February 2003 Volume VII Issue Two

A Triumph for PageRank?
Just how important is Google’s PageRank algorithm – by Leslie Rohde

Question: Why does Nokia, with 12,700 incoming links and a PageRank of 9/10, rank only #3 in a Google search for cell phones?

It’s a very good question. Particularly when considering that the #1 and #2 pages have lesser PageRank and only a paltry (by comparison) 764 and 426 incoming links. Even more puzzling, after analysis, is the fact that Nokia is showing up at all in the top five. Here’s why.

The importance of PageRank

Although a great deal of hoopla has boiled up around Google’s PageRank algorithm, there are other elements affecting positioning within Google’s search results. Often they are overlooked or even dismissed entirely. It is widely reported, for instance, that Google uses more than 100 different measurements in determining the position of any given page in the search results. PageRank (which is often abbreviated as PR) is just one of those measures, albeit an important one. But, just how important is it?

As reported in the Dec’02 SE Bytes, the Nokia main page, despite a PageRank of 9/10, does not rank at the top of Google for the search phrase cell phones — being beaten by pages with a lower PageRank. It was suggested that Nokia’s incoming links’ anchor text might be the culprit and the thought occurred to us that an analysis of Nokia’s link structure might provide an excellent opportunity to illustrate the good and the bad of incoming link structure in respects to Google’s ranking system.

An analysis of Nokia page’s ranking metrics for keyword cell phone

Let’s take a look at the top five cell phone pages out of the 2.66 million reported by Google. In addition to the position, site,title and PR for each of these pages, the table below shows the total number of links pointing to each page (link popularity), the on-page keyword density for each word for the search phrase, and the percentage of links that contain each search keyword (link reputation).

Reference: OptiLink gathered and crunched the data for the last three columns.

The possibility that Nokia’s link reputation might be a relevant deficiency is borne out graphically by the figures in the last column — the keywords cell phone are found in none (0%) of their incoming links. Never mind the question, why isn’t it #1 — based on that finding, one could wonder why it ranks in the top five at all!

Furthermore, there is nothing at all in either the Nokia page title or text links that suggests Nokia even makes cell phones. Of course, the figures on-page and link text measurements are rounded values. So, the Nokia site really does have some relevant text linking but the percentage is displayed as zero because less than half-a-percent of the links contain the search term. And because of the massive number of links (12,700), even the fractional percentage represents a significant actual number of relevant incoming text links.

Regardless, far and away the best asset, search-wise, of this site is its high PageRank (9/10). By every other measure it doesn’t even belong in these results. So it could be said that we’re seeing a “triumph” of sorts for PR because, obviously, the Nokia site is highly relevant for the search phrase cell phones. That means that PageRank did it’s job by elevating an otherwise less relevant page toward the top of the listings based on the element of overall popularity as Google measures it.

The fact remains, however, that it is an expensive way (resource-wise) to win the SE game. The high PR of the Nokia page is mostly achieved through the use of many hundreds of internal pages. It’s a huge site. By contrast, the other pages in the top five achieve their position using a much smaller number of links. How do they do it?

How the smaller sites compete with the behemoth

The vast majority of links on the Nokia site are graphical anchors (image links) and Javascript buttons that provide no text for the search engine to analyze. Such links offer no assistance to Google in determining the site’s theme and reputation. In other words, Google can’t tell what the site is about by looking at the image or javascript links.

In contrast, all of the other pages in the top five enjoy a significant percentage of incoming text links — specifically, text links that contain the exact search term (cell phones). This is the difference that provides these pages with a highly relevant theme or link reputation that Google takes into consideration as it ranks the relevant search results.

Here’s what Nokia should be doing . . .

So how would we help Nokia take the top spot? Well, we could use some link text (duh!), but let’s consider something really simple for a moment: the title text (double-duh!). Since pages with the exact search expression in the title enjoy considerable advantage at Google, it only makes sense to add the phrase cell phones to the HTML title of the Nokia page. We’d suggest Nokia Cell Phones as the exact text that should be placed between the <title></title> tags.

This single change alone, when combined with this page’s currently high PR=9/10, would probably place it at the top of the results. Add-in some keyword-rich link text and the page would be virtually unbeatable in a search for the phrase cell phones.

Here’s how YOU can apply these lessons…

OK, enough about Nokia — how do we use this example to explain and improve our own positioning?

For starters it helps to understand that search engine ranking is not voodoo! …but it isn’t an exact science either. Rather, it exists on a plane of study in a way similar to anthropology, sociology, and psychology. Of course we’d all just love for SE ranking to be as orderly as physics but instead we must learn to work with inexact characteristics that are often more practical than precise.

Precise or not, we all want to know what it takes to get top positioning. The problem is, of course, for every search the answer is slightly different. Fortunately, the search engine tells us (approximately) what we need to do. We just have to make our page characteristics slightly more optimized than the page currently listed on top. The other pages show us various combinations of features that didn’t make it and their positions in the ranking give us clues as to which combinations are more likely to win than others. By looking at enough of these, we see patterns emerge that we can use to predict ranking.

Let’s take another look at our example and see if we can use it to explain, and subsequently predict, ranking.

The #1 page,�despite having a lower PR, appears to beat the #2 page because of superior link text and a title tag that contains the search term. It appears to beat the other pages in the top five the same way.

Note however the very close match between the #1 and #4 pages. The on-page and link text measures are too close to call. The #4 page does have more links, but the key difference seems to be the title text: the #1 page has the complete search phrase, where the #4 does not. We’ll return to this in a minute.

The opposite situation is presented when comparing the #4 and #5 pages.

While the title on page #4 is incomplete (it contains only half the search phrase), it has enough of a lead in link count and link reputation to overcome the better title tag on the #5 page. These two comparisons, 1-4 and 4-5, illustrate that Google is balancing multiple criteria in ranking these pages.

Ranking by Example

Let’s put this study to work. Suppose we want to make the #2 page beat the page currently at #1. What do we have to do? The search engine gives us a pretty good idea.

We could tweak the title a little bit — changing “Phone” to “Phones”, increase the search term frequency on the page, and modify a bunch of our links to include the search phrase. Since page #2 already has a higher PR than page #1, we should be able to take the top spot with on-page and link text numbers that are fewer than what the current #1 page has. In regards to the link text, it is likely that simply changing the internal links on page #2 to say something more meaningful than “Home” would be sufficient to increase our link reputation to at least 75%.

But suppose instead that we want the #5 page to be top ranked, how would we do that?

Our title looks good, and our on-page measures are close enough that we probably don’t need to bother with them either, but we need way more links and they need to contain the search phrase whenever possible.

Okay, so we need more links. But about how many will do? That depends. It’s a balance of multiple factors. If we beat the top page in link percentage, then we should expect to win top spot with fewer total links — all other factors being equal. In this case, if we could attain a 100% link relevance then a link count of less than 764 — say, something around 600 — should be sufficient.

Better yet, if we can increase our PR in the process of adding links (which would normally be the case), we should be able to do even better — maybe with as few as 400 links.

What about the other 2.7 million pages?

Obviously, this strategy is not limited in use to the top five or ten results pages. So, let’s take a look at the page at position 22.

Looking at the table it becomes clear that the link text percentage is too low and needs to be increased. Furthermore, our OptiLink analysis revealed that at least half of all the incoming links are either Home Page or World of Phones — neither of which will give the page a boost for the keyphrase cell phones.

Secondly, the on-page keyword density is also too low when compared to the page in the #1 position. It clearly wouldn’t hurt to increase the on-page keyword density for the desired search phrase (i.e., cell phones)

So, where would this page rank if 80% of the incoming links contained the search term? Our best guess is it would likely move up to #5. In any case, what’s interesting is that this page is a great deal closer to being the #1 page than is suggested by its current position in the search results.

Ranking is non-linear

Look at our top five pages and ask yourself which one is easiest to move to the top spot.

• Is it #2? No. That page needs a “link overhaul”. It has good PR and total link count, but in all other respects it is not well optimized.
• What about the #4 page? If all we did was change the title from Cellular to Cell it might just rank #1. Despite being lower ranked than the page at #2, this page would take less work to move it to #1.
• Similarly, Nokia could very likely move from #3 to #1 by adding two words — cell phones — to their page title.
• Finally, compare our #22 page with our #5 page. Both of these pages need a lot of help in their linking, but the page at #22 has far more links to work with. So, this so-called lower positioned page is actually the easier of the two to optimize successfully.

In conclusion . . .

From a strategic viewpoint of brute force, PR is one way to overcome poor optimization. Conversely, a well optimized page — one with better topic and link reputation — will often beat a page with high PR. Stated simply; efficient optimization still beats high PR in most searches.

But what about those cases where high PR pages beat highly relevant pages? Even then, building a smaller number of quality links will almost always be the more cost-and-time effective strategy when compared to building (and maintaining) the many hundreds of links required to beat the Nokias of the world.

Within your own arena of competition, PageRank may or might not be a critical element in vying for Google positioning. Regardless, the only way to know for certain is to measure your top competitors’ link and page structure so you can custom-design your own linking strategy to beat them.

Best Regards,

Leslie Rohde – Windrose Software
Creator of OptiLink