Very unusual SERP phenomenon

[g1smd]

I've seen the oddest thing in the SERPs a few minutes ago.

For a SERP initially reporting nearly 400 results, there's links to 19 pages of results along the bottom of the page.

Clicking through this initial search reveals only 11 pages before running into the usual "omitted similar pages" Duplicate Content warning.

After clicking this "show all results" link, there's 10 results on each of the first five pages.

At this point there are links to 19 pages of results along the bottom of the screen. So far so good.

Page 6 has only 6 results.

Page 7 has 10 results.

Page 8 has zero results.

At this point there are links to 17 pages of results along the bottom of the screen.

Page 9 has only 4 results.

Page 10 has only 3 results.

Page 11 has only 9 results.

Page 12 has 10 results.

Page 13 has only 9 results.

At this point there are links to 18 pages of results along the bottom of the screen.

Page 14 has 10 results.

Page 15 has only 7 results.

Page 16 has only one result.

Page 17 has zero results.

Page 18 has zero results.

After this there are no more links to further results.

[edited by: tedster at 3:21 am (utc) on Aug 21, 2011]
[edit reason] moved from another location [/edit]

[Robert Charlton]

g1smd - I think that's significant.

It suggests a profoundly different way of returning content when sites have become so similar that distinctions between them are meaningless.

Not sure whether this is a test or a new feature in action, and unfortunately right now I'm in the process of running out the door. That only 400 results were returned initially suggests that Google wasn't seeing enough content worth returning without refinement of the search.

[tedster]

I'm wondering if this search result still showing the same strangeness, or if was it some kind of bug that's been fixed. As Robert said, it could be showing us something or other, and sometimes a bug has quite a story to tell if we can figure out how to listen to its story.

[lucy24]

Yup. Seen that, allowing for orders of magnitude ;)

It's a computer thinking like a human-- but not in the way we mean when we talk about wanting computers to think like humans. This just comes across as "Oops, I thought I had another eleven of these but I guess there are only six." Can't it count to 400? Does it want us to think it's generating results on the fly? A page and a half of its results were removed from the database at the exact instant that you asked to see them?

Anyone other than g###, you'd say "Nice try, but spend some more time on basic coding before you try for a public release."

:: idly wondering how many results would really come up if you were able to call its bluff on the "Showing 1-30 of 1,400,000,000 results" line ::

[Sgt_Kickaxe]

Page 2+ seems to be Google's testing grounds where they gather information about web pages using real visitor behavior. They try to return the top 10 sites on page one but after that the decision engine is likely getting 'confused' by secondary effects such as page testing etc.

ie: we want to show page example.com/blah-blah roughly 15k times in spot #13 of the serps to get a baseline of this pages value to visitors in that position, will adjust accordingly afterwards.

I have several years of data that show Google displaying my content during specific time periods but not at other times, in some cases the content is page 1 and the see-saw effect is extreme. I'm basing the above on a comparison of MY server traffic stats vs what analytics is reporting, it's NOT the same.

[g1smd]

It lasted a day or two but has gone now. I am guessing that various server clusters weren't completely synced.

I have seen similar things several times before (years ago), but not to this extent. It was very odd to see.

[tedster]

These oddities do come up - I remember seeing the bottom result on every page (that is, #10, #20, #30 etc) being the exact same URL, title and snippet. Sometimes these temporary bugs show us something useful about the ranking process, but more of the time it is about coordination of data centers or a coding issue in the display generation area.

[Robert Charlton]

g1smd - Sorry to have been away from this for so long. I realized shortly after I'd asked tedster to move your post to this new thread, that, in my haste, I'd both misread your very clear description of what you were seeing and also edited "possible glitch" out of my original comments. You've since answered the obvious follow-up question, about whether this search was repeatable. So, this isn't nearly as significant as I'd thought when I posted, and it's more about "coordination of data centers" than about ranking... but, FWIW, here are some thoughts....

I'm surprised that the results lasted as long as a day or two. To someone more familiar than I am with the inner workings of the Google database system, consistency for that long a time might not be surprising. It suggests to me, anyway, a cycle of meshing of various databases involved in your particular search that wasn't often refreshed.

For a SERP initially reporting nearly 400 results...

To create searches in a range that small, I tried to produce analogous conditions by searching with unusual or made-up words and word combinations, but I couldn't reproduce anything close to the pattern you reported.

Generally, when not very many good matches are found, Google will stretch spellings and matching criteria if they come closer to fitting other query sets that are in the index. With spelling corrections or not... on searches where not many pages are returned, I did see that Google apparently needs to work harder. Most of these test searches for <400 result queries were taking times in the range of .36 sec to .39 sec, instead of times hovering around .2 sec or less that are typical where I am for queries returning many millions of pages.

Thinking about what might be going on here... I'm assuming that Google partitions and distributes its databases for different degrees of availability and performance. It's likely that common searches, therefore, are optimized for speed, with perhaps redundant copies of lookup tables for most common searches distributed around Google's data centers.

Uncommon searches, on the other hand, might have less priority on speed and more on conserving system resources... and thus a sparser availability of lookup tables, which would more likely result in the kind of anomaly you saw. Perhaps not new thoughts to you, but it is what I'm guessing might be happening.