Change To Link Bomb Sign Of New Link Analysis Shift?

[dannysullivan]

If you aren't aware, Daniel Brandt of Google Watch fame started a link bomb to make the page about Google's management come up tops by linking to it with the words "out-of-touch executives." It's ranked tops until apparently this week.

Brandt emailed me today claiming that Google must have deliberately changed things like this:

If your search term includes the adjacent two words "touch" and "executives," regardless of which order they are in, then the links from my pages with the "out-of-touch executives" are discounted and do not provide any "juice" for the results you see. However, each of these words by itself provides juice, as long as the two words are not adjacent.

I found two exceptions to this. For example, executives out of touch does NOT have the terms adjacent, yet the page is still buried. That runs counter to Brandt's theory.

Meanwhile, google executives touch does have the words adjacent, yet the page comes up number one. Again, counter to the claim.

This all comes at a time when people are reporting a lot of backlink changes at Google. I also noticed yesterday that the Kerry campaign web site is no longer tops for "waffles" on Google, as he has been courtesy of another link bomb campaign.

That makes me think that the change in Brandt's case didn't happen specifically because Google itself didn't like the publicity but instead because Google's trying some new link analysis changes in general. In fact, it seems like they might be leaning toward doing something that Brandt's wanted, to not let link text count (or as much) unless the words are actually on the page.

Of course, the miserable failure link bomb still comes up tops with the Bush bio -- and those words don't appear on that page. That might make you think Google is indeed fixing thing selectively, given that major Democrats like Gore have visited Google in the past. But Michael Moore is hanging in there in the top results as well, and he doesn't use those terms on his page.

It think it's most interesting to focus on the word "executives" in Brandt's case. Consider the fact that out of touch management has also brought that Google page to the top (and still does) courtesy of Brandt's campaign. In fact, the New York Times embarrassingly reportedthis was something people upset and inside Google had caused, something they later corrected.

So, why does out of touch management still work but out of touch executives doesn't? One complication is that Brandt in the past few days has changed his links to "out of touch management."

Another reason may be that the word "executives" doesn't appear on the Google page in question -- while the word management does, and quite prominently.

Consider also that other searches for prominent words on the page combined with "out of touch" still bring the page up:

out of touch larry (1.2 million matches)*
out of touch sergey (14,200)*
out of touch brin (11,800)
out of touch google (605,000)*
out of touch cindy (619,000)

While these don't:

out of touch president (1.7 million)
out of touch products (5.2 million)
out of touch page (6.5 million)
out of touch technology (2.7 million)
out of touch employees (958,000)*
out of touch engineering (822,000)

It may be that a new system is working where the more popular a word is (in terms of appearing on pages across the web), the more required it is to appear on a particular page for link text to also work with it. In other words, "out of touch sergey" would work because there aren't that many pages with all those words on them, so the link text is allowed more influence. But "out of touch page," there are lots of pages with those words on them, so link text analysis might be suppressed more.

I doubt it's this simple -- nothing ever is. But I've talked with Google before about the entire link bombing issue, and they have said they're looking at new ways to avoid some of the problems that can happen. This might be the start of that. Or maybe Daniel's right, and they just don't like that management page coming up

By the way, those searches I noted with * above? Those all bring up this thread from WebmasterWorld where this issue has also been brought up in the past. I thought it was interesting that this page shows up at times with the Google management page doesn't. Key reason that I can see? When it shows up, it has all the words in the query.

I asked Daniel to drop by to share his own thoughts -- and more speculation or observations are welcomed.

[rustybrick]

Very interesting indeed.

I did some of my own tests on the keyword computers.

Dell ranks #1. The keyword "computer" is not found anywhere within the on page copy, only within the meta information. Look at the Google cache for Dell on that search, "These terms only appear in links pointing to this page: computers." The allinanchor:computers brings up dell as #1 as well.

So, if I understand what you are saying, Dell should either (1) not rank for computers, or (2) Google looks at the meta information when determining rankings.

[dannysullivan]

So, if I understand what you are saying, Dell should either (1) not rank for computers, or (2) Google looks at the meta information when determining rankings.

In this case, Google's definitely indexing the text in the meta description tag. Go back to the search on computers, and you'll see the word being highlighted.

That skews things. To explain, let's assume there was absolutely no use of the word "computers" on the Dell page and it was still ranking well, purely on the basis of link text. That has been the case for some pages in the past on Google. It's also why link bombs have worked.

My thought is that Google might have made a change so that for high frequency words, such as computers, link text might only be allowed to be credited to a page if the page itself also made use of that word.

So in this situation, if Dell didn't have the word computers on it, it might not rank well despite all the links point at it.

I don't think the case could be that simple. Google might also be trying to group links together, discount them in various ways, look at the frequency of words and so on. Nothing will ever be so simple. But it may be that they are introducing some new ways to counter link bombing.

Here's a different example: failure. The Bush page still ranks tops yet it doesn't have that word on the page at all, according to the cache. And failure on its own is popular -- Google finding over 22 million matches for it. Not all of those matches are pages with the word on them -- but plenty are.

That's completely counter to what I've suggested. Here's a popular word and the link text is still being entirely credited for it for this page despite the fact that the page doesn't use it at all.

So, more may be going on. Here's something else to consider. I get plenty of matches for bush failure or moore failure (moore ranks tops for just failure, as well) -- both over a million. Perhaps part of the link analysis system is to look for the cooccurance of certain words.

In other words, Google sees lots of links saying "miserable failure" point at the Bush home page. Should I trust this link text for either one of the words or both of them? Let me look at ordinary text on pages across the web. Do a see many pages where some of the words on the Bush bio page (like the word "Bush" itself) also appear near the word "failure." Yep, over 1 million of them. OK, I'll trust the link text for this.

How about Kerry? Well, look for kerry waffels, and you get only 19,700 matches. Far fewer pages out there apparently have the words kerry and waffles in close proximity. So, you might trust less the link text pointing at the Kerry campaign site and saying waffles.

SPECULATION WARNING! The above is purely speculation, may be completely wrong and almost certainly would overlook many issues with how links are actually being used. The point is, there are a variety of things you could try to reduce the impact of link bombing but still maintain the advanges of link analysis for finding popular sites on popular topics.

[Everyman]

Meanwhile, google executives touch does have the words adjacent, yet the page comes up number one. Again, counter to the claim.

I see this too. But to confuse things further, google touch executives buries it again (leave in "google" but switch the order of the other two).

Here's another strange one: touch google executives is at number 1, but executives google touch is buried.

Okay, so the adjacency isn't it, and the order does matter. Two strikes against me. But I still insist that it was designed to zap my bomb. The reason "management" works is for the same reason that "miserable president" works. The word "miserable" alone is sufficient to bring up Bush, and the word "failure" alone is sufficient to bring up Bush. Now all you have to do is find a strong word on Bush's page to go with one or the other.

That's the same reason "management" continues to work. Yes, the fact that I changed my links 48 hours ago should strengthen "out of touch management" considerably. But it's been number one anyway ever since the New York Times wrote about it on June 22. I researched the issue then, and only one or two bloggers had accidently stumbled onto the "out of touch management" one week before the NYT. That's not enough juice. The juice for "management" came from Google's title and headline, and the juice for "touch" came from "out of touch executives."

The way I would phrase it is that while anchor text in links has, for two or more years now, given juice to the target page, suddenly we also have a situation where "negative juice" is also possible. When Google is inclined to rank a page based on external anchor text, perhaps it now has the option of consulting a list of "anti-keywords" found in anchor text for that page. And while I'm forced to agree that it's not as simple as I thought before Danny found his exceptions, I still think Google went after my bomb.

They probably did it before June 22 and weren't even aware that "management" would turn out to be an issue. I wasn't even aware of this until I saw the NYT piece. It's possible that the lists of anchor text and anti-anchor text associated with a page are set fairly early in the update cycle, and it was just too late to throw "management" into the list for Google's page at www.google.com/corporate/execs.html.

I'm wrong in thinking that this is a simple case of adjacency. But the words "touch" and "executives" are so hyper-sensitive for the ranking on Google's page (you're either at number one or you probably aren't even in the top 1000), that I think my claim of being the first Google bomber to get bombed by Google still stands up.

[seobook]

while it may not be related...

there have been a decent number of people reporting home pages not showing any pagerank while inner pages do

they have been showing a different backlink profile for most sites than what they usually showed in the past

some people have recently been complaining about their sites being dropped for something (this happens all the time so it might mean nothing).

i think showing the different link profile might be done to confuse us while bigger changes are occuring. hopefully we are in for a really fun next couple weeks

[Everyman]

I am reconsidering the possibility that the behavior with respect to "out of touch executives" is related to an extra algorithm layer in Google that was designed to help control spam. I considered this two days ago, but rejected it because there was no corroborating evidence from other webmasters observing their own sites. Since then I've seen a report on another forum from someone who got nailed for a single keyword that he started emphasizing in anchor text just two months ago. In every other respect his rankings are normal, including keywords he's used for months, but he's suddenly having a lot of trouble with this one keyword. It drops his page out of Google almost entirely.

Bear with me, as I take you from the obvious to the speculative.

1. It's obvious that Google maintains a list of keywords for many pages. For the most part, this list finds its way into the inverted index, along with the position of each word on the page. Let's call this the on-page word list.

2. It's also obvious that Google somehow adds to this list from inbound anchor text, regardless of whether the anchor text keywords are on the page. Let's call this the anchor-text list.

3. The third obvious point is that search terms related to either or both the on-page list and anchor-text list cause the page to be considered further for ranking on those words, word pairs, or phrases.

4. Here's where it gets speculative. Let's assume that Google installed an additional list, called the anti-anchor-text list. Only a small fraction of web pages have this list, so it doesn't really slow things down. The algorithm makes a pass at the SERPs page, before displaying it, to see if any of the links about to be displayed to the searcher are flagged as having this anti-anchor-text list.

5. If any of the links are flagged, then the anti-anchor-text list is pulled up for that link. Additional rules are applied to determine whether that link is maintained in the display, or rejected. The rules essentially represent some level of interaction between four things: a) the search terms entered by the user, b) the on-page word list, c) the anchor-text list, and d) the anti-anchor-text list. The rule set could be optimized so that you end up with conditions that appear almost random to us. For example, if you apply the rule set to each word in the search term separately, you then arrive at a yes/no decision for each. If you accept it based on the first word, you go on to look at the second word. This could explain what we're seeing with my Google bomb.

6. Now let's assume that Google developed a way to populate the anti-anchor-text list automatically, by using some sort of link analysis designed to snare spam. This would almost certainly be an off-line crunching sort of thing, like the old-time PageRank calculation that took days to do. Maybe my Google bomb got caught in such an algorithm?

The difference between what's happening now and what happened last November, is that this time the crunching is precomputed instead of being done on the fly. This is not a real-time filter (Scroogle is no help at all), but a new level of analysis for anchor text. The flag for whether the anti-anchor-text list exists is checked on the fly, but the anti-anchor-text list itself is precomputed. Essentially, it's a done deal. They could turn off the flag detection, but that would make it all-or-nothing. There's no knob they can turn back slightly.

How does this relate to the fact that so many more backlinks are showing now? Perhaps Google wants to place a layer of fog over this new algo. Perhaps Google feels that they have sufficient potential control now over backlink spam that they can afford to show more information. Perhaps it's unrelated.

How do we confirm this? Webmasters who see a page drop out for their keywords, and who have optimized for these words in inbound anchor text, should play with the placement of each keyword in the search box and see if this changes the results dramatically for the ranking of their page.

[orion]

Since we are dealing with terms co-occurrence and term combinations, a co-occurrence and sequencing analysis may shed some light, at least partially.


Results are only valid for the Google database and may chance in time.

QUERY CONDITIONS

TARGET: GOOGLE
DATE/TIME: 07-23-2004 AT 11:00 AM
CASE: INSENSITIVE
MODE: FINDALL (for co-occurrence analysis, only)

Using the semantic tools described in the http://forums.searchenginewatch.com/showthread.php?t=48 thread we obtained the following (see http://www.miislita.com/semantics/c-index-8.html)

CASE I

Query 1
k1=touch n1=33,700,000
k2=executives n2=6,300,000
k12=touch executives n12=369,000
c-index = 9.31 ppt
touch executives, ef = (292/369,000)*100 = 0.0791%

Query 2
k1=executives n1=6,300,000
k2=touch n2=33,700,000
k12=executives touch n12=370,000
c-index = 9.34 ppt
EF ratio = (17/370,000)*100 = 0.0046%

Query 1 vs. Query 2: CO-OCCURRENCE AND SEQUENCING ANALYSIS
1. Queries show similar degree of co-occurrence (c-indices: 9.31 ppt vs 9.34 ppt).
2. Query 1 shows more degree of sequencing (EF ratios: 0.00791% vs 0.0046%)
3. Query 1 shows more co-occurrence results in EXACT mode (292 vs 17).
4. Query 1 shows more results related to the bomb in the top 10 positions.
5. Query 1 shows documents about the bomb in EXACT and FINDALL modes in the top 10 positions.
6. Query 2 does not show documents about the bomb in EXACT mode in the top 10 positions.
7. Query 2 show documents about the bomb in FINDALL mode in the top 10 positions.


As a "bomb", Query 1 appears to be more "loaded" than Query 2. In Query 1, some documents retrieved in EXACT mode find their way to the top positions in FIND ALL mode.


CASE II

Query 3
k1=out-of-touch n1=424,000
k2=executives n2=6,300,000
k12=out-of-touch executives n12=17,600
c-index = 2.62 ppt
EF ratio = (279/17,600)*100 = 1.5852%

Query 4
k1=executives n1=6,300,000
k2=out-of-touch n2=424,000
k12=executives out-of-touch n12=17,600
c-index = 2.62 ppt
EF ratio = 36/17,600*100 = 0.2045%

Query 3 vs. Query 4: CO-OCCURRENCE AND SEQUENCING ANALYSIS
1. Queries show identical degree of co-occurrence (c-indices: 2.61 ppt).
2. Query 3 shows more degree of sequencing (EF ratios: 1.5852% vs 0.2045%)
3. Query 3 shows more co-occurrence results in EXACT mode (279 vs 36).
4. Query 3 shows more results related to the bomb in the top 10 positions.
5. Query 3 shows documents about the bomb in EXACT and FINDALL modes in the top 10 positions.
6. Query 4 does not show documents about the bomb in EXACT mode in the top 10 positions.
7. Query 4 show documents about the bomb in FINDALL mode in the top 10 positions.

As a "bomb", Query 3 appears to be more "loaded" than Query 4. In Query 3, some documents retrieved in EXACT mode find their way to the top positions in FIND ALL mode.


CASE III

Query 5
k1=out of touch n1=11,900,000
k2=executives n2=6,300,000
k12=out of touch executives n12=312,000
c-index = 17.44 ppt
EF ratio = (279/312,000)*100 = 0.0894%

Query 6
k1=executives n1=6,300,000
k2=out of touch n2=11,900,000
k12=executives out of touch n12=314,000
c-index = 17.56
EF ratio = (36/314,000)*100 = 0.0115%

Query 5 vs. Query 6: CO-OCCURRENCE AND SEQUENCING ANALYSIS
1. Queries show similar degree of co-occurrence (c-indices: 17.44 ppt vs 17.56 ppt).
2. Query 5 shows more degree of sequencing (EF ratios: 0.0894% vs 0.0115%)
3. Query 5 shows more co-occurrence results in EXACT mode (279 vs 36).
4. Query 5 shows more results related to the bomb in the top 10 positions.
5. Query 5 shows documents about the bomb in EXACT and FINDALL modes in the top 10 positions.
6. Query 6 does not show documents about the bomb in EXACT mode in the top 10 positions.
7. Query 6 show documents about the bomb in FINDALL mode in the top 10 positions.


As a "bomb", Query 5 appears to be more "loaded" than Query 6. In Query 5, some documents retrieved in EXACT mode find their way to the top positions in FIND ALL mode.


CASE II AND CASE III ANALYSIS

1. out-of-touch returns 424,000 results.
2. out of touch returns 11,900,000 results.
3. out-of-touch executives returns 17,600 results.
4. out of touch executives returns 312,000 results.
5. CASE III shows more degree of co-occurrence than CASE II (about 17 ppt vs about 2.6 ppt)
6. CASE II shows more degree of ordering (sequencing) than CASE III (by more than 1 order of magnitude)

It is clear that the above use of hyphens introduce a degree of selectivity in the queries, affecting the end results (less documents retrieved). These results also suggest that CASE II and CASE III are different scenarios. In a more general sense, one should expect that the use of delimiters could affect query results, especially queries formulated in EXACT mode. The situation is more complex that it looks, especially when too lose or generic terms are used ("out","of", etc..).

EF ratios measure the degree of ordering present in searches conducted in FINDALL mode. In this particular experiment, the EF ratios of Queries 1 - 6 are too small to claim that search engine positioning was the result of ordering. Furthermore, note that in FINDALL mode Query 2, Query 4, and Query 6 produce top 10 results about the bomb, but in EXACT mode no top 10 results about the bomb are obtained. This confirms that the sequences utilized do not play a significative role.

Why then Query 2, 4, and 6 produce results about the bomb in FINDALL mode? It is possible that the returned documents may (or may not) be using other type of sequences, optimization strategies or linking techniques, including but not limited to "link bombs".


Orion

[jseamless]

If google is being bombed for words which do not even exist on page that must cease. I really hope they are implementing this one.

[Everyman]

If google is being bombed for words which do not even exist on page that must cease. I really hope they are implementing this one.

I agree, but I'm not hopeful. Why start on my puny little bomb? The miserable failure bomb goes to the White House, and John Kerry is even keying an AdWord off of it. (Oops, I answered my own question. Google gets money from it and besides, Eric Schmidt supports Kerry for President.)

I don't understand your point. Orion. Surely the fact that we're observing one particular page (www.google.com/corporate/execs.html) appear variously at number 1 to number 3, or not appear hardly at all (number 380 to 1000+), depending on how, where, and when the term "executives" is entered into the search box, suggests that this situation is rather binary in nature. It's very rare to get anything between 3 and 380.

I now think that "executives" is the word that Google went after, and "touch" may not have been targeted. If "executives" is preceded by "google" in the search box then they let it through on its merits, otherwise they zap their /corporate/execs.html page from the results.

For me the binary question is, "Did Google aim a shotgun at spam and my Google bomb got hit by some stray buckshot, or did Google go after "executives" with a sniper's rifle?" I still don't see enough reports of missing pages from other webmasters to convince me that this was a general attack on spam. It's been almost a week, and you'd think that some webmasters would be screaming by now.

I still claim to be the first bomber to get bombed by Google. Danny is eager to find contrary examples (he doth protest too much?), but I still think the weight of the evidence is with me.

[Marcia]

The thought that came up when first reading the article that prompted the original forum post mentioned was that there's a potential for difference in effectiveness between a bomb for a a phrase that isn't competitive and isn't being targeted at all and one for which at least some people are making an effort to rank, which is useful to see for people who optimize sites.

dannysullivan

By the way, those searches I noted with * above? Those all bring up this thread from WebmasterWorld where this issue has also been brought up in the past. I thought it was interesting that this page shows up at times with the Google management page doesn't. Key reason that I can see? When it shows up, it has all the words in the query.

While it was posted because of reading the article, the first post was deliberately "conservatively" optimized with just a few simple basics to see how it would fare against the bombed page, as stated in message # 7

How much does it actually take to Googlebomb for an obscure phrase, if there's even just a tiny bit of "optimization" and the natural use of language, just in the normal course of conversation?

How many links would it take to move Google out of the #1 position there? My guess is that one, maybe two or three links would do it and they wouldn't all necessarily have to be the exact phrase.

There was very little done in the first post, as also indicated in message #7,

OK, now let's see how hard this is to do.

You'll please notice the title of this thread and the fact that the exact phrase was used in outbound anchor text - and deliberately put in bold font for my own personal amusement.

The rest would be determined by random occurrences in the rest of the thread, as in any normal course of natural conversation, with random density.

As we can see in the Google SERP, where that thread now sits at #1 & #2 indented, the original title read, which is in bold except for the stop word, was

Google: Out of Touch Management

which if we look at the current thread we can see was edited to read

Google Management Article

We can also see that there were originally 16 messages, which more than likely gave a good random number of occurrences and density, and spanning two pages explains the indented result - which in people's efforts to get indented results can tell us something about the use keywords in page titles.

This 16 message thread spans 2 pages: ( [1] 2 ) > >. Google: Out of Touch
Management FUD from the Pittsburgh Post-Gazette. ... out of touch management. ...

There are now only 10 messages left in the thread, so with the change in number of posts, and consequently in the of number of words, keyword occurrences and density, combined with the altered title not including the exact phrase, it will be interesting to see how it fares in near future compared with the original Google page in question, which is now at #5.

What's also interesting to note, regardless of the fact that Google's link: command is now worth less than nothing, is that there is only one rather strange backlink showing up for that thread

Link to Google Management Thread

It's a given that scoring is based on a combination of off-page and on-page factors, but for all practical purpose what remains as a constant for effort and study in site construction is the many factors that go into on-page and site-wide optimization considerations, since in the long run those are generally more easily controllable by the webmaster.

Everyman:

I still don't see enough reports of missing pages from other webmasters to convince me that this was a general attack on spam. It's been almost a week, and you'd think that some webmasters would be screaming by now.

Daniel, people are screaming about missing or PR0 pages, but not necessarily those who have spam issues. Unfortunately, when the net is put out it manages to catch a lot of squid along with the yellowtail.

[dannysullivan]

Why start on my puny little bomb?

Actually, you wouldn't be the first targeted even if you were indeed targeted.

After I did the write up on the miserable failure search, I did a follow-up interview with Google on the issue of how they deal with link bombs and the impact on relevancy. This was a couple of months ago and for a piece on relevancy I'm planning for the future.

One thing that was mentioned is that some time ago, Google blocked certain profane words. If I understand it right, you couldn't point at a page with some profane text to force that page to appear.

That's key reason that the dumb motherf--- search doesn't bring up the Bush campaign store as was the case ages ago. So if anything was targeted specifically, ironically, it was a link bomb that hurt Bush.

Danny is eager to find contrary examples (he doth protest too much?)

No, I'm not. You emailed me that your link bomb was specifically targeted by Google and offered an explanation as proof. So I checked it out, because it's a serious allegation -- and it was unusual for that listing to have changed. It was rather easy to find exceptions to what you said. I didn't have to be eager to find it.

Also this same week -- and before you contacted me -- I also noticed that the Kerry waffles search had changed. It could indeed be that you and the Kerry link bomb were both targeted because Google doesn't like you dissing their management and they all love Kerry -- and I said that at the very beginning:

Or maybe Daniel's right, and they just don't like that management page coming up

OK, adding the smiley face perhaps took away from the statement. But I didn't mean for it to come across like I wasn't taking you serious. It could very well be that behind the scenes, they're doing something to influence these specific link bombs. I don't know.

It could also be something else -- and the suggestions of this are also worth exploring. That's what I do, Daniel -- try to look around at all the explanations I can find and lay them out for people, so people can make up their own minds. And that's why I'm glad you're also posted any further findings you come up with. The more information, the better.

[Everyman]

Okay, Danny, I apologize for "doth protest too much." You've been fairly straightforward and objective with me for a year now, and less inclined to accept Google's spin uncritically, and I should give you the benefit of the doubt.

I agree that the implications of Google playing games with the algo, which is what we're basically suggesting by entertaining even the possibility of zapping my bomb, demoting the waffles bomb, and letting the miserable failure bomb fly high with Kerry's AdWord next to it, are quite serious. If fact, if it were possible to prove this with a smoking-gun internal memo or something, it would be a story. A couple of heads would have to roll at Google, assuming it's just some algo engineers doing it for fun. I've already told one reporter that I don't have enough for a story.

By the way, congrats on noticing the waffles thing. That is suspicious, falling to number 16 like that. But if my bomb was demoted to 16 from 1, instead of falling to somewhere between 380 and 1000+, I wouldn't have raised the issue.

Yes, Marcia, webmasters are always complaining about pages dropping out. And yes, bombs that have little competition for the keywords are many orders of magnitude easier to do. But the reason I'm suspicious is that my bomb was going along perfectly since late March, when it rose to number one, up through the last time I checked it, which was -- to the best of my memory -- about two weeks ago.

When I say it was number one from late March, that means over three solid months of number one, without a hint of anything even coming close to challenging its position. Then all of a sudden, it's crushed.

Perhaps it got caught in an anti-spam effort. But it must be something new that Google is doing with spam, otherwise the bomb wouldn't have been doing so well for three months. And why doesn't "miserable failure" get caught in the same anti-spam effort? I don't see anything really different being reported by webmasters over the last few weeks (yes, there's the backlink upheaval, but I mean pages disappearing for keyword searches). If what my bomb experienced was happening across the entire Google index, you'd have threads at WmW like they had last November. I just don't see it, and I've looked on most of the forums for evidence. I've found two or three posts that might apply, but that's not much.

[orion]

OK, folks, back to the discussion of results...

Everyman (Daniel), the above results I provided were given in a general sense and are valid for the tested cases only. I haven't check combinations with the "google" or "management" terms (not yet). Including the term may (or may not) change the picture.

On other matters,...

The default search mode in Google is FINDALL (also known as AND) anywhere in the document and without regard for sequence or proximity. Thus, including all terms in a query doesn't mean that the retrieved pages are relevant to the sequence entered in the query box.

On the other hand, contrary to popular opinion, queries in EXACT mode are not searches for phrases. (See Keywords Co-Occurrence thread). These are searches with regard for sequecing and proximity. How proximity is defined varies between search engines and depends on many things, among others on the library of stopwords and delimiters to be ignored by the target system.

If one suspect that an exact sequence plays a role, then one must determine the fraction of documents containing the sequence when one search in the default mode (FINDALL). This is done with EF ratios.

Finally, for the combinations without hyphens and marked with an astherisc by Danny (original post of this thread), these queries should return the www.webmasterworld.com/forum86/305-2-10.htm page or related pages in the top 10 positions. This is what we found.

QUERY CONDITIONS

TARGET: GOOGLE
DATE/TIME: 07-23-2004 AT 9:30 PM
CASE: INSENSITIVE

EF RESULTS

1. out of touch larry *, EF ratio = (9/993,000)*100 = 0.0009%
2. out of touch sergey *, EF ratio = (6/14,400)*100 = 0.0417%
3. out of touch google *, EF ratio = (28/630,000)*100 = 0.0044%
4. out of touch employees *, EF ratio = (11/966,000)*100 = 0.0001%

As expected, Results 1, 2, and 3 in FINDALL return the www.webmasterworld.com/forum86/305-2-10.htm #1 or close.

Result 4 in FINDALL shows the www.webmasterworld.com/forum86/305.htm page in position #6 with the ChannelMinds.com site occupying positions #1 and #2.

How this could be possible considering the few results obtained in EXACT mode? Let see.

Result 1 in EXACT mode produced 9 results. Using "repeat the search with the omitted results included" reveals many secondary results coming from webmasterworld posts.

Result 2 in EXACT mode produced 6 results. Using "repeat the search with the omitted results included" reveals many secondary results coming from webmasterworld posts.

Result 3 in EXACT produced 28 results but displays only 3. Using "repeat the search with the omitted results included" reveal many secondary results coming from blog posts, especially the TechnologyReview.com site.

Result 4 in EXACT produced 11 results but displyas only 7. The webmaster.com page in question does not appear in the top 10. In this mode, the ChannelMinds.com site is #1. Using "repeat the search with the omitted results included" reveals many secondary results coming from the channelminds.com site and the rest coming from dissimilar sites.

Finally,...

In Results 1 and 2, the weight assigned to the page ranked #1 in FINDALL mode appears to be supported by "the suppressed" secondary documents found through the "repeat the search with the omitted results included" and containing the exact sequence. In Result 3 TechnologyReview.com site finds its way to the top 10.

It appears that not mere word sequencing is playing a role here. However, it does appears to play a role when links/documents are coming from the same domain pointing to a page from the same domain. This may explain why ChannelMinds.com site is #1 for out of touch employees and why the TechnologyReview.com site finds its way to the top 10 (Result 3).

These results may change over time.

Orion

[orion]

Not sure if this add some real-life perspective to the "out of touch management" thing.

According to this news
http://business.bostonherald.com/tec...rticleid=36942

Dr. Brian Reid is suing Google. In his complaint, Dr Reid, 54, and quote "alleges that the Mountain View, Calif., company fired him as its director of operations last February because he didn't fit in a culture emphasizing ``youth and energy.'' He also claims he was discriminated against because he's a diabetic."

The article mentions that "Page made the final decision to fire Reid, the lawsuit states. Reid said Shona Brown, vice president of business operations, told him he was incompatible with Google's youthful atmosphere. After he left, Reid said he learned he was replaced by someone in their 30s."

Gary has noticed this news in this SEW thread

http://forums.searchenginewatch.com/...read.php?t=742

along with a bio of Dr Reid (http://justus.anglican.org/reid.html)


Orion

However, it does appears to play a role when links/documents are coming from the same domain pointing to a page from the same domain.

Might be interesting to take some of these pages that have dropped like a rock and look just at internal links for a moment (for my purposes I am going to define internal links as links originating from the same class C IP).

I wonder if the pages were penalized because of what percentage of the internal links have duplicate anchor text...

[Marcia]

>>penalized because of what percentage of the internal links have duplicate anchor text...

lots0, I've seen it happen - a problem with a site exacerbated by an excessive percentage of identical anchor text within the site itself.

[Everyman]

At the peak I used ten links on eight different domains. One of the eight was a dot-com, the other seven were dot-orgs. Each domain had a static IP. Four of them were from one Class C, and four from a different Class C. The anchor text was identical: "out-of-touch executives". Seven of the link pages were PageRank 5 or 6; the other three were lower.

I'm not exactly in the same league as some spammers out there!

And why did it happen suddenly after more than three months? People have been talking about identical anchor text and same Class C for many months now; it's nothing new.

[seobook]

Originally Posted by Marcia

>>penalized because of what percentage of the internal links have duplicate anchor text...

lots0, I've seen it happen - a problem with a site exacerbated by an excessive percentage of identical anchor text within the site itself.

many sites use the same phrase link using breadcrumb navigation. did the sites that got penalized usually not have many external links pointing into them? what happened when they were penalized? was it just a term penalty or like an out of the index penalty?

[orion]

I must agree that there is "nothing new under the sun". The method of appending false relevancy to documents with keywords is well documented in this blog

"Attaching Keywords to Any Site"
http://blog.outer-court.com/archive/...85851918769310
The original bomb was using http://www.cnn.com/?-gmail-account

Garrett French also discusses and tried here

http://www.webpronews.com/insiderrep...thodFound.html

French tried with the following delimiters

http://www.google.com/?-teoma-rules
http://www.google.com/#-Teoma-Rules

The technique involves manipulation of delimiters in links and queries and also appears to work with Yahoo.

Is interesting to point out that querying Google with certain delimiters and Daniel's "out-of-touch management" expression brings up Google's www.google.com/corporate/execs.html page. Some delimiters brings nothing and other bring surprising results and as follow (query mode is FINDALL)

1. The following returns nothing

out-of-touch_management
out-of-touch&management

2. The following returns Google's page with many pages in the top 20 positions being quite critical about George Bush's administration. I count this one as a collateral bomb to the Bush administration, since few documents mention Kerry, too.

out-of-touch|management (multiple pipes -eg. ||, |||, |||- produce same results)

This particular query does not return the webmasterworld pages or blogger pages in the top 10 positions.

3. The following returns Google's page and webmasterworld pages in the top results

out-of-touch^management
out-of-touch?management
out-of-touch#management
out-of-touch;management
out-of-touch:management
out-of-touch::management (multiple colons produce same results)

4. The following returns the webmasterworld pages and other blog pages

out-of-touch-management
out-of-touch*management
out-of-touch & management

That bloggers are part of the mix is evident. The following urls appear in top positions in many of the above queries
http://members.cox.net/katheesue/200...le-search.html
www.rugles.com/weblog/archives/000177.html
and many more

So, it appears that SE Bombs are here to stay. Use the above strategy of appending keywords to links pointing to an external site or page, mix well with carefully pre-tested delimiters, link secondary pages with similar tricks, let bloggers link to the primary and secondary pages, and voila. Soon others will be jumping in the vanwaggon to the link scheme. Really, not many initial seed links are required to bring a page all the way up to the top and an algorithm all the way down.

Orion

[I, Brian]

In reply to the first post - the simplest explanation is just that Google devalues links to a page where the majority of the links are same text. This sort of pattern is indicative of link manipulation - link bombing

It's been reported for quite some time now that a filter may be in place to devalue links overwhelmingly sharing the same link text (anchor text).

We've been advising our own link-building clients for some time to use 3-5 different variants in link text. We've also been recommending "naturalising" the links by adding non-keyword elements, in case such a filter is expanded to apply not simply to eqact text matching, but also to too overt repetition of matching key words across link variants.

The surprising thing is that anyone should be surprised if and when such practices may be evidenced - especially when we are possibly looking at nothing more complicated than a lowering of the devaluation threshold.