Why The Google Keyword Tool Is Useless For SEO, Even With Exact Numbers

Not quite on target... Recently there was a bit of a hubbub surrounding Google’s Keyword Tool External (the keyword suggestion portion of AdWords that was made public a couple of years back). It started when a few people, like Barry Schwartz from SERoundtable, noticed that the tool was showing specific numbers for search terms instead of just green bars. Even though at first the numbers only appeared intermittently for people, the official Inside AdWords blog reported the change as a permanent one later that night.

This of course made many people very happy, because if Google themselves were providing the approximate number of searches for keyphrases, then of course those numbers had to be much more accurate that services such as Keyword Discovery, Wordtracker, or Wordze could provide. One would think that, anyways. I mean, who better than Google would know how many times something was searched on, right?

The problem is, however, that those numbers are meant for people doing research into PPC traffic. The numbers shown have very little to do with what people actually search on using Google.com. I learned this the hard way about a year and a half ago, when I decided to try and use AdWords (the internal tool, the one that would actually show me estimated clicks based on position) to pick keywords that I might want to try and optimize for. I identified 3 phrases that I figured would be fairly easy to rank for (which they turned out to be). According to Google, being in the top 3 ad spots (which is as narrow as Google will estimate) I would get an estimated 141 to 180 clicks per day from all three phrases combined. Since natural serps get a slightly better CTR than ads usually do, I figured I would be golden if I could SEO my site to the top for those phrases. Not a ton of traffic, but with the conversion ratios I had on that site for targeted traffic, it would be more than worth it to spend the effort required to get those three phrases to the top.

I managed to make it to the #1 spot in Yahoo for two of the phrases and #3 for the third in a relatively short period of time. Since Yahoo has about 1/5th of the market share that Google does, even without factoring in that natural serps get clicked on more than ads do I should have gotten at least 30 clicks a day for all 3 phrases combined. What I actually got, for the 3 months I was at those positions, was a grand total of 5 clicks for those phrases.

As soon as I saw all of the excitement surrounding the new actual numbers that Google was displaying on the Keyword Tool, I was reminded of my disappointment back then. I decided to set up some testing, and track where all of the actual traffic came from, to help people better understand what those numbers meant, and how they really don’t apply to SEO. I figured I would use my poetry site for this, since I know most of those keywords won’t cost a fortune to test against. I chose 5 keywords, [friendship poems], [birthday poems], [wedding poems], [inspirational poems], and [best friend poems], using exact matches for each. I ran the test for 3 full days, so I could compare the impressions delivered against what the Keyword Tool estimated the search volume was (should be approximately one month’s worth of search volume divided by 3). Some of the phrases got impressions close to what the keyword tool suggested they might. For instance, for [birthday poems] the tool gave a number of 27,100 (which would be an average of 903 searches per day):

[birthday poems] estimated searches
(Click to enlarge.)

and bidding on that keyword for 3 days gave me 2,411 impressions (or 803 impressions per day):

[birthday poems] actual impressions
(Click to enlarge.)

This is fine and dandy if I am only concerned about getting traffic from AdWords, of course. The thing is, if I rely on this data for my SEO efforts I will at best be most likely wasting my time. At worst I will be seriously wasting my time. By analyzing the referrers on the clicks generated during this test we can easily see why this is so.

The Google “Search Network” is much larger than just Google Search

Google said that I had 93 clicks in that time period, but I only had 88 hits to my tracking URL. I’m pretty sure there were some bots thrown in there as well, but since the point of this wasn’t about getting charged for invalid clicks, I am going to ignore that fact for now. Filtering out those clicks would only more so make my point anyways, so as far as this experiment goes I am erring on the side of caution anyways. Of the 88 clicks that did register, only 42 actually came from Google or Powered By Google sites. Those are the only places where you could expect to get traffic from if you managed to rank in Google for your desired phrases. The rest of the clicks came from places such as Ask.com, search.bearshare.com (which appears to be powered by Ask), Shopping.com, etc.

Google includes traffic that wasn’t even generated by searches at all

That’s right, as part of the “search network”, Google will display your searches on parked domain pages. I got traffic from at least 2 of them, including birthday.com:

[birthday poems] actual impressions
(Click to enlarge.)

Notice the links on the side (and on the bottom of the pages, if you visit the site) labeled “related searches”. All someone has to do is click on a link (or visit a cached page that matches one of those links) in order for Google to register that impression as a “search”. It’s a pretty safe bet that a fair number of the “searches” showing on the keyword tool were generated in that fashion. Not only can you not opt out of having your ads showing on sites like that, there is also of course no way to filter them out of the keyword tool.

The number the keyword tool show are worldwide searches

When using AdWords you can of course opt to restrict your campaign to a given country. However, the keyword tool itself doesn’t allow you to filter the numbers like that. I allowed my campaign to run worldwide to illustrate the point. Of the 27 clicks that came in from Google, 1 was from Google Canada, 1 was from Google Books New Zealand, 3 were from Google Australia, 4 were from Google Ireland, and 11 were from Google UK. Only 7 actually came from Google US, and of those, 2 were Google Custom Search Engines (which can vary greatly in the serps they show). That means that when all is said and done, only about 5.68% of the traffic came from unmodified Google.com.

Now, while you might be tempted to do so, you cannot just reduce it down and say, “Ok, just take the number the Keyword Tool shows, and multiply it by 5.68%”. The keywords weren’t distributed at all evenly in that way. [best friend poems], for instance, didn’t receive any official Google traffic during this test. In the end, the Google keyword tool, even with the shiny new numbers it now displays, has to be taken exactly for what it is… a selling tool that acts as a gateway to getting people to sign up for AdWords. Anyone trying to use it as something else will wind up being sorely disappointed.

As for the other keyword tools available out there, they all of course still have merit. While it isn’t free, Wordtracker does have a 7 day free trial for anyone wishing to give them a shot. Li Evans recently turned me on to Keyword Discovery, a service owned by Trellian. And while they don’t have a free version, the Wordze Trial for $7.95 is very reasonable, and probably a good option for someone who only needs to do keyword research occasionally. Even if they don’t have direct access to Google’s secret numbers, at least none of them will try and pass off impressions on a parked domain as a “search” on their network. 😀

Be Sociable, Share!

92 thoughts on “Why The Google Keyword Tool Is Useless For SEO, Even With Exact Numbers

  1. Interesting tests. Granted the traffic numbers reported by Google are “estimates,” and there are the issues of geotargeting and search network. But I’d be interested to know if you checked your exact match impression share for the keywords used in your test.

  2. “… has to be taken exactly for what it is… a selling tool that acts as a gateway to getting people to sign up for AdWords”

    Did you really not suspect this before? j/k 😉

    Nice work Michael 🙂

  3. Very good stuff Michael. I criticize when necessary, but praise when necessary as well. I would have thought the SEO industry would already know what you just posted anyway? Seeing the many SEO blogs saying how great this is… I guess not, huh? The other day when checking the G tool, I totally knew what it was all about… no testing needed, but you just confirmed all my thoughts anyhoo. Nice job!

    Just like years prior, etc, one should always do multiple resources when doing keyword research.. including the new G tool. Nothing has changed.

  4. What do you mean by

    “PPC traffic and actually search on using Google.com.”

    I feel people google.com to search anything …once serps are displayed …people click on either organic or sponsored ads….

    I made this statement with what ever little knowledge i have ….please correct me if i am wrong

  5. Good analysis, it should be definitive proof that in-depth keyword research should not come from only one source.

    And we should always keep in mind, human analysis on the phrases dug up should be applied to determine the intent of the searcher.

    The duration of the test and the popularity of the terms tested had a large effect on the numbers, but the overall conclusion probably would have remained the same.

  6. What do you mean cannot opt out? Of course you can. You can not only opt out of the content network, but also the search network. If you’re getting these referrers from your PPC, you need to call Google for a refund, or open up the campaign settings….

    I find the tool to be very useful when looking for PPC phrases that have a ‘Great’ quality score by entering the landing page URL. Saves me a boat load of time. I still analyze each keyword for convertibility. I almost never pay attention to search volume. If the keyword will convert, bare minimum exact match. You don’t pay for lack of clicks! You pay when people click. Rare clicks are worth many times more than their CPC when relevant….

  7. Thanks for the analysis and post. I think you are putting into words what many of us know intuitively from using the tool, but it does help to see some numbers.

    I would be highly interested in your thoughts on the other tools that you mentioned and I hope you do a post on them.

    – Ryan

  8. “When using AdWords you can of course opt to restrict your campaign to a given country. However, the keyword tool itself doesn’t allow you to filter the numbers like that.” But of course you can.
    I did some similar checks like you, and the numbers turned out to be quite correct. Not in all instances, mind you, but for a set of keywords the numbers fit nicely. Therefore your results are a bit astonishing. I assume your test campaign wasn’t restricted by a daily budget, right?

  9. You reference your impression numbers within AdWords, so I’m just wondering what your impression share is. If you’re only using exact match keywords, then you can run a report to view the share of total possible impressions your account is receiving. If you do not have a 100% share, your impression numbers are not necessarily going to match the keyword tool numbers.

    Also, you actually can opt out of the search network. But I see your point-if the keyword tool data includes search network, that can be misleading for some advertisers.

  10. I strongly disagree!

    You mention above that you used the AdWords traffic estimator to pick your organic keywords. Then felt disappointed by your results in Yahoo. You can’t simply use market share of Yahoo to Google, then predict this traffic. The user demographics that use these two search engines differ. Not to mention that you’re listing could have been the issue.

    “Not only can you not opt out of having your ads showing on sites like that, there is also of course no way to filter them out of the keyword tool.”

    While I agree this distribution network might inflate the search volumes provided by Google, it does provide input on overall popularity. And you can opt out of sites like that. Just turn off the content network, and/ or search network.

    Lastly you’re misinformed about the geo-targeting with Google Keyword Tool. Not only is the following in the description “Results are tailored to English, United States”, but you can edit the location and language. Although I believe Google commented that not all countries currently have data.

    Nice affiliate links on the other keyword tools thought! 😉

    My full thoughts . . .

    (http://www.nonbot.com/blog/index.php/2008/google-keyword-tool-update/)

  11. Ryan (and others), I was in fact wrong about both not being able to opt out of the search network on AdWords, and about not being able to set your location in the keyword tool (no clue how I missed that last one). However, the former has nothing whatsoever about what my test showed. The latter actually means that the estimates that the tool showed were worse than what my testing demonstrated, not better. The tool defaulted to US traffic. The fact that the numbers the tool shows get bigger if I select world wide doesn’t help the tool’s case at all.

    As to your assertion in your post stating that “PPC ad impressions are searches”, that’s simply wrong. An ad showing on sites like Ask.com, Shopping.com, or parked domains have nothing whatsoever to do with traffic you could expect from Google.com. I wasn’t saying that these other sites were mixed in, mind you… they actually composed the bulk of the traffic. Now, searches done on AOL and the like would count, of course, because they are all powered by Google. I counted those already though.

  12. You have taken the quote, “PPC ad impressions are searches”, out of the context of my post. I am referring the PPC ad impressions on Google being general searches. As I stated, this number would be inflated by the factors you mention, but this is not a bad thing. Keyword research is about finding markets and comparing term popularity. Nothing is going to tell you how much organic traffic your site is going to receive. There are simply too many factors at play.

    Even if Google told you that 153,746 people searched for ‘birthday poems’ last month, it means very little for your traffic. You still have to reach the top spot, have compelling listing copy, deal with seasonality / ‘hot trends’ and compete with all other paid and organic listings on the page. Searches does not equal traffic!

    This tool is not a magic 8-ball, but it is a far better representation of online keyword popularity, than the other tools available. It’s Google giving you information, so take it for what it’s worth, and learn to leverage their generosity.

  13. Ryan, I did not take your comment out of context in the least. My entire post is about how you cannot simply use the numbers from that tool as an indicator of what natural rankings will bring you in the way of traffic. Period. If you already knew that, then fine, good for you. A very large number of people in the SEO community will not know that, however.

    Btw, if you want to go spouting your opinion as facts, such as the tool being “a far better representation of online keyword popularity, than the other tools available”, then I suggest that you go test that assertion and demonstrate it with data. Otherwise you’re simply being argumentative.

  14. Wow, do you always attack people in your blog that don’t agree with you?

    I’m simply providing a different opinion, as I don’t entirely agree with your post. I think your research is flawed, or at least bias and it benefits your users to see multiple opinions. If you cut the fat we are making similar points.

    We are both saying the tool does not provide an accurate prediction of organic traffic. I’m just saying that it’s not simply AdWords data, and that there are many other factors that determine search traffic. This is a common flaw with all keyword tools.

    I would rather get my data from Google, over a tool that uses small meta search engines, and calculates volume based on market share.

  15. Heh, I can tell you don’t read my blog much. That’s not even close to an attack from me. 😀

    I provided data and my reasoning for my conclusions. You are coming here (and other places) stating your opinion that I am flat out wrong, but offering literally nothing in the way of evidence to back that up other than “it’s Google giving you information”. Suggesting that you run some tests to prove your point is not an attack.

  16. I think the take away here is that the numbers are not 100% accurate. But they are accurate when it comes to scale and are directional!

    That being said, I think Google’s tools is the best and most reality based tool we have out there right now.

    Keyword Discovery and the the other tools are garbage when compared to the Google tool, but will give you mid tail and long tail terms.

    -Sri

  17. We noticed the same things you are talking about Ryan, great post. We generally use at least 4 different sources for keyword comparisons and research. KeywordDiscovery, the Google Keyword Tool, Quintura, sometimes stumbleupon, facebook, and squidoo as well.

  18. Great post and you speak my mind as well.

    There are a ton of other great keyword discovery tools out there to use. Of course Google has to support their AdWords program and make it easy for their advertisers to do research and spend more $$. They have shareholders to answer to too y’know!

  19. Hello !

    Does anybody know how could it be that in Google’s “External Keyword Tool” near one keyword (exact phrase) I see 2900 searches on average per 12 months and in Google Analytics I see that via this keyword (exact phrase) I have 6789 users to my site per 2008 June? Is it possible that 2900 means unique queries and in Analytics it is only sessions? How to interpret that?

  20. I enjoyed your article.

    However I’d like to point out that Wordtracker’s 7 day free trial is indeed free. You have free access for 7 days without being charged and can cancel at anytime if it’s not for you.

    -Mike
    Wordtracker.com

  21. Well theres alittle bit more you need to throw into the mix,and thats how many actually click on your SERP.AOL says that if you rank #1 that only 22.6% actually click thru.Heres more of a break down:

    Rank Percent
    1 22.6%
    2 6.4%
    3 4.5%
    4 3.2%
    5 2.6%
    6 2.1%
    7 1.8%
    8 1.6%
    9 1.5%
    10 1.6%
    11 0.35%

    Ye s this doesn’t add up to 100% only apx 50% …

    AOL says that almost half (45%+) all queries resulted in no click thurs

    The user just did not find what they were looking for.
    ( *from http://ww.realnichekeywords.com )

    Keyword research is not and exact science.The tools listed above will all give various results for the same word,it depends on how you collected that data (as you have made this great post stating so)

    Its kinda like fishing…
    🙂

  22. AS you said Keyword External is for Adwords users, and no other use. It is not intended as a general keyword tool so I don’t understand the point you are trying to make. Use Wordtracker if you want a general keyword analysis tool, but not Google’s Adwords tool. I find Adword’s Analyzer the best in any case, but prefer Digital Point Solutions for my free tool.

  23. Peter, my point was that most people would assume that it was not for AdWords only. I had people making comments to me like Google showing exact numbers “pretty much puts wordtracker, wordze, et all out of business”. That’s why I decided to illustrate exactly why that should not be the case.

  24. Interesting approach and valuable info. I am using google keyword tool but from other side as adSense publisher to target quality content on my sites and blogs which realy helps me to make decent results..Just my 2 cents

  25. Very good experiment. I would never have suspected that it was counting impressions from parked domains as searches. I started suspecting that those numbers were somehow inflated but now that I read your post I am sure and I should assign less value to the numbers it is giving. Thanks.

  26. Nice post.

    While I was already planning to not rely only on Google’s Keyword Tool for my web site’ SEO strategy, since I am still at the beginning (of both my site lifespan and my SEO knowledge) that tool would have played a major role. Nice to know that I should take such information with caution.

  27. good Post

    I *think* that for Google gives search numbers for that phrase and all phrases that contains those words, not just the exact term. For example: if you search for “How To” Google’s numbers will include all searches for “how to travel”, “how to drive”, “how to do SEO”… Actually maybe very few people are searching for just “how to” without any other words in the search.

    I am basing this just looking at the numbers and a few simple tests. This also makes it not very good for SEO since you could think a short phrase gets a tons of searches when actually it is simply included in a bunch of different longer phrases. So your page optimized to that short phrase ends up getting very little traffic.

  28. Arthur, what you are talking about has to do with the differences between Broad (include stemming, such as -s, -es, -ing, etc.), Phrase (which is what you discussed, where the phrase contains the target terms), and Exact matching. I ran my tests using Exact matching, which should mean that the numbers only reflect those exact searches.

  29. Well doesn’t that just put the cat amongst the pigeons! VERY annoying! I’m going to have to go back and check ALL my keywords now and see what I come out with.

Leave a Comment

*