What Is A Link Worth? Part 1: Valuing PageRank

Every serious search marketer instinctively understands that links are hugely important. But it’s difficult to quantify the value of link-building efforts in a systematic way, especially in a way that can justify the expense of link building efforts to clients or management. This post is the first of a series that will attempt to quantify the value of links—in this case, by measuring the value of links in terms of other links’ PageRank (e.g. how many PR4 links is a PR5 link worth). A later post will cover market pricing of links with statistics from various paid link markets, and other posts will cover what links are worth in terms of effort and resulting traffic. By the end of this series, we’ll have a complete model for valuing linking activities and determining their ROI.

The problem with valuing links

Some years ago, several SEOs calculated and published PageRank tables based on Brin and Page’s original paper. However, I couldn’t find any evidence of anyone ever doing measurements to check or calibrate their tables; you can easily find some if you do a Google image search on “google pagerank table.”

Most of them appear to show that each level is worth 5.5 times the previous level—for example, a PR5 link is worth 5.5 PR4 links and so on.

To verify this, it seems to be extremely difficult to figure out how or why an individual page has a particular PageRank, short of spidering the entire web yourself and reproducing the PageRank calculations yourself, or using a tool such as LinkScape or Majestic-SEO.

A clever trick we can use

However, there is a way to examine entire websites (or at least, their home pages); Google already provides PageRank to us of a sort, via their Toolbar PageRank metric. Google has also already spidered the entire web for us. With a simple Google query, it is trivial to figure out how many backlinks a website has (the industry-standard notation of brackets means that you should type in every character you see between the brackets when doing your search):

[“searchengineland.com” –site:searchengineland.com]

This query will return all web pages that reference “searchengineland.com” that are not located on searchengineland.com domain itself. This includes web pages that have links to the home page as well as web pages with links to deeper pages. It also includes references to the website that are not a link (i.e. if someone references searchengineland.com in an article but does not make it a hyperlink), but many in the SEO industry speculate that the search engines then convert that into a link and count it, which I am also assuming here.

The theory

Knowing that Brin and Page’s original PageRank paper specifies PageRank is a logarithmic measurement, I then theorized that a graph of website home pages’ toolbar PageRanks versus the log of the number of backlinks to those sites should be roughly linear, could give us insight into the value of links at different levels of PageRank, and could perhaps aid us in building a PageRank table based on actual measurements.

One might argue that for small websites, the PageRank of links coming to them might diverge significantly from the average (for instance, a small website may have a few PR8 links and get a big boost in PageRank), but intuitively, large sites should not diverge from the average by too much. The larger you get, the more you should resemble the average website—if you have millions of links, they can’t all be PR8 links and are more likely, on average, to be similar to the “average” link profile of the entire web.

So, I decided to simply measure the Toolbar PageRank of 50 websites and to plot it against the number of backlinks that they each have, on a logarithmic graph. One would think this graph would exist somewhere in the industry but I have never seen one (if you know of anyone who has done this before or any papers, please comment below).

What I got was the following remarkable graph:

Toolbar PageRank versus number of backlinks

Testing the theory

Any good theory should have some predictive value, so, armed with the equation for a linear fit to this graph, I checked how many links there are to a co-worker’s new website (it’s a great place to Buy Baby Announcements) and found it had 177. The equation predicts the website, or at least, its home page, should have a PageRank of 2.6, and sure enough its Toolbar PageRank shows as 2.

Try it yourself with a few sites: take the log of the number of backlinks to a website, multiply by 1.4063, subtract .4747, round the result down, and check that against its toolbar PageRank.

A caveat:It’s important to note that a site may be ~1-2 PageRanks above or below its calculated value depending on the average PageRank of the links coming to it, i.e. if you are doing a great job and have very high-PR links, your site will be above its predicted PageRank. Quite often though, this simple equation is surprisingly accurate.

An interesting observation

The high-toolbar-PageRank sites that were below the line (i.e. whose reported PageRank seemed lower than it should be based on their links) included Dailymotion and Wikipedia—notably, both user-generated content sites. High-PR sites that were above the line included usa.gov and cnn.com. This is certainly not a significant sample but suggests some further study on UGC sites versus news, government, or .edu sites is probably worthwhile. It could also argue for a “hidden variable” in the numbers such as TrustRank (Google discloses on their website they use 200+ variables in their organic ranking algorithm on their website so this seems highly likely).

So, how much is a link worth?

Well, we can’t say what a PR1 link or a PR2 link is worth in absolute terms, since the equation measures the “average” value of links on the web. However, the slope of the equation is the same regardless of how you value a PR1 link, so we can say with some confidence that each level link is worth (drum roll please…..) 5.14 times the previous level. Those old-time SEOs armed with the PageRank paper calculations were not too far off with their estimates of 5.5 after all!

“Average” number of links required to reach each toolbar PageRank level

So, per actual measurement, a PR3 link is worth 5.14 PR2 links, a PR7 link is worth 5.14 PR6 links, and so on. And as it turns out, the SEOs of years gone by that calculated their PageRank charts weren’t too far off with numbers like 60 million or 80 million links to become a PR10 site. On a log scale, those are really close to 28 million. Note that all the PR 10 sites came in above the curve, so for instance, one made it with only 11 million links—but all of this is close enough for our purposes.

This is of great significance to link-building campaigns—if you have a choice between sending an email to ten webmasters requesting 10 potential PR4 links, versus two webmasters requesting a single PR6 link, it actually is more efficient to spend significant time crafting your email for the PR6 link as it is worth 5 x 5 = 25 PR4 links.

Future posts will dig closer towards placing actual monetary value on links. Hopefully this series will inspire others in the industry to do some analyses in this area which sorely needs the attention!

Postscript: Caveats and assumptions

If Google reported toolbar PageRank to an additional decimal point, then it would probably show that some sites are really 10.1, 10.5, 10.9 etc (think about it—two PR10 sites clearly don’t have identical PRs). The only effect of accounting for this would be to shift the Y-intercept up; if you assume all the PR10 sites were really “average” PR 10 sites, then they should be, on average, PR10.58—halfway to 11 on a log scale. All the PR3 sites would be assumed to be PR3.58, and so on. Although the graph as shown does not reflect this assumption, the equation disclosed does include this adjustment—and either way, it has no affect on the slope which gave us the “each level is worth 5.14 times the last level” factor.

This entire exercise also assumes that Toolbar PageRank actually means something (it’s still rather vague, and the industry has no clear consensus on this point). It also assumes that Google’s “520,211 results found” metric is actually reasonably reliable.

Finally, some PageRank 10 sites that were considered were excluded if they would obviously warp the calculations, such as Adobe (due to its Reader) and Mozilla.com.