Response to SEO Questions by Rand Fishkin
First, a little note. I mentioned this to Rand when he was considering adding new companies to the SEOMoz Recommended Vendor’s page. At Virante, we use a brute-force method of research. When we want to know something, we run real experiments in the search engines so that we can exclude assumptions regarding the myriad extraneous factors that go into ranking. In our experimentation, we have covered several of the questions he mentions in his most recent post, Questions in SEO that I Can’t Answer. Check out Number 8 if you get a chance.
1. The Diminishing Value of Anchor Text
The theory goes that if a page is linked to multiple times on a single page, the subsequent anchor texts (assuming they are unique) will carry less weight, or perhaps no weight at all in the rankings. For example, if I linked to this groovy thing on utensil art, which is coolerific, the search engines would give more weight to the anchor text “groovy thing” than “utensil art” and “coolerific” – true or false?
Answer: Anchor Text Value Diminishes for Multiple Links on a Page
This experiment was more complicated than I expected. 1 new domain with 3 subdomains was constructed. Each subdomain was a 4 letter nonsense word like jkdh. (ie: sdjd.ajaksadlfn.com) The content on each page was a random 250 7-letter nonsense words. No title or meta were used, and the target keywords were not used, to avoid on-page factors from affecting the results. We then chose 3 separate non-sense words and used them as anchor text in links back to these domains. We paid careful attention to make sure that the links were always in order such that the first site received links in the order of (anchor 1, anchor 2, anchor 3), the second site received links in the order of (anchor 2, anchor 3, anchor 1) and the third site (anchor 3, anchor 2, anchor 1).
Several interesting things occurred. First, when searching for each anchor, only the page that carried that anschor as the first on a page showed, the others did not. This indicated that, orginally, the search engine ignored the subsequent two links. Eventually, all were listed, but they maintained perfect order respective of the order of anchor text on inbound-linking pages.
2. How Far Does Synonymy Go?
In a search like this one for Bryan Gumbel (purposely mis-spelled), note how Google appears to treat the search as though it were the more likely phrase, “Bryant Gumbel.” In this situation, assuming I really did want to optimize for “Bryan Gumbel,” could I conceptually rank well by having a stronger page than the top few for the mis-spelled phrase, or is Google giving weight to listings that have the proper spelling?
Answer: Google Gives Weight to the Mispelled Page When Not Stemming
This one is dependent upon stemming. Using a keyword like “Bryan Gumbel” is different from many mispellings in that it is a stem: Bryan(t). If you were to mispell “Bryan Gumbel” with an I and keep the T, as “Briant Gumbel”, you would notice that the mispellings rank much better. This is, in may ways, a shortcut method for search engines to deal with posessives, plurals, and contractions and other alterations to root words in the language.
3. Can Link Removals Hurt Rankings?
Let’s say you have a site that an editor adds at Wikipedia. A few weeks later, another editor decides to remove that link – the search engines spider and discover the addition and, later, the removal. Will the temporal loss of a high quality link affect you more adversely than if the link had never been placed at all? Is removal from Wikipedia (or other high authority sites) a factor in hurting rankings more than simply taking that link’s juice out of your site/page’s strength? (note – Wikipedia might not be a great example, since links are often added and later removed)
Answer: Yes (assuming that link is one of the most powerful and few links pointing to your page)
This experiment was fairly simple. Take a page, build about 50 links to it, and see what happens as we “pulse” a front-page link from one of our more popular, powerful properties. Like clockwork, Google ranked and unranked the pages. It was quite dramatic. Of course, the interval was slow (we did a 2 week pulse over an 12 week period)
4. Does Sharing Registrants with Spammers Hurt You?
Matt Cutts recently alluded to the idea that Google knows about all of your domains. Since this is the case, would they be inclined to penalize a high-quality, white hat site if they noticed that dozens or hundreds of other domains from the same registrant were manipulative, low quality spam (assuming you had never attempted to use the spam domains to build link pop for the legit site)?
Answer: No (assuming that link is one of the most powerful and few links pointing to your page)
Don’t ask me how I know this, but variation of whois information, privacy protection, hosts and ips should keep you safe. However, do not register a white-hat domain with the same personal information as a black-hat domain.
5. How Much of the Original PageRank Formula Still Matters?
Obviously, the toolbar PageRank is barely worth watching, but what about that original formula? This very well researched page on the subject recently made the top of Digg and shows through example how PR would be calculated in several linking examples. But, what I’m curious about is what role that portion of the “over 100 factors” PageRank plays. I had been a skeptic for a long time that it was anything more than 5-10%, but recently, I’ve been wondering if there might be a bit more weight to it than that (possibly a full 20-30%).
Actually, the relationship between PR and ranking is synergistic and dependent upon the other factors – namely domain age and no penalties. I do not have a very good answer for this one. Sorry.
6. Yahoo!, MSN & PageRank
Let’s say that Yahoo! and/or MSN decided to adopt an exact replica of PageRank in their algorithms. Yes, I know it’s patented, but how would Google ever know or find out so they could issue a cease and desist or lawsuit? Do you think they could puzzle it out to a degree that it could serve as hard evidence? Even if it’s not happening now (and I doubt that it is), is it possible that years ago, a major search engine did decide to copy Google’s patented methodology?
Answer: Who cares?
Just kidding. It is an interesting hypothetical, but the truth remains that the ranking factors for the top search engines appear to greatly differ from one another.
7. Text Placement Weighting
Remember Michael Murray of Fathom’s quote from last week? While I’m not a big believer that text placement in code is given any significant measure of weight in the algos, I do wonder about how the content structure of a page might impact rankings. For example, let’s say you had three unique article pages with three paragraphs each and each had one paragraph on lemmings. If the first article had the first paragraph on lemmings, the second made the middle paragraph on lemmings and the third used the last paragraph for lemming discussion, which one would rank best, or would it matter at all?
Answer: Yes, content at the top matters more.
It blows my mind how many sites get thrown into duplicate content hell just because the top 30 lines of content on every page are a nav bar, a company name, slogan, and login form. It is ridiculous. Nevertheless, on one of our major properties we moved the “tags” for the article to the top of the content of the page, rather than the bottom. Rankings skyrocketed.
8. Higher Links = More Weight
Take a page with 6 outbound links, all to different, never-before crawled web pages, and use images as your links (thus giving no anchor text). If all of those pages were targeting exactly the same nonsensical KW phrase and each had precisely the same degree of rank-worthy content (though not duplicate), would the higher-placed link (link 1 of 6) from your linking page carry more weight than lower-placed links (link 2 or 3 of 6)?
Answer: YES YES YES!
This was the primary method used to create the “Five SEO Excuses” campaign we ran a few months back. The phrase “Five SEO Excuses” was used as anchor text for 5 links placed sitewide on a Virante property. This keyword phrase did not occur at all on the original five sites we were attempting to rank. This is what happened…
Anyway, with a little research and a lot of time, there are good answers to most of these questions.