Expanded Broad Match and The Google 1-2 Punch

By Mike Churchill
(Special thanks to my colleagues Jim Gilbert and Liana Evans in researching this article).

There has been a lot of discussion on Google’s recent changes to the way that they handle expanded broad match (at WebMasterWorld , High Rankings Forum and other places).

We have come across a different issue that relates to the investigation of expanded broad match, and has wider-reaching repercussions for your PPC campaigns, as well as understanding oddities in interpreting Analytics. We have been seeing this issue for the last month or so (since Aug-Sep 2007), and according to the Google engineers with which I have discussed this: “the search results […] are the result intended behavior. When determining which ads to show on a Google search result page, the AdWords system evaluates the user’s previous search query as well as the current search query.”

The Google Match Problem

Google is combining the search queries from two successive searches when serving up the PPC ads. If a Google visitor makes a search, then uses the search box on the first search’s results page, the original query AND the second query are BOTH used to determine the ad to display. My colleague Jim Gilbert refers to this as the “Google 1-2 Punch”, and it can end up costing you money and leaving you confused if you don’t take steps to combat this new algorithm change.

Why is this a problem? Ads may be displayed for inappropriate searches, resulting in unnecessary expense for the advertiser.

Examples:
Using Google, search for:
Golf clubs
Then on that results page, search for:
Women’s perfume

Here is what I see:

women's perfume Google search results showing the Google 1-2 punch

In another example, I searched for:
mp3

And got these results:

Google search results for mp3

These results look substantially different if I search for:
accessories
And then search for:
mp3

Google search results for mp3 after searching for accessories

What Is Happening with Google AdWords


If the Google visitor uses the search box from one search to make a second search, when the second request is sent to the Google site, the HTTP_REFERER field is filled in with the URL of the first search request. In this second example, this means that the HTTP_REFERER line in the header would look like this:


GET /search?hl=en&q=mp3&btnG=Search HTTP/1.1
Referer: http://www.google.com/search?hl=en&q=accessories

(There are other lines in the HTTP Request, but these two are the relevant ones.)

I believe that the Google site is seeing the q= parameter from the HTTP_REFERER in addition to the q= parameter from the actual GET request, and is using BOTH terms to determine the relevant PPC ads to display. Note that this issue is NOT reflected in the organic search results.

Try it for Yourself and See


It is easy to reproduce this issue for yourself: Search in one of your current PPC campaigns for a phrase match KW that is at least 3-4 words long (let’s say “Okra ice cream sundae”). (It also works for two word phrases, but this is my example). We want to select a phrase that does not include a shorter phrase that might trigger your PPC campaign. Therefore, insure that the phrase consisting of only the last half of the KW (e.g. “cream sundae”) does not trigger one of your ads. Starting with a newly opened browser, search for the first half of your KW phrase (“okra ice”). Then, using the search box on the results of that query, search for the 2nd half of your KW phrase (“cream sundae”). Your ad should appear (assuming the campaign is active, has remaining budget, etc.,etc.).

This is a factor for broad, phrase, and exact match KW phrases in your AdWords campaigns.

Why do I think it is the HTTP_REFERER field?


The problem is only seen if the visitor uses the form on the results of the first search to perform the second search. If you pass the same two URLs requesting the two searches, but either type the queries manually, cut and paste them, or use a bookmarked query, the organic search results are the same, but the PPC search results reflect ONLY the second query’s terms. In all three of these cases, the HTTP_REFERER is not filled in by the browser when sending the search request.

Insidious Part


Search engine marketers (SEMs) researching the effects of Expanded Broad Match tend to make a lot of back-to-back queries to see which of their PPC ads are being displayed. Since this effect only comes in to play when you make a 2nd search within the search box on the results page of an earlier search, SEMs are likely to have been tripped up by the Google 1-2 punch.

Both expanded broad match and the 1-2 punch have the effect of having an ad displayed for words that are not in the current search query, so many SEMs researching expanded broad match may actually have been seeing the effects of the 1-2 punch instead (or in addition).

What Can You Do About It?


Unfortunately, not a lot. There is nothing in the Google AdWords user interface that allows you to opt out of this new feature (nor the expanded broad match feature, which is dearly desired by many, but is unlikely to happen (See related artcile on Expanded Broad match)

There is a partial defense: the traditional defense against expanded broad match is to use negatives in the campaign to explicitly request that your ad NOT be displayed when one of those related terms is searched for. Thus, if I were selling stainless steel knives, I might include “pots” and “pans” in my negative list to keep my ad from matching to searches for stainless steel pots.

With the Google 1-2 punch, it appears that using a negative exact match phrase in the adgroup may prevent the ad from being displayed – even if that negated phrase is NOT the phrase being searched for in the second case!

For example, I set up a test PPC campaign with a single phrase match KW “okra ice cream sundae”. Searching for “okra ice” followed by “cream sundae” caused my ad to display (as expected). Here is the good news: Adding “–[sundae]” to my negatives list (which should prevent my ad from displaying if someone searched just for sundae also keeps my ad from displaying for the Google 1-2 punch. Curiously, using “-[ice]” keeps the Google 1-2 punch from happening, even if “ice” is not a part of the second search.

This is a work-around, and it will add needless complexity to your PPC campaigns. It will also increase the processing for the Google servers as customers’ negative lists expand extensively to counteract this issue. I firmly believe that the best solution is for Google to give AdWords customers the option of opting out of both expanded broad match and the Google 1-2 punch, and have the Google AdWords system work as advertised.

We will continue to research this issue, and provide updates as we learn more.

Share and Enjoy:
  • Print
  • Digg
  • StumbleUpon
  • del.icio.us
  • Facebook
  • Yahoo! Buzz
  • Twitter
  • Google Bookmarks

10 thoughts on “Expanded Broad Match and The Google 1-2 Punch

  1. Hi Mike,

    I notice this now and again and I blogged about it a couple of weeks ago. I hadn’t been able to see any real pattern to it but I’m able to replicate the test you describe with examples from my PPC campaigns so it gives me more of an idea now what’s happening. It results in some really odd mix of ads at times! Not good.

    Regards, Christine

  2. Another thing to consider is the impact this will have on dynamic keyword insertion whereby a search keyword is entered into your ad.

    This dynamic insertion feature, which in my opinion is quite a good one, basically becomes nullified by expanded broad match if it actually enters you newly search upon irrelevant string into the headline.

  3. Adding on to my above post, has anyone come across this actually happening ?

    It’s only speculative on my part.

  4. As long as the referrer has the erroneous keyword you can still build negatives…but probabalistically, precedent search habits cannot be predicted, thus it’s touch and go and trust google to some degree.

  5. Very Interesting!
    But what does Google want to accomplish with this move – beside earning more money.
    I can’t see how this will gain the quality of the searches, only mixing the bidding war between different markets.
    Best regards,
    Christian

  6. Thanks for this, Mike. I think you have described it perfectly.

    Note to Christian: I guess Google’s intent is to determine the searcher’s intent further, because (except when we’re talking about search professionals) often when people perform two searches in succession, the second search is a refinement of the first one. Once again though, as with Expanded Broad Match, advertisers did not opt in to this feature and are not given the option to opt out. Unlike Expanded Broad Match, Google didn’t even bother to announce or document this “enhancement”.

    Note to Pierre: It will have not have the implied impact on dynamic keyword insertion, because the keyword that is dynamically inserted is the trigger keyword that appears in your Adwords account, not the query that the searcher typed (or the extended query derived from the referrer query and the current query). The only impact it will have is that some keywords will be triggered that you would not necessarily expect to be triggered.

  7. Rather than assigning evil intent, or profiteering as motivations for Google. perhaps we could also consider that they are running a test of behavioral ad targeting on direct search results?

    The premise behind behavioral ad targeting is that you get to present ads that reflect what’s on the person’s mind even when they are doing something else. This how the behavioral ad networks operate. Look at a car site one day, and all of a sudden every website you go to seems to have an ad about cars.

    While it may seem unlikely that this type of behavioral targeting would translate well into regular paid search results on a search engine, how would you know unless you tested it? Seems like a good thing to test, if you believe in behavior targeting in its current iterations.

    My opinion is that behavioral ads mixed in with keyword relevant ads would be a long-shot to perform well, but how can you know unless it gets tested? As outrageous as an idea may seem, it is quite possible it could turn out to be an effective generator of effective broad matched traffic.

    Just a theory, at any rate.

    Matt VW

  8. Hello Pierre,

    I don’t have empirical data on keyword insertion as we don’t use it much, but I can reply to Matt that in our markets we’ve found the the additional clicks from expanded broad match were next to worthless in sales generation (insurance and real estate).

    Next to useless means three times higher advertising costs and 25% more sales. I’ve written up my own experience on how to combat Google’s expanded broad match cash grab.

  9. i’m a adword’s publisher and i think the query generated in the google search are always “pertinent” to that the people are looking for.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

*
To prove you're a person (not a spam script), type the security word shown in the picture. Click on the picture to hear an audio file of the word.
Anti-spam image