WARNING: This post is very long
Just wanted to share some good news here - one of my (many) sites hit on March 15th is seeing a nice 20% up-tick in Google referrals since yesterday. Googlebot's activity has also increased 10 times.
This is a far cry from full recovery (lost 90% of G* traffic in March) but the first upward movement in almost 2 months.
Basically, what I'm seeing is a return of the same keywords to approximately the same positions (although they now tend to fluctuate a lot, as many pointed out here). Just as if the data about the pages was lost for two months, then some of it found.
I have also done some analysis of my G* referrals before March 15 crash and after. What I was specifically looking for is whether the type of the keywords that continued working after March 15 differs in any way from those during "normal" times. Meaning, has the traffic become more "long tail" or less. In this study I sided with those people that measure the "length" of the "tail" not in the number of words but in the number of searches for those keywords. This is because some of my best performing KWs would be ridiculously long tail if you measure in words - 5 to 7, and that was the case both before the crash and after.
So, I took the total number of keywords I received G* traffic on and came up with percentages for the keywords that only occurred once (produced referral once) in a month, then 1-5, 5-10, 10-50, 50-100 and more than 100. For all intents and purposes those above 100 are the "shortest tail" or "trophy" and 1 - the "longest tail".
In addition, I looked at two sites - one that has been affected and the other one that has not.
I'm going to have to spare you a lot of details (that I don't feel comfortable sharing in the open anyways) but, after a couple days of grep
ing and sed
ding the end result was the exact opposite of the hypothesis I started the study with:
- The affected site was slightly less dependant on long tail than the non-affected one
- All percentages for keyword frequency ranges for before the crash and after were very similar, within the last significant digit or so (i.e. 0.015% vs. 0.011%), in other words no significant change in "length of tail" was observed.
- Non-affected site was no less different in the percentages for keyword frequency ranges than the affected one, in other words, any difference can be explained by randomness.
The only thing that was actually very different was the total number of keywords that sent traffic. So, I'm squarely back in the camp that believes in G* loosing ( retiring, invalidating, corrupting, whatever it is that happened) some part of its index during the move to Caff infrastructure. Sh.. happens, so to speak.
They do seem to require heavy re-indexing in order to restore that lost data and they are doing it top-down, meaning, the top dogs (who's amount of backlinks still makes me cry) get the crazy re-indexing first and we, bottom feeders, get it later. Could be that the spammers and the penalized folks may not even get the re-indexing thus falling out of the index in a "natural" way.
The last paragraph is completely my speculation - if you have another way of interpreting the results I got, I would be very interested in hearing your point of view.
P.S. if anyone's interested, I ran the same compilations for Y! and Bing referrals for the same periods. Both show slow but steady increase in the total number of referrals and the type of keywords is remarkably steady across the three month period I looked at. So, it looks like Google's changes are working. For Microsoft.