In the never ending quest to interpret the behavior of Google Ads A.I., Google’s Responsive Search Ads are the latest mystery. If you have been following the “recommendations” which the Google Ads algorithm continuously presents, you have been creating a dynamic ad with maybe 15 different headline variables, times 2 or 4 description combinations.
The idea here is that Google’s artificial intelligence will predict and/or interpret searcher behavior such that it will serve up the most enticing combination of headlines and descriptions to the most relevant searcher. Don’t be alarmed if this seems to be working … but isn’t.
Yes, as we stretch our imaginations to “add more keywords in headlines” (which of course means adding more headlines) it seems that our organic, human searchers do indeed click on the ads, yet they don’t seem to convert as much as when we are only running well crafted expanded text ads.
In our tests (and this is all the detail we can provide and still preserve some competitive intel) we tend to find Google Ads serving the responsive ads much more frequently than the older expanded text ads that used to perform so well.
We recognized this problem to the extent that we simply paused the responsive ads altogether. Lo and behold, expertly crafted expanded text ads were suddenly on fire, giving us 4 times the conversions at about 4 times less cost.
Once again, Google Ads artificial intelligence is quite remarkable … if you do not know how to contextually craft a compelling static text ad. Google Ads responsive search ads will attempt to do that for you, artificially … by shuffling your random attempts to create more random possible combinations. In the process it loses the human element of contextual relevance. And when you really think about it, this makes sense. You’re being asked to create all these variables for the sake of variables and they get mixed up and create ads that humans will naturally process as seeming “artificial”. In a world that is saturated in the artificial or “derivative”, human beings much more effectively respond to, or trust, genuine, organic messaging.
Not to be some Messiah of the failings of technological virtue, but we still need to at least mention that our analysis shows that machine learning does not (yet) communicate with organic searchers in the same compelling context as a clear and determined, person to person “call to convert” message. Casting a wider net does not always catch the best fish.
At least Google machine learning doesn’t. Or has your android’s latest update really impressed you with something ground shaking lately? Wordstream also has a very good article about how the system is supposed to work. Maybe we’re just bad at adding keywords to headlines, but what we have found may be something for paid search advertisers to consider.