Lots of people criticize the reliability of the Alexa toolbar – see the comments on Matt Cutts’ blog or more recently from Om Malik. I would like to offer some qualified support for the Alexa toolbar – with some data to back it up.
At SLI we run the site search for lots of sites that have a wide range of Alexa rankings. Assuming that the number of site searches on a site is roughly proportional to it’s traffic we can use the site search data to test the accuracy of the Alexa data. This graph plots the Alexa rank against the number of site searches performed over a 30 day period for about 50 sites (randomly selected from our customers).
As you can see, there is a strong correlation between the Alexa rank and the number of searches that happen on a site, although there is a reasonable amount of variation.
We use the Alexa rank to get an order of magnitude estimation of the number of searches for a site, and it’s normally pretty good.
The assumption that number of searches performed on a site is proportional to traffic isn’t going to be always right. Some sites lend themselves to searching – particularly larger sites.
Some sites have completely unreliable Alexa rankings – particularly those that attract more (or less) people that have the Alexa tool bar installed. Case in point is the SEMPO web site. As others have pointed out, a higher proportion of search marketers have the Alexa toolbar installed. The SEMPO site has an Alexa ranking of 19521 (today) and had less than 5,000 searches over the last 30 days. This model would predict around 150,000.
I think Alexa is a useful service, providing that you use it carefully. I think there are a lot of people out there who have dismissed it as flawed and may be missing out on some valuable data.