Google, Ad Transparency, and Query Relaxation: Should Advertisers Care? Probably

April 20, 2020

You need information about Banjo, a low profile outfit in Utah. Navigate to Google and enter the query Banjo law enforcement. No quotes for this query. Banjo has a Web site, and the phrase law enforcement is reasonably common and specific. (It is what is known as a bound phrase like White House or stock market; that is, the two words go together in US English.)

Here’s what the system displayed to me on April 20, 2020, at 0918 am US Eastern time:


The search results are okay. The ads do not match the query or the user’s intent: Law enforcement is not even close to a $1,000 musical instrument in a retail store.

Notice that the first result is to a Salt Lake Tribune article in March 2020 about Banjo’s allegedly “massive surveillance system.” The second result is from the same newspaper which reports a few days later that the Salt Lake City police won’t share data with Banjo. So far so good. Google is delivering timely, relevant results.

But look at the ads. The query Banjo law enforcement displays to a person wanting information about a policeware company the following for fee, pay to be seen ads in front of a buyer with an interest in Banjo:


These advertisers are betting money that Google can get them relevant clicks when a person search for a banjo. Maybe? But when someone searches for the policeware company Banjo, the advertiser is going to be “surprised.” Do advertisers like surprises?

Here are the advertisers whose for fee ads for people interested in law enforcement software (policeware) had displayed in front of a Google user with a vanishingly low probability of purchasing a stringed instrument whilst researching a specialist software vendor selling almost exclusively to police and screened quangos (quasi non governmental organizations):

  • Banjo Ben Clark
  • Deering Banjo Company
  • (note that our Banjo is
  • Banjo Studio
  • Instrument Alley
  • Sweetwater
  • Guitar Center

These companies paid for ads as a result of query relaxation. Google’s system does not differentiate the Banjo policeware outfit from the music products.


Are there parallels between games in which a person can win money by guessing which cup hides the ball? These games of chance are often confidence operations. In this context confidence means trickery, not trust.

Why? There are url distinctions; that is, versus; there are disambiguation clues in’s Web page; there is the metadata itself with the keyword surveillance a likely index term.

This is a very big deal because these advertisers are trusting Google to deliver relevant clicks. The same type of query relaxation or what I call “query repurposing” going on across the Google properties.

The user wants one thing and may get it. The ads displayed, however, are often not correlated with the user’s interests or intent (a big concept for smart search).

The reason is that stray clicks will deliver a law enforcement type to a guitar page. And there is a possibility that that stray click will result in a sale. But that’s putting lipstick on the pig. The banjo vendor wants to sell banjos to people looking for musical instruments, not to a random person intrigued by a $1,000 Goodtime Blackgrass banjo.

The query relaxation (repurposing) is designed to burn through ad inventory, generate clicks no matter what, and get the advertiser to replenish the ad spend account.

Stated simply: Google has to burn through ad inventory to generate revenue. After two decades of trying, Google is as Steve Ballmer observed, “a one trick pony.”

Now navigate to the Wall Street journal either the dead tree edition or the online version and read “YouTube Spars with Auditor over Transparency on Ads.” The write up asserts on the basis of a single “leaked” memo:

Google wants to substantially limit the information a key auditor of YouTube can share about the risks of advertising on the video service…

Close but not the whole cigar of relaxed and repurposed queries. Ad matching is a variant of search. Search matches a user’s query to a term in the index. An exact match is what most people assume is happening. Applied to advertising, the musical instrument vendor assumes that the word banjo will be displayed in the context of a music centric query.


YouTube does the same thing: Ads don’t necessarily match the index generated for the advertisers’ ads. Furthermore, data about skip ads after five seconds is not publicly available. Nor is there data about click aways when ads unrelated to the user’s video appear in the middle of a video.

The write up quotes a Google professional as saying:

We know how important it is for our industry to have a healthy third party ecosystem of trusted independent solutions for driving and measuring marketing performance on YouTube.

There are several issues in play:

  1. Google wants to make sure that advertisers’ messages are not displayed in an inappropriate context; for example, criminal videos, crazy news about disease, etc. But finding such mismatches between videos and ads is a tough job. Using the pandemic play, Google suggests that some mistakes might slip through. Just try to find those slip ups.
  2. Google wants advertisers to keep spending. This means that data about ads has to be happy data.
  3. Google has to burn the ad inventory.

Back to query relaxation or repurposing. The basic idea is to shift from exact term match to a fuzzier approach; that is, a term appears in a general category. The relaxation means that an ad can apply to terms not closely related to what the advertiser expects. The other aspect is what is called query repurposing. The relaxed match is applied to content that is related to the relaxed query. This is moving the criteria for a match farther ways from the exact match. This is like a friend of a friend in a social graph. You want to sell to the closest friend of a target. Instead you are selling to a friend of the friend; that is, a couple of steps away in closeness or relevance.

What’s interesting to DarkCyber is that the flaws of the core ad approach are easily visible in the Banjo law enforcement example.

Also, the ineffectuality of the third party services to hold Google’s paws to the bright flames of precision and recall.

Maybe someday. But for now Google will try harder and the hand waving of third party outfits will have to suffice.

Who loses?

  • Users, of course, because the old school promises of Oingo (Applied Semantics) are lost like the secrets to the stone cutting on display in Machu Picchu.
  • Advertisers because the individuals these companies must reach to survive may not see the messages in the time, place, and frequency the advertisers assume the messages will appear.
  • Google. Plagued with soaring costs and competition from annoyances like Amazon and Facebook, the GOOG has to generate revenue and leave no digit unflipped. At some point, the marks may quit the game.

Net net: Watchdogs, think about semantic relaxation and repurposing. Do more with facts and less with PR.

Stephen E Arnold, April 20, 2020


Comments are closed.

  • Archives

  • Recent Posts

  • Meta