Seems counterintuitive, but his tests appear valid. I guess it's just a question of how much or little so those factors weigh into rankings. Great post, though.
What would be really interesting is if more people repeated the test and published the results. I doubt the results would be exactly the same, but seeing if there was a dominant trend over a wide range of sites would be a nice piece of data to have.
And on what kind of site? Well established site/pages?
There are so many elements going on in the algo these days, and so many changes on a regular basis, and with the possibility of time lagging of links, even outbound links ...
We did a similar test in reverse earlier this year: We removed all outbound links from a highly ranking page. Results were as you might expect (drop in rankings), but it took three months to show up, and who knows what else may have been at play... ;-)
The files are less than three months old, the site's about a year and a half. If you look at the test files you'll see the pages are dead ends and don't interface with other parts of the site. Each page also has ~200 words. I tried to exclude as many variables as possible.
I will admit I'm still scrathching my head on the "click here" results.
What does this mean from an SEO standpoint, well I wouldn't be linking out from pages that you want to rank for a particular term.
This is a poor conclusion IMO. The problem I see from the start is that the pages (on-site examples) do not all link to the same site. Further, the content of the linked to URLs does not reflect the anchor text used. Why would an SE want to rank YOUR page higher for KEYWORD just because you use it in an outbound link to a page that has nothing to do with KEYWORD?
That spells possible spam and the non-preference that is resultant could likely be a safety measure dampening applied.
On page phrase versus outbound phrase:
Pages with NO outbound links do best. This may convey that a page about keyword X that focuses on keyword X but does not link to other pages that do not focus at all on keyword X with Keyword X as anchor text (potentially misleading the user) are of more value than the others.
Outbound Link Anchor Text Phrase versus Click Here:
Inconclusive somewhat because of Control Group1's results... but this may
convey that a page about keyword X that focuses on keyword X and has an outbound link to a general authority site without the inference that that site is about Keyword X (when it clearly is not) are of more value than the others which may be conveying to the user relevance which does not exist.
Combination Test On Page versus outbound with click here or anchor text:
Same as above in respect to Google because this engine may value the fact that a link to a known (general) authority site does exist on the page; whereas the others (Y! & MSN) may not award (any/enough) ranking value based on the authority score of the linked to URL(or site) and only on the keyword usage or lack thereof.
This is a poor conclusion IMO. The problem I see from the start is that the pages (on-site examples) do not all link to the same site.
I wanted to test if linking to a page with the anchor text on the recieving page had any effect. In retrospect linking them all to fake MSN spaces would have been a better option, but I didn't get there right away, see my next reply.
Further, the content of the linked to URLs does not reflect the anchor text used.
I came to the same conclusion after the first batch of tests so I created the second batch of files. Look at file 14, the site it links to has the keyphrase on page.
Why would an SE want to rank YOUR page higher for KEYWORD just because you use it in an outbound link to a page that has nothing to do with KEYWORD?
My expectation was that file 14 was going to rank the highest of all the files in the second test, it linked out another page that shared the same keywords. But that wasn't the end resutlt.
Very cool graywolf, I notice that you have gained a new level of respect from Mr. Cutts. I find all of you guys extremely interesting. Just don't give away too many of your secrets or we will have billions of sites from india ruling the serps. ;)
A good test to try would be the effect of linking out to sites that are authoratative for the subject of the test page. I have been doing it for a long time now and although I've never tested that one factor by itself it seems to help.
A good test would be the effect of linking out to sites that are authoratative for the subject of the test page
A few people have brought to my attention that I wasn't linking to true authority websites for the search terms. The difficulty is I picked terms with no other search results to keep the test as free from external factors as possible. I'm not sure how to run a pure test in isolation, involve true authorities and still acomplish my goals. If you have an idea how I to do that let me know in the comments or PM.
I good way to do it might be to still pick terms that have no search results but create a theme for the entire page that can be used to find authority sites to link too. My guess would be that its not necessary to link to sites that are authoratative for the particular keyword but for the theme of the entire page.
For example:
Keyword: unana bulobop
Theme of page: Money and Banking
Authority Outbound Link: .Gov site dealing with money and banking
My guess is that because the outbound link would be pointing to an authority site that matches the theme of the page, that the whole page would get a bump.
'click here' isn't that counter intuitive really. After all any links page can use any anchor text but you're only going to say 'click here' if you're actually talking about the subject on page.
But it is a strange result. Needs a nice big test set I guess :)
The fundamental flaw with this experiment is that it was designed to test links in relation to hubs and authorities, yet the links do not point to known authority sites for the terms measured. The page linking could thus not possibly get a Hub bonus.
Try a genuine keyword with links to genuine authority pages so that there is the possibility of hubs and authorities within the topic.
I'd suggest a search term that gives at least 1,000 results, and not a nonsense term. After all, the experiment is to see which ranks higher, not which one reaches #1 position.
I'd suggest a search term that gives at least 1,000 results, and not a nonsense term. After all, the experiment is to see which ranks higher, not which one reaches #1 position.
As I noted in this and other experiments Google's ranking is made up a combination of factors, and I'm trying to isolate them down to their smallest parts. By playing is a space that is pristine I can eliminate or marginalize a great many of those factors.
Slap me down like a rabid monkey if im wrong about this. but if the only difference between the pages in question is that some have outbound links, then dependent on the number of outbound links, the pagerank of the page that has the links on it will be reduced. and because the links are not to authorities, meaning your site is not a hub, all the experiment shows is that by reducing your pagerank you reduce your position in the serps.
Maybe, as others have suggested, it would be wise to do this on a competative search term, and to maintain a constant to ensure that the test is fair, put links on one page to sites that are not relevant (for PR issues) and links on another to authorities/resources. And if you really want to be clever, try using relevent and irrelavent anchor text.
p.s. out of interest i have read somewhere about a guy who made a 1 page site and just put 150 outgoing links on it to ontopic resources with the correct anchor text. oh and the site had no incommiing links. yet the site achieved PR3 which dropped to PR2.
i am still sitting on the fence about the outbound link thingy but i will say this, the web is a network and all links are important in any network.
I was wondering if you used the 'nofollow' tag at all?
I think it would be an interesting test to start with pages indexed in google both without outbounds and then use both nofollow and follow links as well as no outbounds to see if it makes a difference. I also would like to see the difference if you used authority sites for your links.
Thanks for the information though, very thought provoking.
Seems counterintuitive, but
Seems counterintuitive, but his tests appear valid. I guess it's just a question of how much or little so those factors weigh into rankings. Great post, though.
Wow
That's one of the more interesting tests I've seen done in a while. Never would have expected those results. Makes you go hmmmmm
What would be really
What would be really interesting is if more people repeated the test and published the results. I doubt the results would be exactly the same, but seeing if there was a dominant trend over a wide range of sites would be a nice piece of data to have.
Ummm, I could be missing something here but...
Over what period of time?
And on what kind of site? Well established site/pages?
There are so many elements going on in the algo these days, and so many changes on a regular basis, and with the possibility of time lagging of links, even outbound links ...
We did a similar test in reverse earlier this year: We removed all outbound links from a highly ranking page. Results were as you might expect (drop in rankings), but it took three months to show up, and who knows what else may have been at play... ;-)
The files are less than
The files are less than three months old, the site's about a year and a half. If you look at the test files you'll see the pages are dead ends and don't interface with other parts of the site. Each page also has ~200 words. I tried to exclude as many variables as possible.
I will admit I'm still scrathching my head on the "click here" results.
Interesting test...
My 2c:
This is a poor conclusion IMO. The problem I see from the start is that the pages (on-site examples) do not all link to the same site. Further, the content of the linked to URLs does not reflect the anchor text used. Why would an SE want to rank YOUR page higher for KEYWORD just because you use it in an outbound link to a page that has nothing to do with KEYWORD?
That spells possible spam and the non-preference that is resultant could likely be a safety measure dampening applied.
On page phrase versus outbound phrase:
Pages with NO outbound links do best. This may convey that a page about keyword X that focuses on keyword X but does not link to other pages that do not focus at all on keyword X with Keyword X as anchor text (potentially misleading the user) are of more value than the others.
Outbound Link Anchor Text Phrase versus Click Here:
Inconclusive somewhat because of Control Group1's results... but this may
convey that a page about keyword X that focuses on keyword X and has an outbound link to a general authority site without the inference that that site is about Keyword X (when it clearly is not) are of more value than the others which may be conveying to the user relevance which does not exist.
Combination Test On Page versus outbound with click here or anchor text:
Same as above in respect to Google because this engine may value the fact that a link to a known (general) authority site does exist on the page; whereas the others (Y! & MSN) may not award (any/enough) ranking value based on the authority score of the linked to URL(or site) and only on the keyword usage or lack thereof.
Fuck, did any of that make sense? :-P
This is a poor conclusion
I wanted to test if linking to a page with the anchor text on the recieving page had any effect. In retrospect linking them all to fake MSN spaces would have been a better option, but I didn't get there right away, see my next reply.
I came to the same conclusion after the first batch of tests so I created the second batch of files. Look at file 14, the site it links to has the keyphrase on page.
My expectation was that file 14 was going to rank the highest of all the files in the second test, it linked out another page that shared the same keywords. But that wasn't the end resutlt.
gotcha
hey, I missed those later files I guess. Gotta run but i'll check 'em and respond later.
Very cool graywolf, I notice
Very cool graywolf, I notice that you have gained a new level of respect from Mr. Cutts. I find all of you guys extremely interesting. Just don't give away too many of your secrets or we will have billions of sites from india ruling the serps. ;)
Another Good Test to Try
A good test to try would be the effect of linking out to sites that are authoratative for the subject of the test page. I have been doing it for a long time now and although I've never tested that one factor by itself it seems to help.
A good test would be the
A few people have brought to my attention that I wasn't linking to true authority websites for the search terms. The difficulty is I picked terms with no other search results to keep the test as free from external factors as possible. I'm not sure how to run a pure test in isolation, involve true authorities and still acomplish my goals. If you have an idea how I to do that let me know in the comments or PM.
I've done one other so far the Google image search case study and I do have a few more up planned.
I good way to do it might be
I good way to do it might be to still pick terms that have no search results but create a theme for the entire page that can be used to find authority sites to link too. My guess would be that its not necessary to link to sites that are authoratative for the particular keyword but for the theme of the entire page.
For example:
Keyword: unana bulobop
Theme of page: Money and Banking
Authority Outbound Link: .Gov site dealing with money and banking
My guess is that because the outbound link would be pointing to an authority site that matches the theme of the page, that the whole page would get a bump.
Just one idea
I think "unana bulobop"
I think "unana bulobop" would have to be on the linked to page.
In that case what about
In that case what about optimizing some pages for a combination of unique text that is only found on a particular authority site.
For example:
Keyphrase: "HSR Introductory Guide"
Authority site: http://www.ftc.gov/bc/hsr/hsr.htm
The only site ranking for that keyphrase is the authority site.
..
The isolation is a tough one when your making the results public.
As far as linking to authorities, I would think that linking to google.com or one of the google regionals would solve that problem.
I guess
'click here' isn't that counter intuitive really. After all any links page can use any anchor text but you're only going to say 'click here' if you're actually talking about the subject on page.
But it is a strange result. Needs a nice big test set I guess :)
The fundamental flaw with
The fundamental flaw with this experiment is that it was designed to test links in relation to hubs and authorities, yet the links do not point to known authority sites for the terms measured. The page linking could thus not possibly get a Hub bonus.
Try a genuine keyword with links to genuine authority pages so that there is the possibility of hubs and authorities within the topic.
I'd suggest a search term that gives at least 1,000 results, and not a nonsense term. After all, the experiment is to see which ranks higher, not which one reaches #1 position.
I'd suggest a search term
As I noted in this and other experiments Google's ranking is made up a combination of factors, and I'm trying to isolate them down to their smallest parts. By playing is a space that is pristine I can eliminate or marginalize a great many of those factors.
Slap me down like a rabid
Slap me down like a rabid monkey if im wrong about this. but if the only difference between the pages in question is that some have outbound links, then dependent on the number of outbound links, the pagerank of the page that has the links on it will be reduced. and because the links are not to authorities, meaning your site is not a hub, all the experiment shows is that by reducing your pagerank you reduce your position in the serps.
Maybe, as others have suggested, it would be wise to do this on a competative search term, and to maintain a constant to ensure that the test is fair, put links on one page to sites that are not relevant (for PR issues) and links on another to authorities/resources. And if you really want to be clever, try using relevent and irrelavent anchor text.
p.s. out of interest i have read somewhere about a guy who made a 1 page site and just put 150 outgoing links on it to ontopic resources with the correct anchor text. oh and the site had no incommiing links. yet the site achieved PR3 which dropped to PR2.
i am still sitting on the fence about the outbound link thingy but i will say this, the web is a network and all links are important in any network.
but
you don't reduce PR by linking out :)
I was gonna quote Gray on
I was gonna quote Gray on this article he made, but looking at the date, I saw it's a little old ..
With the recent changed in the algorithm I think we need an update on those.
what about nofollow?
I was wondering if you used the 'nofollow' tag at all?
I think it would be an interesting test to start with pages indexed in google both without outbounds and then use both nofollow and follow links as well as no outbounds to see if it makes a difference. I also would like to see the difference if you used authority sites for your links.
Thanks for the information though, very thought provoking.
Veeeeery subtle link drop
Veeeeery subtle link drop there... :rolleyes:
pruned that.
pruned that.