are no longer universal and are now segmented for different types of searches users do, the complexity has grown.
From what’s been said publicly by Google about machine learning, the feeling I’ve gotten is that they’re working on it, but that we shouldn’t expect things to happen quickly, and my guess is because of the level of complexity behind integrating this technology into all of the various parts of organic search.
Overall, it’ll be interesting to see just how quickly Google will move now & in the future as their algorithm becomes increasingly complex, especially when most of it seems to still be driven by humans, not machines.
Because the above evidence listed in various places throughout this post is far from substantial, I’m only confident in my conclusions from the standpoint of where we are today, not 5 years from now.
Here’s what I believe:
- Links still work, today. This is based off strong evidence of link campaigns we’re currently running (in April, 2016), Google’s March 24th, 2016 statement, and as AJ Kohn noted earlier, Google’s active pursuit in combating link spam.
- Tomorrow is not a guarantee. As we’ve seen, Google can move very quickly. With that said, even if Google decided this very morning to move away from links as a significant factor, I highly doubt they could make a major change within a ~12-18 month timeframe, just because links are so foundational to their search engine.
- The concept of RankBrain is not a major threat to links. I even think there’s a very real chance that it’s not even a minor one.
- There doesn’t seem to be a golden knight to replace links. The most talked about new set of factors is UX, and I have seen more than a few examples of specific UX signals being easily manipulated, even more so than links.
- The real threat is more foundational than links. Justin Briggs explained it best in his response earlier. The aspect of ranking a page organically in Google’s results has slowly declined in value, both because of other SERP features & search ads. There’s still a ton of money to be made, but we should work like we’re living on borrowed time.
Things do change quickly. But for now, I won’t be hopping off the link bandwagon in the near future.
Thoughts? I’d love to discuss what you think in the comments!
Published at Wed, 06 Apr 2016 18:42:30 +0000