tedster - 6:49 pm on May 23, 2010 (gmt 0)

Interestingly, all articles written on Feb 1 and 2 were ranking well. Jan 28,29,20 did not rank at all - while I must say I still could not find a pattern, it is interesting that there was a clear demarcation at this point.

Thanks for that - I'm going to take a closer look for the same.

Connecting words (eg: in, at, on, among) seem to be having a lot more relevance now

Yes, I am definitely seeing that pattern. Earlier in this Mayday discussion [webmasterworld.com] I made an educated guess that something had changed in the phrase-based indexing algorithm, and I based that guess on this kind of observation.

Suppose you search on a 4-word phrase (a 4-gram) - word1 word2 word3 word4. If a page on a strong site contains word1-word2-word3 in one group and word4 appears elsewhere on the page, it can still outrank pages with the full phrase on "weaker" sites, but not as easily as before.

This pattern seems to include "connecting words" more often - words than might previously have caused the full phrase to be ignored as "a good phrase". The observation points to a shift that is quite complex in the indexing/ranking logic. Maybe something that required more computing power to bring live?

It certainly has shaken up some long-tail traffic patterns. Other reports in this thread show both directions of long-tail change, significantly up and significantly down.

Interestingly, very large sites with lots of UGC seem to be the ones that went down, (admittedly this is based on a very small sample). I wonder if that drop was because pages with the exact, full phrase now get "more credit", even when other signals are relatively weak.