Site keeps yo-yo-ing. How best to troubleshoot?

[1script]

Not sure if "yo-yo-ing" is an actual word but I hope you got the idea. G* traffic drops 90% one day. 10-14 days later it returns. Then 3-4 days later it drops again to the same level. 10-14 days later it returns ... And so on.

This has been going on since June. At first I though might have caused the effect myself by starting to work on the site when the traffic returned (well, you know, an overworked webmaster paying attention to only those sites that earn any money ...). I was doing some minor changes to page layouts, move some ads around, that kind of stuff.

After the second cycle I decided enough is enough and just kept watching, no changes to the site made during neither low nor high times of the cycle. But the cycles still continue.

G*bot visits site as usual, pretty regularly. There are peak visits sometimes but they neither coincide with nor precede/follow the traffic cycles.

Can anyone offer an idea about the best course of action in this case? I would like to get to the bottom of it and stabilize the traffic somehow but not even too sure about what are the telltale signs I'm looking for.

All the parameters I'm usually looking at - PR, BLs, # of indexed pages via site: or via GWT all stay pretty stable. All other sources of traffic are very stable, too. Only G* referrals keep doing this yo-yo thing.

Any idea is greatly appreciated!

[maximillianos]

Perhaps G is picking up some new pages on your site that periodically rank well for a short period of time.

It would be helpful to know what type of volume you are talking about. 90% of 100 is only 90 visitors. Not much data to work with. 90% of 100,000 is quite a reasonable amount of data to work with.

[1script]

@maximillianos:

Yes, I know what you mean and should have given that info in the first post.

Traffic goes from 1200 uv/day to 120 uv/day and back, give or take 100 on the high side, 10 on the low side.

Regarding which keywords are ranking: as far as I can tell, the very same keywords that ranked during last "high" time rank the next. None of the high performing KWs are ranking anywhere within the first 50 results during low times.

It looks similar to the loss of long-tail traffic that was reported en-masse after Mayday except this site had suffered that long before - on March 15th and was on the low side up until the oscillations started in June.

In fact, I think that by comparing log files during low and high times I can glean some insights useful for other sites that lost long tail traffic. I'm just not too sure what would be the most useful bit I'm looking for - difference in the number of ranking KWs during high/low times or perhaps instead find KWs that don't move (if there are any) and dig around those for BL anchor texts, title tags and such.

What would you look for?

[aristotle]

Is the site close to a borderline between penalty and no penalty, so that random events cause it to move back and forth across the line?

[1script]

Is the site close to a borderline between penalty and no penalty, so that random events cause it to move back and forth across the line?

I wish I knew an answer to this one ... Anyways, I think you meant "filter" as in automatically applied "discount" as opposed to human reviewer penalty so to speak. I guess, it is possible. It may actually be possible that G* is tweaking those filters as we speak and this site finds itself on different side with each tweak.

Anyways, the only way I know to check for a penalty: search for the site's domain with the quotes and without, with the .com and without, shows the site's homepage on position #1 - so looks like it's fine as far as penalties are concerned. If there is another way to look for a penalty, I'd be glad to learn it!

[tedster]

Before June did you have consistently high rankings? Or is the top of the yo-yo the first time you got to the top?

[1script]

@tedster:

The timing goes like this: the current top level held for couple years before March 15th. On 03/15 it went down to 10% of what it was. Then on June 10 it went up almost to the former high level and started oscillating after that.

[tedster]

In that situation, I'd look off-site first. Something may have thrown your backlink profile into a lower trust situation. Look for strong backlinks you've lost (be sure your link target URLs are working correctly!). Then look for a new flood of crappy backlinks. Finally, work for a few real quality backlinks of diverse sources. Depending on the competitiveness of the SERP, I've seen even one solid new backlink stabilize things.

If you never ranked well, then there's little for you to diagnose - you're getting a sort of try-out or audition on the first page. What I'd do in that situation is monitor the bounce rate and A/B test the heck out of the page to bring it lower. Enhance the value for the page as much as you can.

[Hoople]

You may wish to rule out a Google dataset issue. Record keyword search results for a few Goole data centers during the high and the low period.

Does Google Webmaster Tools give any indication of crawl errors or html suggestions?

[mromero]

We are seeing something akin to this yoyo thingy. One site that we monitor is very high on page one when we check it on the Caffeine I.P. posted somewhere else here.

Using the Google Global extension in Firefox today we noticed that the rankings in the U.K., Canada, Australia etc. match that of Caffeine.

However running it on the U.S.A. the ranking is distorted and just like the old rankings of about two months ago with mostly government and government funded websites at the top. The site we monitor is private and independent of government control and devoted to general, news and travel information of a country.

This country BTW has a low ranking in an index on corruption that is maintained by an independent and highly respected foreign policy watch group in Washington.

Logic would tell me a search engine would consider this a factor in ranking government websites from countries where the government is known to be engaged in funny business - expropriating private business it does not like, installing family members in high positions of power, doling out million dollar website contracts to connected individuals - you know - the old drill.

So from our standpoint, at this moment in time in the niche we monitor, Caffeine and the Google results from most other countries have it right.

The results from the U.S.A. do not.

[1script]

Thank you for your input, guys. Based on your suggestions I would like to add more data here.

GWT does show some crawling errors but it is a relatively small amount, certainly less than on some of the other sites that work reasonably well. I think I can rule out crawling errors.

Regarding geographical distribution of visitors during high and low times - I have verified that and the percentages are all within less than 1% difference. The traffic is and has always been predominately US and it stayed that way all this time, so the drop in traffic is spread evenly between all Google DCs

Loss of important links - nothing I track disappeared throughout all this time since March. However, junky BLs are a whole different matter. I have sponsored a few WP themes in the past and now my links are all other the Net. I was sort of hoping that the themes will bring, well, themed links but that did not really happen - the blogs that link to me via those WP theme "Sponsored" links are about anything one can think of. Am I now forever doomed to fight those junky BLs? There is approx 150,000 of those, there is no chance in hell I'll be able to acquire this many legit links to balance those out, at least not in a reasonable time frame. Unfortunately, most those links point to the homepage so I can't really 403 it to stop the flow of those links.
Any idea about fixing the sins of the past?