Posted on December 4th, 2010

Google’s semi-radical shift, to-instant-while-you-type SERPs, mostly served to codify the dumb-em-down effect caused by herding users via the search suggest Ajax box.  In tandem with Google’s sad business decision to neuter the AdWords Keyword Tool, all but the best PPC marketers are now funneled to operate predominantly on the short tail. This represents big G’s jihad to consolidate inventory and make more money.  Some would say, with the advent of Modified Broad Match, this is not all bad. To me, it mostly sounds like some scrape and bake fun-time testing!

Recently in that spirit AIMCLEAR, along with a handful of industry colleagues, took a few spirited cracks at hacking ‘Instant to see if we could somehow break it, to the #PPC marketer’s advantage.

OK, let’s start hack 53. Testing the suggest REST API, reveals the personalized nature of suggestions, most obviously by the geography of users’ internet access. The word “first” is a great suggestion API test because the keyword has inherent geographic contingencies.

We wondered if Google Instant offered any opportunities for #PPC sharks to exploit seams, as users type.  The objective was to discover if we could leapfrog competitors, get ads in the SERPs sooner and/or find a way to lower costs.  After tons of testing (52 tries)  our friend Jon Thralow discovered this little hack and thought our readers might enjoy it (special thanks to Jon and our friends at mozenda). Also PPC savants Matt Van Wagner, Brad Geddes and David Szetela gave input to this test.

Scraping Protocol
The Google Instant seam is easily illustrated by the keyword, “data extraction.” First we emulated common human search behavior by typing one letter at a time. For the first letter, “d,” we used mozenda to scrape the PPC ads that appeared in the instant Google SERPs.  Then we repeated the process for “da,” again extracting PPC ads in the SERPs. We repeated the process for each new letter as we typed the keyword and scraped the PPC ads for each fractional combination of formative letters.
data e
data ex
data ext
data extr
data extra
data extrac
data extract
data extracti
data extractio
data extraction
We built a database that shows input (the letters typed), the suggest term (in the search suggest Ajax box), the PPC ads from the top and right hand SERPs.  Here’s a screen capture of the DB.
Our goal was to find a gap, where a partial of the keyword “data extraction,” returned no PPC ads.  That’s exactly what we found for the partial “data ex.”  At that point in the keyword typing process, Google suggested “data execution prevention,” which was exactly the keyword we then bid on.
Sure enough, bidding on “data execution prevention” (not our actual KW) at Broad Match showed our ad. This occurred before either the user or Google suggestions got to “data extraction,” a much more competitive and therefore more expensive KW.  As a side note Exact Match did not trigger the ad.

The ad showed, at different times,  in both the top ad RH positions.

At the beginning, before the quality score was established, the result was impressions, clicks and a cost that, while impressive, was not as exciting as what we hoped for.

Then the clicks added up and the quality score improved. We did not artificially inflate the clicks. They were natural.

The results were fascinating. Essentially we displayed our ad early, for a KW that was uncontested and much less expensive.  The clicks were in addition to normal clicks received bidding on the highly contest “data extraction” keyword.  We assume that is the case because of the clutter in the ‘data extraction” SERPs.  The cost was $.20 average CPC, as opposed to $9.00 +, the typical CPC for bidding on “data extraction.” While additional low-cost clicks were a relatively small percentage of overall clicks for the full KW, they absolutely diluted the campaign’s overall CPA.  That’s what we were after!

Also, it is possible that the early display of the ad, usurps other “data extraction” SERPs players, by branding early in the SERPs as a user types.  We will study this lift at a later time.

Tactical Takeaways
  • There are freaky geo-considerations to this tactic. Not many folks search for “data extraction Minnesota.”  Geo considerations would would render this technique impossible to manage.
  • It won’t work for many keywords because not all SERPs have uncontested gaps to target as the word is built.
  • I presented this specific case study at a conference so we would expect Google to shut it off. Google Instant algorithms may be in active revision. Expect changes.  Google won’t like this.
  • Is it scalable? Nope it’s probably not.  This technique is just a little fun between PPC monkeys.
  • Does this information justify systemic screening of extremely expensive keywords, just for the heck of it? Probably yes. In the odd case where this technique works, a marketer could save a lot of loot.
  • Dear Google, please give us a Google suggest API, where we can filter by region, unpersonalized.  Also you might throw in restoration of the old KW tool. The new one sucks.

Subscribe Today

We'll keep you updated on the latest Aimclear musings & appearances

  • Mike Marshall

    Very cool and very well laid out as always Marty! Great read – I like the methodical approach to the experiment.

    • Marty Weintraub

      @Mike Marshall: Thanks dude. It’s great to see you here and thanks for reading such a twisted technical rant.

  • Lauren

    This is so geek-chic, I love it.

    • Marty Weintraub

      @lauren: Dude, I know you are but what am I? 🙂

  • Lauren

    @Marty: Game, set, match. Well played 🙂

  • Justin

    Really interesting experiment and a nice discovery. Appreciate the clear and informative post, thanks Marty.

    • Marty Weintraub

      @justin: We’re glad you appreciate it. Without testing, life itself would be impossible. Thanks for checking out the post and the kind words.

  • Olivier Amar

    I tested this 2 days after instant went live on a big Forex site and kept my mouth shut about it..

    Nice Marty 🙂


    • Marty Weintraub

      @Olivier Amar: I understand. We used it for months and presented it at SMX East, before publishing in the blog :). That wasn’t to be coy though, just don’t have all the time I want to blog! Thanks for stopping by and rock on!

  • Izzy

    Great post! have bookmarked and shared your site amongst my team. Good to read I’m not the only one that thinks the new Google Keyword tool is rubbish… I often use the Wonder Wheel scraper and Google Suggest scraper to build highly targeted keyword lists. Keep up the good work guys!!

  • Cormac


    I’m using this technique by broad bidding on “windo” for a client that sells window blinds. The Ad Preview tool displays the ad when I type in windo but when I go to Google Instant it doesn’t appear. Strange.

    Any ideas why that might be the case?

  • Sean M

    Exact match didn’t trigger the ad but broad did – what query is getting credit for the impression then? Is it [data ex]?

  • Jon Thralow

    @Sean M: When I tested this, one of the things I found most interesting was that google displayed the ad based on the amount of the bid. If “data ex” was exact and bid higher then “data execution prevention” broad it would trigger “data ex” Google is nice and tricky like that. Google always optimized the amount of money they were making.

  • Bryan

    Awesome post guys… Keep up the good work.

    The fact that you ad copy wasn’t really relevant to your target keyword wasn’t a huge road block for QS?! Did you always use the keyword in the ad copy? Wouldn’t that lead to rather nonsensical ads?

    • Marty Weintraub

      @Bryan: …which was what we thought at first. However, that tells ya’ what you need to know about QScore now doesn’t it? :). The CTR drove QScore. Google mostly cares about money, which (I guess) equals “relevance.” Thanks for the kind words and thanks for stopping by.

  • Bryan

    Did you have any trouble getting the ads to serve initially? I am having a problem because the keyword relevance is too low for the ad to serve. Thoughts? Thanks!

  • mike rhodes

    Hey Marty – great post dude.
    Just to echo Sean’s comment – what actually did show up in the SQ report?
    Also did you manage to test this for modified broad, or just broad?
    very interesting
    mike (aka ppc monkey down under)

    • Marty Weintraub

      @mike rhodes: Jon ran the test we published. My tests accomplished it at modified broad, broad and phrase, but not exact. It’s a wacky flacky deal and I’m not sure you can put rules to it. Just plain works some times and not others. Let us know what you find.

  • Eric

    Great post. We were wondering about this at our company and you post confirmed my thoughts that you have to bid on the keyword that comes up on instant and not just part of it.

    We are now testing it in a very competitive niche in various countries. Just launched it yesterday in fact. We have the keywords in both broad and phrase as you suggested. We’ll see how it goes.