Forum Moderators: open
ABC-Widgets.com or Widgets-ABC.com
where ABC is another non keyword (such as a brandname)?
choose doamin name based on what the surfer is searching for. If that is your company name OK. If not forget the Co. name
Then by both hyphenated and the type in domain.
AND if it is important to block leakage buy the .net, .org, .us etc
Do i call it 'Direct-Widgets' or 'Widgets-Direct'
From a Google Point of view please.
Do i call it 'Direct-Widgets' or 'Widgets-Direct'
If you expect searchers only to search for 'widgets' (or 'direct' for that matter), then my experience is that you should have the most important keyword first.
If you think searchers may start searching for 'widgets direct' then it's going to be pretty useless if your domain is 'direct widgets'. Are you calling your company 'direct widgets' by any chance?
In any case, you may end up wanting to register both widgets-direct.all-tlds and direct-widgets.all-tlds to stop your competitors getting their hands on them :-)
webdoctor
Thanks for giving a straight answer too!
It is an interesting question though. Here's the answer looking at an index a couple months old using a few thousand rankings for a few hundred keywords:
[widget-abc.com...] Score: 458
[abc-widget.com...] Score: 505
I don't know where you got the information about wanting to use hyphens, but here are a couple of other unsolicited answers:
[abcwidget.com...] Score: 670
[widgetabc.com...] Score: 656
[awidget.com...] Score: 638
[abwidget.com...] Score: 685
[abcdwidget.com...] Score: 582
[abcdewidget.com...] Score: 638
If you are really going to name your domain just for the tiny amount of ranking you might get from Google as a bonus for your great naming tastes... I would go with:
[abwidget.com...] Score: 685
Can you explain this "score" thing?
Let me give a fictitious and simplified example which shows how one could discover the same for yourself (an excerpt from an article I wrote over a year ago... sticky me for the entire article):
1. Pick twenty popular search terms.
2. Enter them at the selected search engine (Google in this case).
3. Review the first twenty results for each search
and write down the answer. Write a 1 if the answer
to your question is true for that listing and a 0
if it is false.
4. If the answers are all the same (either true or
false), you may need to use more than twenty results
for your study. If so, be consistent with all
search terms.
5. If the answers for the second half of your results
for all search terms is zero, you may need to include
some more search terms until you have a non-zero
result in the second half of your results.
6. Total the answers for all search terms for the
first half of your results. Separately total the
answers for all of your search terms for the second
half of your results.
7. Divide the total from first half of your results
by the total of the second half of your results.
8. If your answer is very close to the number 1, then
you will need to expand your search to more than
twenty popular search terms... or review more than
the top 20 results... or possibly accept that the
answer to your question is that this particular
factor is not very important to this search engine
either way (positive or negative).
9. If your answer is significantly more than 1, then
the answer to your question is true.
10. If your answer is significantly less than 1,
then the answer to your question is false.
So you take those steps for dozens of questions and then you have a "score" that can be applied for those dozens of factors. Now it isn't correct to apply equal weighting to all of those factors so you must then discover the weighting. There are a variety of ways you could do this, but the following is the simplist way I can think of explaining it:
1. Apply random (or sequention between 1 and 1000 or something like that) weighting amounts to each of your factors and then score your original list of SERPs (or a sampling of them).
2. Look through the resulting scores and keep track of scores that are "out of order". The goal is to be predictive of Google's ranking. So you want the original set of data to show up in the same order when sorted by the score times the weighting factor.
When you find a set of weighting factors that produces an ordered set of scores that is almost identical to Google's own ranking order... you have validated your set of questions (factors) and your weighting factors.
If you can't come close to predicting Google's ranking, then:
1) You haven't used enough questions (factors)... or...
2) You haven't applied correct weighting factors to the questions you are researching.
So tell us jbrausch,
How close have you been able to determine Google's latest Dominic weighting factors? What were Cassandra's weighting factors and what has changed? And finally, what's it worth to you? :)
Ingenious! :)
Thanks; but it's just crunching the data...
How close have you been able to determine Google's latest Dominic weighting factors?
As close as any other update according to my stats. Dominic was a little special in that I had no idea when to take my snapshot. Things kept changing. So... my Dominic may not be the same as your Dominic. This is the first time that has been an issue. Generally there is a time that you can say that a particular Google update is over.
I'm not sure how to quantify that in a way that would make any sense. I look at area under the curve between a linear trend line drawn through a line graph of the scores of a statistically significant sample of fresh SERPs. The area under that curve (ie: between the trend line and the plotted data) is currently less than 14% of the area of the entire square represented on the top left by the beginning of the trend line and the bottom right by the end of the trend line. I wish I could draw a picture and shade in the area I'm talking about. I think it would make more sense that way.
What were Cassandra's weighting factors and what has changed?
Sure; I'll just post them right here. :) Actually weighting factors are useless unless you know all of the factors used and their weight. They are comparative, not absolute. So, I would have to publish the results of my entire research in order to answer the first part of that question. That's not going to happen.
If you want to know the second part of that question, there is a discussion going on in the supporters forum about one factor that changed significantly and I found interesting. Put up your money to support this fine forum and check out the supporters forum to find out.
And finally, what's it worth to you? :)
A lot.
I read that thread, very interesting... especially now since I understand better as to how your scores are created.
Any other "trends" you noticed? Feel free to post them wherever you want, I'll be sure to find them :)
[added] I also know exactly what you mean about the weighting factors not being absolute, since it's impossible to know if you've accounted for them all. But you can say for example, inbound link text is now worth less than h1 text or something like that, if you wanted to compare the 2 weighting factors in 2 seperate data sets (updates) correct? [/added]
I've been thinking about your question. I don't think I really can. To do that, I would need to be isolating the factors and preventing overlap. I currently make no attempt to do that. Because of that, the weighting factors (not the answers to the factors themselves) can't even be compared one to another. They only work predictively when applied in concert with every other weighting factor.
I can answer the question "Do sites having the keyword at the beginning the domain rank better or worse?" and get a definitive answer (whether Google actually uses this factor directly or not... there is an answer to that question). I can then ask the same question for the end of the domain name without the TLD (ie: without the .com) and get a definitive answer.
If those were the only two questions I asked, I could attempt to then assign weightings using the predictive outcome as the guide. However, since I know that 2 factors don't even come close to a predictive model, I never do that. I have hundreds of factors in my predictive model. The combinations of each pair of factors is overwhelming.
I make no attempt to keep those hundreds of factors from overlapping in their influence. I simply blindly assign the weights across the entire universe of factors using only the predictability of the outcoming score as a guide to assigning the factors. There is no intelligence in the part of the algorithm that assigns the weights.
Let me see if I can come up with a simple example that would show that comparing the weighting of two factors is invalid when more than two factors exist in the model. In our pretend world, let's make up a really simple Google that ranks sites alphabetically. We don't know that in advance, so we ask a bunch of questions to create our primary factors. Let's say we end up with the following questions being good enough predict the rankings of 10 sites perfectly 34% of the time:
1. Do domains beginning with an "a" rank higher or lower than sites beginning with some other letter?
2. Do sites having vowels in the domain rank higher or lower than domains having no vowels?
3. Do sites with domains ending in a consonant rank higher or lower than sites ending in a vowel?
Rememember that we were blind to how this fictional Google ranked sites before we did the study. We just randomly found out that by asking these three questions and then setting weightings would predict the rankings of 10 random sites perfectly in order 34% of the time. Some other percentage of the time, only two of those sites would be swapped. Some other percentage of the time, three of those sites would be out of order, etc.
Let's say that the weighting factors for those three questions ended up being 15, 1, 4 respectively. Is it valid to say that starting your site with an "a" is 15 times more important than having vowels in the domain name? No; it's not. "a" is a vowel. There was no isolation of those two factors. They overlap.
In addition, it is obvious that the three factors don't even come close to the actual universe of factors actually used.
Because of the overlap and the missing factors, the weightings only make sense when used along with all of the other weighting factors using in the model.
Sorry.
If you are really going to name your domain just for the tiny amount of ranking you might get from Google as a bonus for your great naming tastes... I would go with:[abwidget.com...] Score: 685
Maybe I'm just not understanding your Google algo decryption method properly :-) but are you saying you think Google can parse keywords out of a domain without punctuation?
Webmasterworld is a domain in the for <keyword><keyword><keyword> yet searching for 'masterworld' doesn't find it.
I don't think Google can do this, so I don't understand how you think abkeyword.tld could be a better domain than ab-keyword.tld
webdoctor
I am probably just too stupid for this.
[abwidgets.com...]
Furthermore, I make no claim that Google parses domains in the following format either:
[abc-widgets.com...]
To be perfectly clear, I also make no claim that Google does not parse domains in either format.
My claim is that domains in the former format actually do rank higher in the SERPs than domains in the latter format.
The cause for that is a matter that can not be addressed by pointing out a statistical coorelation.
For example, let's say I crunched megabytes of crime data and recorded the professions of people present within a one block area of murders one hour after a murder had been reported. Almost certainly one would find a very strong coorelation between murder scenes and police officers. Does that mean that police officers are guilty of murder more often than other professions?
No. Coorelation doesn't prove causality. It takes much more analysis to do that. If one looked at enough factors, one would come to the much better conclusion that police officers respond to murder scenes.
Does that make a study of statistical coorelations worthless? No. What if I am not even interested in discovering causalities? What if my interest is to show that it is possible to predictably achieve desired ranking by becoming part of the coorelation group regardless of the causality of the coorelations?
In the example of the murder scenes and police officers... even though the police didn't cause the murder... if I avoid places where there are a lot of police officers, I probably have a slightly greater chance of avoiding being the victim of murder. I don't really care that the actual cause for that is that city governments assign more police officers to high crime areas and by avoiding the high crime areas, I am avoiding the likelihood of being the victim of murder.
Does that make sense?
It is not possible to keep all other factors equal in this game.
When I remove the hyphen from the domain name, it becomes one letter shorter. It is not possible to remove a hyphen and keep the domain the same length. Therefore, one factor changed another. They are interdependent by nature.
Now let's look at "on page" factors. How can you keep all factors equal and change only one item on a web page? The two web pages must be hosted on different URLs or at different times. Once they are on different URLs, the anchor text chosen by the webmasters of sites that link to those pages might be influenced by the URL, etc, etc.
One could probably use techniques to try to discover the overlap and bias it out. I currently don't try. The set of factors I study have massive known and undetermined overlap. That's just the nature of that data.
I think this may be an attempt by google to try to even out the ranking people get because of their domain name while still taking into account the relevancy of the name.
Make sense to anyone?
Maybe its because I am testing in German language.
In German an aha in the function of an uhu with the habitat of an ihi is an ahauhuihi.
If a widget owns all this its a widgetahauhuihi.
You can write it like that, but some people find it more readerfriendly to write it widget-aha-uhu-ihi or widget-ahauhu-ihi or whatever combination they find appropriate. And they may leave the hyphen and just write separate words.
I have tested this with widgetaha.
Now if you type in German widgetaha, google asks you "do you mean widget aha", which you can click, and after that question lists everything written widgetaha first and then the other variations of writing this.
I would be very cautious about mixing a good brand name with a competitive keyword. All the results I am achieving in building the brand would be me leading people to my competitors in google.
I would be very cautious about mixing a good brand name with a competitive keyword. All the results I am achieving in building the brand would be me leading people to my competitors in google.
I really dont know what you mean by this. By having the brand name and the keyword in the title and thorughout the page, surely you will increase the chance that people will find your site by typing in the brand name OR the keyword.
I think this is the (near) perfect solution, no?
If the website is important for me I would try to make the domain name stick in peoples minds. Now do I really want the brand be connected with this one keyword only or mainly?
If yes, how unique is the brand name? I checked this out briefly - obviously people are using the strategy - and found some real spam. Nevertheless the firm using it has got a brand name which is a not too uncommon first name in Germany. Typing in brand name keyword brings some very small competitor in the keyword business to no 1.
If the brand name is very unique ... after brand name keyword all competitors with keyword will be listed, no?
well if the brand name is absolutely unique google actually doesn't show any competition with keyword. The keyword is of no use though either...
I think you fail to see the point here. If i have a well optimised site for both the brandname and the keyword of the product it sells, the naturally if someone knows the brand they should find my site in a search, and even if they dont i can use the keyword part of the name in a well optimised site to target people searching for the product.
I hope this isnt too difficult for you to understand. I realise that english is your second language so i hope that this is now clear.
