DISCLAIMER: This article refers to the original use of “the SEO method” as established in 2006 and 2007 on the SEO Theory blog, not a marketing program that has been sold on the Internet by that name since 2008. Neither Visible Technologies nor I endorse the marketing program, as we have no connection to it.
The SEO method is pretty simple: You experiment. You evaluate. You adjust.
I’ll admit that I am disappointed in how poorly so many people seem to understand the experiment part. Despite recent brouhahas over experimental methodologies and reporting flaws, most people in the SEO community (I am tempted to say “all”) experiment with search results quite often. Every time you publish new content, every time you place a new link, you are being experimental.
The SEO method does not require that you compile normalized statistical measurements and examine derivative scores on a Six Sigma ruler. The SEO method is nothing more than an extension of the scientific method, which calls for an organized approach to studying phenomena.
The SEO method calls for experimentation first — which can be either informal (as most of us do it) or formal, as only some of us do it (and then probably only occasionally). The formal experimentation must conform to the scientific method if it is to be credibly proposed. The scientific method begins with a question such as “does X produce Y?” or “what if X does Y?” These questions must lead to research, study, observation and analysis.
Only after you have observed and analyzed can you really postulate a good hypothesis. People in the SEO industry just seem to leap over the first three steps very quickly. They ask questions but they ask those questions poorly. The scientific method tolerates simple questions (why do apples fall to the ground?) but when you prequalify the question with unattested assertions, you’re trying to ask a question with an obvious answer (if links help improve search results, is search engine optimization all about links?).
Your hypothesis must be simple enough (or its complexity supported by established scientific knowledge) that it can be tested independently of anything you say or do. Let’s take PageRank Sculpting, for example. We can begin with the question, “Is it possible to manage the flow of PageRank about a site with precision, such that the site obtains some benefit not evident in unmanaged PageRank flow?”
That question simply asks if sculpting PageRank is even possible. If there is no observable benefit, then it’s not possible. That is, if you achieve nothing with the sculpting, then you’re just spinning your wheels — you’re not accomplishing anything.
We can do the observation and analysis pretty easily in several ways. For example, Matt Cutts said that Google sculpted PageRank on YouTube by managing how content was linked to (they even used “rel=’nofollow’”). Observe, however, that it was Google who managed the flow of PageRank around its own site — Google, which alone of all among us has the ability to track and measure PageRank.
Based on that observation, one must reasonably conclude that it is not possible for anyone outside of Google to manage the flow of PageRank through a Website with any precision.
Of course, as has been noted by many, we have (for years) had to observe the predicted effects of black holes in order to see that they are there — as the theory of black holes says we would be unable to see them. Ironically, we can now see black holes so that’s no longer a good metaphor for how to detect the flow of PageRank around a Website.
Science has learned to detect and observe black holes by studying spectra of energy that we mere humans cannot perceive. PageRank flow, however, is more complex than a black hole. Here is why: Google updates its internal PageRank assignments more frequently than the few times a year that it publishes its Toolbar derivative PR date.
And that is all anyone outside of Google can say with any reasonable credibility (except for ex-Googlers who have steadfastly remained silent on the details of Google’s PageRank machinations).
In other words, we don’t know how Google updates the internal PageRank data for individual pages, Websites, sub-domains, domains, communities, vectors, victors, and other essential parts of the Web which Google has indexed. Does a page in the Supplemental Index accrue PageRank? Does PageRank for one page on a site flow to its siblings immediately or are the links queued for processing further on down the road?
Have you ever considered the possibility that Google might not evaluate the PageRank for every page on a site at the same time?
Given so little knowledge and so many unanswered questions, we cannot claim to have observed much and our analysis leaves us wanting more. Hence, we’re unable to create any testable hypotheses regarding the management of PageRank flow around Websites. That is, until you can predict when a page’s PageRank is modified by Google you cannot devise a scheme for effectively managing the flow of PageRank.
Let me demonstrate quickly: Let’s say you have a 10-page Website. You use your root page to link to the other 9 pages, and they all link to each other in groups of 3. If you cannot track and measure PageRank you can at least track and measure the internal anchor text. So let each of the 9 pages use unique anchor text to link to its siblings. They should all also use unique anchor text to link to the root document. And the root document should link to each of its 9 children with unique anchor text.
In all, you’ll have 36 links with unique anchor text.
You cannot begin your PageRank sculpting test until you confirm that all 10 pages appear for all 36 unique expressions in Google’s index. These expressions, being unique, can only appear once — in link anchor text — and nowhere else.
Now, there is no guarantee that Google will pass PageRank if it passes anchor text (and vice versa). It’s a leap of faith we have to make — an assumption that must be noted in the conditions of the test — that we’re treating passed anchor text as an indication that PageRank is being passed from document to document.
What does it take to get a 10-page Website to appear in Google’s index under these conditions? If you only link to the root URL once from some established site, will that be enough? Let’s assume for the sake of discussion (not as a condition of the experiment) that we decide we need 5 links from trusted, aged sites to get all 10 pages indexed for 36 anchor expressions. Now we can start fiddling with our PageRank.
For example, what happens if you take one group of 3 pages and add an extra link on each page to another group? So group (A..B..C) links to group (D..E..F) such that A links to D, B links to E, and C links to F. Now D,E,F all have 1 additional link and anchor text passing to them. What happens in Google’s index to the original 36 listings?
But here’s a complication that the setup for the test did not take into consideration: what if Google devalues or deindexes one or more of those links pointing to the root URL of the domain? You can’t know when that happens for sure, although you can set up some controls to help you see when links may lose value (have those pages point to other sites with unique anchor text).
Here’s another complication: what if Google decides to recrawl only part of your test Website? What happens to the other pages’ standing within the search index?
It is impossible to stabilize the backlink profile for a Website. You can keep building links but if you don’t know how much PageRank is being passed to a Website you have no way of controlling where the peanut butter is being spread.
It is equally impossible to determine from an outside perspective how frequently Google will recrawl each page on a Website. Yes, you can watch your server logs for fetches from Googlebot — but what timeframe do you use? 1 day? 1 week? 1 month? 1 year? If you don’t know how much time to allot for a complete cycle of recrawling, you have no way to establish a window for estimating the flow of PageRank.
You can arbitrarily choose a multi-month period — many people do just that — but that multi-month period may encompass several recalculations of PageRank for only parts of your site. Remember: you have no way of knowing what the PageRank is or how often it is recomputed.
You can, of course, try to observe when your pages stop appearing for unique anchor text during your multi-month period. Knowing this, you could attempt to correlate such losses of anchor text with drops from the index, changes in cache value, and other factors that may indicate that Google has changed how it looks at your pages.
But here’s the problem with that: in the midst of testing your hypothesis that you can sculpt PageRank, you’ve now found yourself needing to form (and test) a hypothesis about when PageRank changes. And that shows that the experiment as proposed is flawed because you cannot isolate the one factor you’re trying to test.
So what you really need to do, before you embrace the ambitious goal of proving that PageRank can be sculpted, is to craft a solid method for tracking and measuring PageRank.
And that is not a Catch-22. You can set up a measurement that tells you something about how a search engine evaluates a Webpage. All you have to do is document the changes that the search engine reports to you for each Web document, and note the events that precede and follow the changes.
The key principle is to establish that change in how the search engine treats Web documents is uniform and predictable. Once you achieve predictability you can look for correlations with the passing of anchor text. When you can isolate a mode of change that correlates with the passing of anchor text you have a reasonable mechanism for tracking and measuring a pseudo-PageRank.
As long as it’s consistent, your pseudo-PageRank will be good enough for your further testing of PageRank-centric hypotheses. After all, if you ever get the ability to track and measure real PageRank you can just plug that into your tests. Your previous work will help you predict what should happen on the basis of the changes in PageRank flow.
So let us now take our pseudo-PageRank and pass it around our Website. We add links, we subtract links. On the basis of our previous work we can now predict with precision what should happen. If our predictions turn out to be wrong there are two possible explanations: either our pseudo-PageRank is not as predictable as we had hoped OR we are not sculpting PageRank.
Random stuff happens all the time, so you need to test this experiment in as many different ways as possible. Let’s say you set up 100 Websites to test this model. And let’s boldly assume that you observe the same results across all 100 sites.
You still haven’t shown that you can sculpt PageRank. Why? Because your link profiles are not stable. In fact, your search engine is not stable. Google claims to make about 10 algorithmic changes per week. We cannot assume that all those changes have anything to do with PageRank, link anchor text, or how documents are indexed. But we do know that these changes may affect what we see in the search results.
In other words, our observations cannot be trusted. And if our observations cannot be trusted our predictions are not going to be reliable. If our predictions are not reliable, then what does it mean if we predicted changes in anchor text-based rankings 100 times and the predictions were all accurate? It means that we overlooked something.
Too much consistency burns the experiment. Consistent systemic behavior cannot occur within any dynamic structure. If during all these months of changing Web sites and not changing Web sites you don’t get some erratic results, either Google has stopped changing stuff or your experiment is flawed.
So we cannot assume for the sake of discussion that all 100 Websites perform as predicted — not every time. There have to be some instances where the predictions come true. Otherwise we’re allowing something we haven’t identified to influence the search results we’re trying to influence.
Albert Einstein is credited with having said the equivalent of “No amount of experimentation can ever prove me right; a single experiment can prove me wrong.”
Scientists understand that you cannot prove that Sculpting PageRank works. Google can prove it because Google owns the search index and controls PageRank. Google can arbitrarily set PageRank any way it wants. On the basis of Google’s authority we can reasonably say that PageRank Sculpting works — but that is all we can say.
No one outside of Google can credibly show that they are sculpting PageRank. You don’t have enough information to manage the flow of PageRank with precision. But maybe you don’t need to go that far. After all, given a large enough Website you should be able to invoke the economies of scale.
That is, if you do things often enough and efficiently enough, you should be able to channel your PageRank through the sheer weight of volumes of links. Now, the intuitive argument against PageRank sculpting is that, if you want to channel more PageRank to any particular page you should just link to it from more pages. The intuitive argument for PageRank sculpting is that if you want to make a page seem less important then you should remove links pointing to it — that is, remove PageRank-passing links. Of course, you want visitors to still be able to find the document.
That is the fundamental flaw in PageRank sculpting. It hypocritically argues that certain pages are not important and yet insists that they be linked to for the user’s benefit. The search engines are trying to use PageRank-like values to figure out what is most important to people, not to rankings.
Let’s go back to our original Website. Instead of using only 10 pages we’ll use 10 million pages. Now let’s say we want page 900,000 — a deep leaf-node page — to become as important as the root URL. How do we do that? First, we have all 999,999 other pages on the site link to page 900,000. Second, we obtain as many inbound links for page 900,000 as for the root URL (page 1).
Are we channeling PageRank to a specific page? Yes.
Are we doing this with precision? No.
Can we refine the link placements so that the channeling becomes more efficient? Yes. We can arbitrarily remove groups of links from page 900,000’s backlink profile. But here’s the rub: What do we mean by “as important as the root URL”?
People in the SEO industry talk about PageRank as if it actually means something but they have never figured out what, exactly, it means. Ask most SEOs today what PageRank represents to them and they’ll likely say something about trust and authority. But what are trust and authority?
Some people also talk about managing crawl but crawl, though influenced by PageRank, is affected by other factors (under the Webmaster’s control). So crawl is not a good indicator of PageRank-based value.
In short, we have no real way of knowing what PageRank does for us. We know it seems to determine which pages are placed in the Main Web Index. We also know that it (originally was used to) weight(s) the Information Retrieval score that Google computes for documents in resolving queries. That is, PageRank is (assumed to be) added to the IR score, such that two very similar IR scores might be adjusted to reverse the ordering of the listings on the basis of PageRank. We don’t know which factor is more significant, PageRank or IR score, except that many low-PR documents do outrank many high-PR documents — which implies that IR score plays the larger role in determining Google’s search results.
And that would be an interesting test in itself — which implies that we have not yet gone deep enough into the algorithm to devise an atomic test that tells us something we can incorporate into another test.
In other words, scientific experimentation is not as easy as the SEO pundits would have us believe. Before we can test something as complex and mysterious as PageRank Sculpting we have to establish axioms and systemic processes that are as ironclad as the failure to disprove them makes them — that is, they have to be good enough for science, real science. Science does change its axioms on occasion, but it’s a real struggle to institutionalize any assumption or finding as an axiom.
People question scientific studies all the time. They go back and forth over the tiniest minutiae. It took 300 years to solve one of the simplest and intuitively most easily understood riddles of mathematics, but in order to do that mathematicians had to invent several new branches of math. I’m talking about Fermat’s Last Theorem, the proof of which is so esoteric I could not even begin to explain it to you. Like most other people, I just have to take it on faith that the math is correct, the science is real.
But Andrew Wiles’ proof of FLT is much more trustworthy than the claims people make about PageRank Sculpting. They lack the millennia-long scientific history that modern mathematics arises from.
If we lack the tools to properly conduct PageRank sculpting experiments, is there anything we can do to study how links affect Websites? Sure. We can conduct lots of little experiments. But until you learn to isolate factors in your observations any claims you make based on your research will be suspect. Your findings are only as good as your observations, and you’re not ready to observe PageRank in motion.
None of us can do that. Probably, none of us ever will be able to observe PageRank in motion — not as long as search engines care about using PageRank-like data to influence how they rank Websites.
At best all you can do is say, “I did this and observed that”. If enough people corroborate your observations, we might reach a consensus about something. But I’ve looked at enough screwed up sites that were attempting to sculpt PageRank to understand that what people think they are seeing is not real.
That is why, Google tells us, they changed how they handle nofollowed links a long time ago — a change that all the SEO tests failed to detect. You’re in no position claim that PageRank sculpting still works since you were never in a position to claim it did in the first place.
Only Google can sculpt PageRank. Only Google can make it work.
But we can all use the SEO method to learn a little bit more each day about what seems to work. I advise you not to waste any more time spinning your wheels on PageRank Sculpting, but hundreds of thousands of people devoted a lot of time and effort to failing to prove Fermat’s Last Theorem. You’ll have plenty of company if you want to pursue a dream for which you are not equipped.
Written by
Michael Martinez
