• Register
• Log in

SEO Theory - SEO Theory and Analysis Blog

Algorithm analysis, Web community relationship analysis, SEO practices and techniques, industry news, etc.

• Home
• About SEO Theory
• SEO Glossary

How your virtual Googles shape your SEO ideas

Posted by Michael Martinez on July 8, 2008 in SEO Theory

Every now and then someone will mention some SEO technique to me and watch my reaction. I’m not sure what people expect to see or hear from me if I disagree with a technique. Just because I choose not to use an idea doesn’t mean it won’t work.

In fact, search engine optimization is filled with ideas that work in mysterious ways. As a community we can explain less than 10% of what we do with any real authority. I seriously doubt there are any individuals in the field who can explain more than 20% (if even that much) of what we do with any authority.

By authority I mean you have to know what you’re talking. I mean that your explanation has to be the correct one (and I mean that there are correct explanations). People like to box themselves in with rules and boundaries, avoiding the ambiguities that I love to caress. Nonetheless, the SEO community wallows in ambiguity.

We know there are several dozen search engines and hundreds of directories out there that send traffic to Web sites — but we almost always confine our discussions to Google, Live, and Yahoo! and most of us just talk about Google a majority of the time. Search engine optimization is not about Google but you’d think it was all about Google if you just depended on what you find in the SEO blogs and forums.

We know that links help with search engine rankings but people stumble quickly when trying to explain why. Too many SEOs still refer to Hilltop, LocalRank, TrustRank, and other SEO myths as if they know how the algorithms are used (technically, all these algorithms have been documented but the SEO community doesn’t know where they are used).

If it were as simple as “just get links”, everyone could do search engine optimization about as well as everyone else and all our optimization efforts would pretty much be in vain. If the playing field were really level there would be no way for anyone to gain an advantage through links. So if you’re beating your competition through links, you really don’t know why.

Let’s take a look at the recent “only first links count in Google versus Google passes second link’s anchor text” brouhaha. I knew when I entered the discussion that people would be appalled at the so-called test I used to prove that that proofs we have been shown were invalid. In other words, I resorted to the simplest test of formal logic to challenge the validity of all the tests that have been published: I showed that every logical argument fails at the first flaw.

The first flaw in the tests the “only first links count in Google” crowd have shared is that they don’t disclose the sites they use in their tests. Quite frankly, if you don’t reveal what the queries are and the sites you’re using are, your tests have all the credibility of a drunken sailor staggering into a bar screaming out, “I saw a UFO! I saw Bigfoot on a UFO!”

Proof is not in the presentation: proof is in the verification of the stated claims. None of the people (Rand Fishkin, Michael VanDeMar, SEO Scientist, and others) who are claiming to have proved their points have offered any proof at all. They’ve made unsubstantiated claims.

I took a chance and shared two sites that I control in a test anyone can evaluate. My two links don’t prove anything other than that I can contrive to manipulate Google’s search results pretty quickly (and I covered myself quite cleverly — as I write this post tonight I can no longer find Xenite.Org in either of the “test” queries, while other sites are beginning to move into those queries as people debate the validity of what I did).

Showing that an argument is flawed doesn’t prove anything. It merely shows that an argument is flawed. There are a minimum of two flaws in the tests that Rand and Neyne have published: first, they failed disclose the sites they were using; second, they didn’t use sites that were capable of getting a second link to pass anchor text. One might suggest the second flaw is a circular argument; however, as I have shown, some sites pass value pretty quickly and you can see the anchor text at work, even if the search results subsequently change.

Michael VanDeMar’s passes the disclosure test. He showed us the queries and named the site he was targeting. However, he introduced other flaws into his test, such as mixing two elements (meta data and links); and he failed to establish a pedigree for the pages he was using. His nonsense words are another flaw, since he failed to test in natural search results.

A bad SEO test consists of placing a link on a page, or setting up a few pages that are “optimized” for some unique expression. The worst tests, in fact, are usually conducted on nonsense expressions with little to no content. Without an environment that replicates natural search results, your testing won’t show you anything useful for search engine optimization analysis.

Most SEO tests repeat this bad pattern. I had to be careful last week with my two links because I needed to use expressions that are not naturally found on SEO Theory (they are not) but which are comprised of fairly common words (in fact, all four words are found on this blog’s pages, as Neyne was quick to point out), and the expressions had to be natural enough to generate some sort of search results.

You can create artificial search results all you want, but since no one else is creating content for your keywords or optimizing for them your tests won’t tell you what makes natural search results tick. SEO testing works best when you statistically isolate factors rather than try to isolate them artificially — that was why the SEO Scientist test seemed so convincing to me. He went to great pains to ensure that the results were replicable, hence proving cause-and-effect.

Alas! The proof has not been disclosed. We’ve only seen a claim that the proof exists. That’s equivalent to saying, “I’ve got the theater tickets at home, Mr. Usher, may we still please watch the play?”

One link really can pass anchor text and change a natural search result. However, in most cases (where I have evaluated other people’s links or my own) that doesn’t seem to happen. Why is that? As I mentioned in “Why your linking tests suck”, there are a lot of possible reasons. Most links don’t seem to pass value in Google’s index any longer — not value in terms of PageRank and anchor text.

When SEOs create test domains they pretty much invalidate their testing from the start. Unless they invest the time to create unique, random content for those domains and obtain unique, random links from relevant (supporting) content on other sites, their test domains are about as useful as a child’s chemistry set would be to a HazMat team. You’re bringing a water pistol to a gun fight. Your test domains are outclassed, outgunned, outmanned, outsophisticated — in short, your test domains suck.

If your test domains won’t rank for “Britney Spears, SEO, and Pizza” all the tests you use them for are worthless. You’re not looking at natural search results; you’re looking at virtual search results.

You’re playing with a virtual Google that doesn’t serve up results to anyone but you. It’s not real because it’s not natural. In natural search results you really don’t know why a particular page appears first, or second, or last. You can make some darned good guesses if you are experienced at this sort of thing, and your SEO techniques may help you adjust the search results. But the bottom line here is that you’re just plugging in links and content and hoping the algorithm favors your combination of links and content.

Hence, we see a lot of half-baked ideas get passed around the SEO world. First, someone performs a test on a nonsense keyword on a “test” domain that isn’t capable of ranking for anything other than nonsense keywords; then the person writes a blog post — perhaps incorporating lots of pretty pictures — making an unsubstantiated claim (and offering as proof only the assurance that they really did perform the test). Subsequently, blog fans shower the post with tons of “Great post!” compliments and Voila! someone Sphinns the post and there you have it: artificial proof that some invalid claim is correct.

Absolutely no definitive truths have been revealed in any of this discussion about whether Google does or does not pass anchor text from second links. There may be several truths behind the various test results that people have reported. In fact, the more factors you throw at a test (new domains, nonsense terms, rel=’nofollow’, do-undo iterations, etc.) the more complex the test becomes.

A complex test requires very careful analysis. To vindicate your findings, however, someone else has to be able to perform your test and get the same or substantially similar results. As soon as someone shows different results, you have to start asking questions. And that is where the SEO community fell down on the job with this business about how many links on a page can pass anchor text to the same destination.

People immediately became defensive (or abusive) and no one bothered to stop and ask, “Hey, why did the Xenite –> SEO-Theory test results contradict the reported results?”

The answer doesn’t present itself. None of us knows why. We only know what happened in the search results. My Google test performed differently from other people’s Google test. I’m in no position to explain why (although my initial thought is that I may have been the only person to use well-linked, trusted sites). I don’t know why I got different results.

What I do know — and what all you who have been following this debate also know, though you may not realize it — is that the hypothesis as stated has been proven false. If it were true that Google won’t pass anchor text from a second link, then my test results would have been no different from those reported by other people.

What that means is that we need a new hypothesis (perhaps more than one new hypothesis). The existing hypothesis no longer fits the available data. That’s the point.

But we don’t all see the same search engine. In fact, I was both amused and frustrated today by the fact that Google AdWords’ keyword research tool shows query numbers for one of my staff but not me. We use the same network to reach out to the Internet. We’re on the same IP address. I see the numbers from home, but didn’t see them at work.

I’ve been on phone calls and have exchanged emails with clients in the same city who have sworn their sites were at position X in the search results (or nowhere to be seen) while I was looking at their sites in position Y. It doesn’t even have to be a different data center (although usually I assume that’s what we’re seeing). Once in a while I find that someone’s PC has magically reconfigured itself to point at a specific IP address for Google. I’ve never figured out how that happens (as it’s NOT happening in a hosts file).

We have known for some time that Google updates portions of its indexes at different rates. You see very active blog sites like SEO Theory, SearchEngineLand, etc. move into the Main Web Index within minutes or hours of posting new content, whereas your static content sites (or less popular blogs) may take days or weeks to update in the Main Web Index.

Many blogs and forums never appear in Main Web search results.

The real-time updates ensure that we probably never see exactly the same results two days in a row for highly active query spaces. Universal Search ensures that new news, blog, and/or video results will enter (and exit) query spaces in the Main Web search results in rapid-fire sequence. Combine these complexities with the dozens of data centers from which Google serves results, as well as their attempts at providing geocentric search results, and is there any reason left to wonder why two people sitting side-by-side may not see the same search results?

We’ve got personalized search with search history, without search history, and other factors to throw into the mix as well.

All of these search environmental issues impact the search results that we see every day, and that’s just for Google. Many of the same factors can be found in Yahoo!, Live, Ask, and other search engines (A9 was the first to introduce universal search last year).

SEO testing thus takes on a rather small value in light of all these challenges. That is why I strongly suggest that people test openly on well-established sites. Of course, testing with poorly established sites, poorly crawled sites, weakly linked sites, etc. helps to show us just how much the Google Supplemental Index has remained a part of the searchscape.

If you are comfortable with the idea that it will take a month or longer to see changes to your pages appear in Google, you’re probably stuck in the Supplemental Results zone. There is no reason for why an active, trusted site should have to wait a month to see Google acknowledge changes on the site. SEO testing won’t tell you which pages are Supplemental, but it can show you how often your pages are being indexed.

Googlers have told us (more than once) that a page may be fetched more often than its cache is updated. Just looking at your server logs won’t tell you anything about how often your page is reindexed. Frequent caching is important in today’s search environment because without it you’re dead in the water. It’s good that Google has continued to work on improving the caching frequency for the Supplemental Index, but there is still one other problem with that index.

Most of the words on a Supplemental page will NOT be indexed. If you use nonsense words or rare, complex expressions on your Supplemental pages you improve your chances of finding those pages in the search results but then your linking tests will be flawed because you don’t know if your pages are Supplemental. There is no guaranteed 100% accurate test for determining whether a page is Supplemental or not, but I’d say it’s a safe bet to guess a page is NOT Supplemental if you can find it for a few relatively common words.

Like the four words I embedded on Xenite.Org for the quick little test I performed last Friday. Those words are not rare, they are not nonsense.

Of course, the Google results I saw Saturday evening no longer exist. And soon I’ll take the links off my page and move on, so the linking data will eventually be dropped from Google’s working indexes altogether.

In a few more days and weeks we’ll all have formed new ideas about what happens when you place two links to the same destination on one page. Those ideas will be shaped by the blog posts we have read, by the forum discussions we have followed, by the tests we have run. But there will be more than two ideas between us all because now we have seen something contradictory in the data.

We now know that it’s not so cut-and-dried, and no one is justified in saying that Google will only pass anchor text from the first link. That simply isn’t true. What may be true is that Google probably only passes anchor text from the first link under certain conditions, but we have yet to identify those conditions.

And maybe by the time people have devised tests that help us identify the conditions Google will have revised its handling of link anchor text anyway.

So the next time you see someone draw a conclusion based on an SEO test they performed, take comfort in the knowledge that they are still practicing last year’s SEO. In 2008, the lesson we have learned about SEO testing is that it ain’t so easy as all that, and the conclusions we draw from it should be regarded with great skepticism until everyone can produce the same results from the same tests.

A valid, correct hypothesis will be proven if its predicted results occur every time under the same test conditions. When we see variation in the performance of the test, we know we need a new hypothesis.

More importantly, to understand how complex our competitive environment is we have to acknowledge that no two people are using the same search engine. Our search perceptions form the basis for our SEO realities, and that may explain why sometimes people disagree over what feel like self-evident principles.