Panda key algo changes summarized

[pontifex]

Folks, I have been reading a lot, thinking a lot and analyzing a lot. I am still not sure, how to get the US traffic back to pre-24th of February levels! But I think it is time to summarize the key theories of the algo change in the US:

- Internal links devalued, only external count really

- Thin pages cause substantial bigger problems for a domain

- Duplicate content snippets on your page cause substantial bigger problems

- Too many external named links "widget keyword" instead of "more..." (eg) cause penalties


are what kept me working in the past 4 weeks. Do you have some additional meme?

P!

[crobb305]

I took a look at a few patents and now it's getting all clear

2macarena

Good move :) That is what everyone should be doing. I firmly believe that the secret is in the sauce.

For anyone interested in researching the patents or just reading bullet summaries, [seobythesea.com...] does an outstanding job. Their documentation, analyses, and summaries have been referenced on WebmasterWorld many, many times. The patents are impractical to read line by line (it would be hundreds/thousands of pages over the past 10 years); but the summaries can really get you thinking the way Google thinks (statistically and non-linear).

Also, here are some of those patents you mentioned:
"Document Scoring Based on Document Inception Date" filed on Nov. 20, 2006 (by Matt Cutts)
"Document Scoring Based on Document Content Update" filed on Nov. 21, 2006
"Document Scoring Based on Query Analysis" filed on Nov. 22, 2006
"Document Scoring Based on Link-Based Criteria" filed on Nov. 30, 2006
"Document Scoring Based on Traffic Associated With a Document" filed on Nov. 30, 2006
"System and method for modulating search relevancy using pointer activity monitoring" filed on July 13, 2010

[Jane_Doe]

While sites I haven't touched for 2 years that I couldn't give a crap about have been given a boost.

I have the same experience. I have a small site that has not been updated in years ranking in the top 20 spots where many of the other spots are national magazines or nonprofits. The magazine above my site in the serps for a reasonably competitive two word term has 9 million backlinks in Yahoo. My site has less than 30 backlinks for the whole site. Thirty backlinks versus 9 million, and those 9 million include many national magazines and other authority sites.

Clearly, at least in some cases, Google is looking at scoring metrics way different than what we are used to these past few years. So how is this site ranking for this key word? In past years this never would have happened, and now it has not only happened, but happened without any action on my part.

[tedster]

I took a look at a few patents and now it's getting all clear

We've also got some good reference links about Google patents and academic papers at the bottom of the Hot Topics page [webmasterworld.com].

[johnhh]

"Document Scoring Based on Link-Based Criteria" filed on Nov. 30, 2006
"Document Scoring Based on Traffic Associated With a Document" filed on Nov. 30, 2006
"System and method for modulating search relevancy using pointer activity monitoring" filed on July 13, 2010

The dates are interesting - 2010 is a complete move away from webmaster control through links,content,traffic into user areas.

_{thinks - maybe there was a reason for collecting all that wifi data personal data}

[DirigoDev]

Anyone looked at the ratio of internal to external links. I'm tracking the top 200 health websites and see that 84.2% of Panda II losers have very high (uneven) ratios (e.g. above 65%). All of the winners have much tighter ratios (e.g. sub 40%). This is the only common theme that seems to be universal.

On a positive note, 8 out of 8 of the top 80 health websites hit by Panda have recovered a bit week/week between 4/17 and 4/24. This includes those with uneven ratios. That is, all but one which is clearly a farm. None have recovered anywhere close to pre Panda II levels.

[tedster]

Do you mean that the Panda demoted sites have proportionally more outbound links, or that they have proportionally fewer outbound links?

[johnhh]

Do you mean that the Panda demoted sites have proportionally more outbound links, or that they have proportionally fewer outbound links?

more internal links than outbound ?

um it might apply to "sites" where the whole site has been hit - where specific pages have hit I can't see any relationship i.e. I have pages with the same ratio internal to outbound - some have been hit some have not

therefore my theory is that there is another re-shuffle to come ( I live in hope )

_{hence total confusion and headache here}

[tedster]

We do need some clarification on this if we are to discuss it clearly. For example, is this a measure of home pages only - or is it all pages? Even 40% of all links being external seems extremely high to me except for a directory, so I'm thinking I must not be understanding what is being measured.

[TheMadScientist]

And, adding to what tedster asked are you taking into account 'nofollow' links which are dropped out of the graph, or is it links in general?

I think to really make a determination you would have to have a bot that is 'nofollow complaint' and spider each site in question, then calculate the number ... (Do you have one?) ... I guess you could view the source code on each and every page, but for a site of any size it doesn't seem like there would be time for it...

[DirigoDev]

I’m seeing exaggerated ratios in both directions – super externally focused or super insular. Sites that cross the 65% line got hurt almost universally. The obvious farms sometimes follow a different pattern. I know fully that the algo change is far more complicated than link ratios.

I’m looking at the top 200 health information websites (those with the most traffic (the smallest having ~38k uniques per week) that got hit hard. I’m mining Hitwise for the winners and losers. I have two employees doing the analysis by hand full time. We’re a little over 10 days into looking for trends. I’m looking at home pages, top level directory pages and a random sampling of internal pages (~200 pages per site). Sixty-two sites moved up and 58 moved down significantly in the category. That's a lot of movement in the top 200 sites. In all we’re working with 120 sites. We’ve measured links, nofollow links, dates, number of ads, adsense, author credentials, date, meta, GA, breadcrumb type... The nofollows are in SQL but we’ve not crunched the numbers on yet. A quick glance tells me that the nofollow is mostly on the external links (as expected) and that it will not materially change the outcome (will certainly lower the ratio). We’ve not fired up a bot just yet because we’re still making our first pass over the sites – looking at every suspect. We’ll circle back and go deeper next week. I do have a bot. I'll need to retool it a bit for this purpose.

In my opinion, more than 65% of the top 200 health sites hit by this algo change have nothing to do with being a farm. They do have lots of content. So I’m probably looking at around 30 sites inappropriately dinged. Eight (About or WebMd) subdomain microsites lost big while eight in the same family gained big – still trying to figure this one out. I have not looked at link ratios of sites that had very little movement.

Am I barking up the wrong tree or seeing some artifact of some another change?

[browsee]

DirigoDev, thanks for the information. I am just curious about sub domain micro sites. Are you saying that some sub domains of About and WebMd gained , some lost? I have been trying to find out the impact of Panda on subdomains.

[DirigoDev]

I’ve been watching folks post about template design, colors, ad placement, adsense, background colors and missing certifications. I don’t think that this stuff has much to do with the change. I say this because I’m looking at families of microsites that use the same design/templates/advertising model… Some subdomains moved up significantly and others down. WebMD and About are great examples. I'm sure there are lots more examples. How do you explain this?

[DirigoDev]

Are you saying that some sub domains of About and WebMd gained , some lost? I have been trying to find out the impact of Panda on subdomains.

Yes, this is what I'm saying.

I'm using Hitwise to identify winners and losers. Seems to be a pretty accurate benchmark - correlates to change in the SERPs. Change is week ending 4/9 and 4/16. Positive moved up and negative moved down. Rank in health information category is based on traffic.

Yes, I'm saying some moved up and others down. I'm not showing any sites that moved less than five positions. One of my sites moved down 10 positions and we lost 60% of our Google traffic.

Many of the WebMD sites almost fully recovered for the week ending 4/24. Best I can tell, WebMD made no changes. Bet there is a story here.

I hope I'm allowed to post these. I removed the actual domains.

Change | Rank | Company
10 | 54 | The WebMD Health Exchange
16 | 113 | WebMD Arthritis Health Center
23 | 125 | About Dermatology
12 | 168 | About Depression
7 | 169 | About Diabetes
52 | 182 | WebMD Lupus Center
21 | 188 | About Heartburn / Acid Reflux
9 | 199| About Colon Cancer
-7 | 49| About Orthopedics
-17 | 62 | About Pediatrics
-16 | 64 | About - Arthritis
-13 | 68 | WebMD Diabetes Health Center
-19 | 81 | WebMD Men's Health Center
-15 | 107 | About - Sports Medicine
-5 | 155 | WebMD Physician Directory
-9 | 183 | About - Men's Health


Same design. Same graphics. Same ads. Hmmm.

[browsee]

Thanks DirigoDev, I've checked some sites, everything looks fine. I guess they recovered after 4/24.

Looks like Google treating subdomains as separate entities.

[synthese]

Thanks DirigoDev. I run a medium sized health site that got punished severely. Reading your posts with interest.

[crobb305]

I’ve been watching folks post about template design, colors, ad placement, adsense, background colors and missing certifications. I don’t think that this stuff has much to do with the change.

None of what you have shown during your sudden emergence into the discussion disputes what we have been speculating about for 8 weeks. You joined the cause after Panda 2 and after some sites started to see recovery.

You definitely provide useful insight, which we can add to our list of possible factors, but your results do not disprove our suppositions.

Incidentally, some of us who have speculated about those things, and made changes accordingly (following information gleaned from the Google patents and website credibility studies), are seeing remarkable recovery. My traffic is 70% restored just in the past 72 hours. Perhaps they are factors after all...among hundreds of factors working in non-linear/statistical models.

[Planet13]

Sixty-two sites moved up and 58 moved down significantly in the category. That's a lot of movement in the top 200 sites. In all we’re working with 120 sites.

But what about the sites that DIDN'T move significantly? What if your findings say that XYZ is a factor in the sites that moved down, but if the other 80 sites that you DON'T analyze have that same factor, then what?

[Shatner]

Interesting data cropping up here.

The thing to definitely keep in mind is that we're all pretty sure it's not any ONE factor.

Probably just about everything discussed in the six pages of this thread may or may not play a factor depending on the site.

Maybe Dirgo has uncovered another factor to add to the list. I'm a little confused exactly about his specific number though, to be honest.

[TheMadScientist]

I do have a bot. I'll need to retool it a bit for this purpose.

I need one of those! lol

[pontifex]

I have to add - in the light of recovery for some - that our traffic has recovered slightly. I am down now 10% compared to the average of 2 weeks prior 24th of Feb. BUT: My improvements on the site has increased the conversion rate by 20% compared to the same time frame. So, I am back on track because of Panda, which triggered improvements on my site.

The whole user behavior aspect of Panda is the most promising theory so far. There are also some simple improvements you can do along the list we gathered here and they helped for me.

Maybe the conversion rate improvement will impact the ranking or not - yet I feel much better :-)

[indyank]

DirigoDev, you information is interesting but confusing. When you say 65% and 40% what is the base for this ratio? Is it Total outbound links or Total Incoming links?

If it is outbound links, I still have the same question as Tedster as I too feel that even 40% of all outbound links being external look extremely high.

[DirigoDev]

I'm just trying to add value. Telling you what I'm seeing.

The ratio is "Internal Links - External Links / Total Links" I'm NOT subtracting nofollows. In my data, high numbers, both negative and positive, got hit hard.

But what about the sites that DIDN'T move significantly?

I'm working on it.

...results do not disprove our suppositions

I'm not trying to disprove any suppositions. I just want to know how to account for the fact that some About and WebMD sites moved up and others down while having the same design, color, number of ads, etc.

This is a complicated update with lots of moving parts.

[DirigoDev]

I'd like to remind you that I'm looking only at health information sites. This page looks to have 67 links with only one external link - so it too fits my model yet it did not get hit by Panda. There are clearly more factors.

[browsee]

Just checked hubpages in quantcast(directly measured), they are still down. I did not see any significant change in my traffic either. Anyway, crobb305 congrats.

[elguiri]

DirigoDev, great data, and a great piece of investigative work.

I imagine you haven't been able to differentiate navigational links, from body links, from footer links, and are just working with total link count.

[randle]

The whole user behavior aspect of Panda is the most promising theory so far.

Certainly an algorithmic dynamic we have all been thinking about for some time now, before Panda. Anyone out there got any sort of real facts or research to support it though?

[crobb305]

Just checked hubpages in quantcast(directly measured), they are still down. I did not see any significant change in my traffic either. Anyway, crobb305 congrats

I want to mention a few things. My recovery is phrase-specific, but increasing each day. I have recovered about 70% of the traffic that I lost to Panda. Also, my site is small (under 150 pages). From what I have seen, most who have seen recovery also report that their sites are small (correct me if I am wrong on that point). At the rate I am seeing recovery, one internal page at a time, and a few phrases at a time, it is my guess that larger sites will indeed take much longer.

Given the enormous amount of work that I have put into site changes over the past 8 weeks, I can only imagine what a daunting task it would be for very large sites (especially those with thousands of pages). For a summary of changes that I made, read each of my posts throughout this thread; everything that I theorized, I did (the only thing I did not do was move to an image-heavy template...I kept my tables, cleaned up the html, and improved rendering). Until I see the traffic back to 100% and after I know the rebound is permanent, it's probably too soon to speculate any further. I've done what I can.

[elguiri]

BTW, I think there are two different dynamics in this thread that probably need to be separated:

1. Panda I in the US (Feb), and Panda II in English language sites outside the US.

2. Panda II in the US.

In addition, this change also goes deeper into the “long tail” of low-quality websites to return higher-quality results where the algorithm might not have been able to make an assessment before. The impact of these new signals is smaller in scope than the original change: about 2% of U.S. queries are affected by a reasonable amount, compared with almost 12% of U.S. queries for the original change.

The important bit here is "new signals". i.e it was something else. Given that there sites in the same sectors, and targetting the same terms, that were affected by one, and not the other, reiterates that.

While the thinking behind Panda I and Panda II was pretty much the same, the medicine, I believe, is rather different.

In terms of the additional 2% of search terms, I think that would affect some sites with the Panda I treatment in sectors not previously touched. i.e. if your sector was untouched by 1, but you got hit by 2, you'll need to understand the changes that went on in both.

Is this making sense?

[indyank]

Sites like hubpages will never move up as long as panda is active.The kind of content on their site is total crap. And a huge percentage of it is scraped. But if they move up leaving others behind, only god can save google.

[crobb305]

Sites like hubpages will never move up as long as panda is active.The kind of content on their site is total crap. And a huge percentage of it is scraped. But if they move up leaving others behind, only god can save google.

The mere size of those sites makes recovery even more problematic. I think the probability of recovery may increase with decreasing site size. Smaller sites are easier to clean up, and will probably represent a larger fraction of depandilization...assuming they are/become valuable (e.g., not a "farm"), and many may have been borderline from the beginning.

[edited by: crobb305 at 5:57 pm (utc) on Apr 28, 2011]