The root of all success in search engine marketing begins with
keywords. Period. Get them wrong and virtually everything about
your online endeavor will fail. Only by targeting the right keywords
can one expect to ride that exhilarating magic carpet to online
Stating the obvious you say? ...well, if so, then why is it that
virtually everyone - professional and amateur alike - is
oblivious to the fact they are selecting, and frequently buying,
keywords based on highly skewed numbers?
The fact is that very few online marketers understand the results
supplied by the two most basic keyword selection tools. These are
the very same tools being used globally to hone keyword choices
into supposedly laser sharp focus in an effort to keep pace with
the challenges of increasingly keen competition and ever-rising
keyword pay-per-click costs.
The critical differences — Overture's STST vs. Wordtracker's KSS
As one of Wordtracker's technical support team
(http://www.wordtracker.com/moreinfo.html), one of the most
frequent questions we receive these days is...
Why are the keyword search query numbers supplied by Overture's
search term suggestion tool (STST) so incredibly different than
those supplied by Wordtracker's keyword selection service (KSS)?
Frankly, there isn't a better search engine related question one
could ask. And, now's a good time to pay close attention because
the surprising answer will likely change forever how you evaluate
First: Understanding Their Motives.
To help you understand the details we're about to reveal, let's
examine the motives of the services that are providing the
keyword query numbers.
Motive Analysis: Purpose
On the one hand, there's Overture's STST whose purpose is to help
customers buy keywords.
On the other hand, there's Wordtracker whose purpose is to help
customers select keywords.
Overture's STST suggests what keywords to buy from them.
Wordtracker suggests what keywords to use in your optimization
efforts and/or which to buy elsewhere.
Overture's success depends on you believing there are LOTS of
search queries for whatever you are selling.
Wordtracker's success depends on you getting accurate numbers
upon which you can reliably base your optimization and keyword
Overture's STST is free. Overture profits by selling you the
keywords that STST reports on.
Wordtracker's KSS is fee based. They profit by selling you access
to accurate and impartial information. Since they don't sell the
keywords, there's no vested interest in query numbers beyond
It's important to note there is no good-guy, bad-guy here - just
two companies that provide information and do so with different
incentives in mind.
Second: Understanding The Artificial Skew.
In researching the search term "keyword," Overture's STST
indicates there were 180,468 searches for the 30-day period
ending the last day of December. Of course, when we divide
this number by 30 (days), one naturally assumes that's an average
of 6,016 combined searches per day for the term keyword -
Now, if you happen to be in a business that sells keywords (like
Overture) then 6,016 pairs of eyeballs per day is a pretty
encouraging number indeed! The problem is, there isn't anywhere
even close to 6,016 per-day queries for the search term(s)
keyword(s). In fact, the actual number, which we'll share with
you in a minute, will no-doubt shock you!
But, for the moment, let's look at why that number is skewed.
Reason #1 — Artificial Searches
Overture's STST numbers are increased upward by automated
queries. These include automated bid optimizers, position and
ranking monitors, page popularity analyzers - anything other than
a real person manually performing a search is considered an
automated query. Monitoring a site's positioning at, say,
AltaVista for the search term "keyword" tallies a "hit" within
Overture's STST system for that search term. That's in spite of
the fact that it was actually automated software that generated
the hit. The same holds true for page-popularity checkers, pay-
per-click bid optimizers or any other machine generated monitor
or tabulator that queries an engine for a "pet" keyword and
generates a hit in the process.
Then, when the same positioning query is done at, say, MSN
(another Overture partner), STST records yet another hit.
Understandably, STST cannot differentiate between automated and
human queries. Neither can they tell when the auto-query has
already been queried at another partner's site.
Now, when we take into consideration all of the position
monitoring, page popularity checking and pay-per-click bid
analyzing - there are well over 15 automated and semi-automated
bid checking software programs alone - it's staggering to realize
the significant effect these automated queries are having on the
overall search term query tabulations.
However, artificial searches are only one aspect contributing to
the artificial skew (defined as: the inflation of actual search
queries for specific keywords performed by anything other than
Reason #2 - Duplicate Searches
As you most certainly must know, Overture's strength as a viable
advertising medium for online businesses lies in the fact they
are provide results to "tens of thousands of Web sites" which
include AltaVista, Yahoo, MSN Search, HotBot, and AllTheWeb just
to name a few. They claim to reach more than 80% of active U.S.
Potentially, this is great for advertisers! ...yet this very same
structure is what so greatly contributes to the artificial skew
leading to extremely over-inflated reporting of keyword queries.
According to Overture itself, statistics on searches in any
previous month are compiled from Overture's partner search
engines. To further understand how partnering tends to facilitate
skewed query counts, let's examine what happens when a visitor
conducts a search at AltaVista.
What's actually happening is that two searches are being
conducted at one time - one at AltaVista, and another that lists
the SPONSORED MATCHES supplied by Overture's pay-per-click
Although it is next to impossible to know the exact figures,
suffice it to say that a single human often generates multiple
queries when doing a single search as calculated by Overture's
STST. In some cases that same human could even generate
additional "hits" for a given keyword simply by conducting the
same search again on a different engine if such engine is also an
For instance, searching Yahoo, then searching again on MSN, then
searching again on AltaVista, then again on AllTheWeb.com would
tally at least five "hits" for the selected search term. In
comparison, if Overture (like Google, for instance) counted only
the searches that were done "on-site," such duplicate searches
would not be counted and their search query numbers would be far
This scenario, combined with the myriad artificial duplicate
searches conducted by the various softwares (explained above),
severely pumps up the number of queries for virtually every
legitimate search term imaginable.
Reason #3 - Plurals and Singulars
Remember our STST example (above) regarding the 180,468
"searches" for the term "keyword"? Well, another factor to
consider is that Overture's STST combines both the plural term
(keywords) and the singular (keyword) in compiling that number.
And, Overture's STST not only combines the plural and singular
versions of "keywords," they also combine upper and lower case
searches as well. Obviously, these two factors also exert an
upward effect on the query count tabulations.
Third: Examining The Alternatives.
So now the obvious question - Is there a "better" way to tabulate
search term query counts? ...let's examine the alternatives.
Meta-engines - a better way to accurately tabulate queries.
Obviously we'd like to eliminate artificial and duplicate
searches from our tabulations, and fortunately there is a way to
do so. The solution is Meta-engines.
Composite (Meta) engines, like Metacrawler and Dogpile, are
search engines that query all the major engines simultaneously.
One of the key differences is that the ratio of human queries to
automated queries for a meta-engine is much higher than for a
major search engine. That's because it doesn't make sense for
anyone to point their auto-bots at meta-engines.
Position monitoring, bid-optimizing, popularity checks, etc., are
typically conducted directly at the search engines themselves. It
would be pointless to conduct such automated queries on a meta-
engine because meta-engines do not "add-url's" nor do they offer
pay-per-click options. They are simply a search engine that
queries other search engines. And, since there is no
"metacrawler" of meta-engines, the search query counts are
unlikely to be artificially skewed by such artificial searches.
Furthermore, duplicate searches are eliminated because the query
counts are being tabulated from a single source instead of
combining results from myriad partners.
Therefore, query counts taken from meta-engines are far, far more
representative of the number of searches conducted by actual
people - but even this is not yet a perfect solution due to a
relatively obscure form of keyword spam.
Keyword spam (in this case not to be confused with word stuffing
or repeating keywords within a Web page) refers to the practice
of using cgi-scripting to manipulate the Metaspy metacrawler
voyeur to artificially promote certain products or services.
By entering a flow of terms or phrases at predetermined
intervals, such spammers hope to inflate the importance and
significance of certain search terms thereby artificially
increasing the value of such terms related to their products.
In a perfect world, adjustments should be made to filter out this
flavor of spam. In a minute we'll share with you how such
filtering is done but first, let's address the issue of combining
plurals with singulars and upper with lower-case searches.
Plural, singular, upper, and lower-case searches represent a
decision-point for search engine optimizers because sometimes
it's good to combine the search query numbers while other times
For instance the search terms "keyword and keywords," whether
singular, plural, or in upper or lower-case, are similar enough
in meaning that they could arguably be combined into one search
However, the search terms "tap, taps, Tap, and TAP" can have
entirely different meanings. Take a look at the results for the
search term "tap" on Overture. The following references were all
found within the top ten sponsored listings:
Machine threading taps,
Tap / Rap support software
TAP A Stock
TAP Terminal Phone Numbers
Note that none of the above has any relation to the others!
Obviously if we are selling any of these items, we'd want more
specificity regarding the search queries than the simple 10,485
searches that STST reports were conducted in the past 30 days.
The example above illustrates the importance of obtaining search
query tabulations for each version of a selected keyword
independently of the other.
After all, it's easy to manually combine the numbers while it's
impossible to break them out into their own categories once
they've been compressed by Overture's STST into a single search
term regardless of potentially different meanings.
Finally: Making sense of the numbers (here comes the shock).
Ok, now that you understand the artificial skew and the
alternatives that can correct for it, let's move on to analyze
the numbers given by Overture's STST and Wordtracker's keyword
selection service (KSS) using the search term(s) keyword(s).
An in depth look at Overture's STST numbers...
Overture's STST shows 180,468 searches were conducted. This
represents the combined count of the search terms keyword,
keywords, Keywords, KEYWORD and KEYWORDS - the combined total of
all singular, plural, capitalized, upper and lower-case searches.
When we divide Overture's count (180,468) by 30 (because
Overture's figures are for a 30-day period), the inference is
there are 6,016 searches per day that meet this criteria. In
actuality, they receive just 40-60 per day total (are we shocked
Here's how we're crunching the numbers.
Fact: Overture's STST suggests a combined average of 6,016 page
views took place between Overture and its major partners - e.g.
AltaVista, Yahoo, and others - each day for the month of December
'03. We're referring to search result pages like:
Fact: Each of these results pages lists between 10 and 40 URLs
Factor in Zipf's Law which predicts that traffic for any
particular keyword on a search engine will be proportional to its
Factor in how the title and description affect a user's
propensity to click on a Web site.
Factor in the Penn State University's findings that 55% of users
check out one search result only, and 80% stop after looking at
Factor in known elements leading to an estimated, but educated,
conclusion as such...
Since it's a fact that Wordtracker's Web site appears in the top-
ten of Overture's results throughout their partner realm, they
should be getting a guesstimateed 10% of the overall click-
throughs from all major engines, pay-per-clicks, and directories.
That would equate to about 602 visitors per day.
However, Wordtracker is currently ranked 1-10 on only about 25%
of the major engines, directories and pay-per-click portals for
the search term, keyword(s)... Calculate the estimate...
...therefore, the Wordtracker site should expect roughly 25% of
this predicted click-through traffic, which is 150 visitors per
Compare calculated estimate to known facts...
In fact, Wordtracker receives 10 - 15 visitors per day for the
search term(s) keyword(s). In fact, Overture's STST overestimates
this search query by a factor of 10.
Furthermore, since Wordtracker is estimating they receive
approximately 25% of the total traffic then that would put the
total traffic generated at 40 to 60 per day (25% of 40 to 60 = 10
to 15 visitors a day).
In fact, Overture's STST overestimates the total search query
count by a factor of 100 ...based on 6,016 being more than 100
times greater than the 40 to 60 figure suggested by Wordtracker's
Experience shock and awe at the difference between the numbers!
Wordtracker's service provides very different numbers...
Using the same search term(s) keyword(s), we pulled a
representative result from the Wordtracker database (on January
13, 2004) that predicts searches per day conducted throughout the
major engines, directories and pay-per-clicks on the Internet.
The results were...
keyword - 93 searches (lower case, singular)
Keyword - 39 searches (Capitalized, singular)
keywords - 187 searches (lower case, plural)
Keywords - 184 searches (Capitalized, plural)
KEYWORD - 115 searches (UPPER case, singular)
Total Predicted Daily Searches for all Engines = 618
This figure - 618 - Wordtracker compiled directly from results
taken from Meta-engines, Metacrawler and Dogpile in order to
eliminate the artificial skew.
Wordtracker further adjusted the number downward by filtering out
keyword spam (as defined above) based upon a proprietary formula
used to identify search terms that are being searched at
intervals too regular to have been conducted by actual humans.
These suspiciously regular and assumed to be artificially
generated searches are therefore discounted in arriving at the
final number - 618.
Even when taking into account such dependent variables such as
position, title, and description, we would expect (logically
guesstimate) the website to receive about 10% of the total
traffic due to top-ten placement, targeted title and relevant
And finally, we should expect no better than 25% of that total
traffic, due to the fact that Wordtracker has top-ten placement
in only 25% of the relevant engines.
So the calculations show...
618 x 10% = 61.8 x 25% = approx 15 visits per day.
This is more in line with Wordtracker's actual 10-15 per day
average number of visits generated by the 5 variations of the
search term keyword across all of the major engines.
So, whose numbers should we trust?
When it comes to trusting the numbers, you should take into
account what you are using them for. If you're looking to
determine relative popularity of a given item, service, topic, or
category, then Overture's STST can fill the bill nicely - and for
For instance, Overture's STST returns the following numbers for
the following searches...
58,312 home insurance
57,315 home owner insurance
233,854 auto insurance
570,337 car insurance
This tells us (for free) that car insurance gets about twice as
many searches as auto insurance. It also tells us that home
insurance gets about the same number of searches as home owner
insurance ...and that searches for car insurance is TEN times
more popular than home owner insurance.
No doubt about it, when researching what to sell online, this is
valuable preliminary information that Overture's STST provides
However, based upon what we now know about artificial skew, we'd
want to get a third-party-review of the search terms - one that
adjusted the numbers for skew - before we bought advertising on a
pay-per-click engine or spent good time and money optimizing a
site for organic (think Google) Web search results.
After all, if Overture shows 6,016 "hits" per day out of which
Wordtracker is experiencing 15 visitors, then reality suggests we
should do the math (i.e., apply the information) that distills
the raw numbers into useful data. Let's first decide if "15"
visitors per day will pay the advertising bill (duh!) ...and, if
the reality count is anywhere near 6,016, we'll be ecstatic,
Always remember it's the amateurs that believe optimistically
romanced numbers just before they lose their wallets on the way
to bankruptcy. Professional marketers learn to err on the
downside of expectations and then smile when the pleasant
surprises shower down riches.
They know that nothing beats accurate information - the most
powerful marketing tool on earth.
Grab a free SEO
Tip of the Day - It's
Robin Nobles is Co-Director of Training for Search Engine
Workshops. She has trained several thousand people in her hands-on 2-5 day optimization and
marketing workshops at locations across the globe and her online
SEO courses. Robin is also a member of Wordtracker's special technical support team, and she partners with
John Alexander whose eBook, Wordtracker Magic, offers unique strategies for
applying the Wordtracker service to generate profits for online marketers.
Click here to
go back to the index of search engine marketing
This work is licensed
under a Creative Commons