My Satisfying Approach For Making Quality Responsive Search Ads (RSAs)

I have some issues with the approach of creating Responsive Search Ads (RSAs) as soon as you start a new campaign in a Google Ads account.

First, RSAs use the headlines and description lines you input. How do you know if those inputs are any good? Until you have data that proves it to be the case…you don’t. You may have a general sense that they’re good. But you don’t really know.

Also, you don’t know how the Google Ads system is going to react to those inputs. I’ve used ad elements before I thought were great for my audience and the system just wouldn’t serve them as much as other elements for some reason. The system gets to decide what it thinks about the relevance of your ad based on its imperfect semantic technology.

Second, you get no conversion data. How do you know if the elements and combinations that are getting the most Impressions in your RSAs are actually the best performers? You don’t.  How do you know if lead quality differs by combinations?  You don’t.

All you know is the combinations the Google Ads system likes best. But not necessarily because they’ve resulted in the best performance.

picking responsive search ads

Third, the Google Ads system makes decisions too quickly. I’ve created Responsive Search Ads at the beginning of campaigns only to jump in after a couple of days and very minimal Impressions to find most of the Impressions already heavily weighted toward only 1 or 2 headlines or descriptions.

Fourth, there can be too many combinations to test. RSAs are supposed to take the ad elements you input and act like a multivariate testing tool that puts them together in unique combinations to show to searchers on your target keywords.

If you follow the recommendations to add a bunch of headlines and descriptions, there will be thousands of combinations to test; each of which will need many Clicks to get any idea of how it would perform over the long haul.

Sure, you can pin elements in certain positions to cut down on the possible number of combinations.  But do that too much and you’ve essentially eliminated the reason for using Responsive Search Ads in the first place.

Especially if you have a smaller budget account, getting enough data on how combinations perform would take forever. You may not even be managing the account by the time any conclusions can be made :).

Combine human & machine abilities

Because of all of these issues, I’ve started using the approach of not creating RSAs as soon as I start a campaign. Instead, I take an approach that combines our human ability to make rational judgments and come up with quality inputs with the machine’s ability to use those inputs in an optimal way.

The reason this type of approach is important is because machines aren’t good at understanding and communicating with humans in complex ways.  They’re much better at finding patterns and making decisions with how to use those patterns.

Responsive Search Ads are a good example of this. Machines can’t make good decisions about what the inputs should be. They can only work with what a human gives them.

If a human gives them 6 pretty crappy headlines, the machine will work to find out the level of crappiness each headline contains over time and show the most appropriate crap to each human as it notices patterns in performance. Your best elements are being shown, but they’re only relatively good crap compared to the rest of the crap you input.

So…here’s what I do to keep this from happening.

My step-by-step approach to Responsive Search Ads

I start with 3 ad variations whose elements differ fairly significantly and I put the campaign on the Rotate Indefinitely ad rotation setting.

For example, for a B2B advertiser offering a virtual fitness platform, I might start with these 3 ad variations…

Click to enlarge

I run these variations until they get fairly significant data, but not complete statistical significance. Unlike an A/B testing situation, I’m not testing these ads and looking for a winner. I’m simply wanting to make sure the variations I’m using are of good enough quality to use in my Responsive Search Ad.

Remember, I only want quality inputs.

After running these ads for about a month, here’s the data on them…

Click to enlarge

The top 2 ads from the ones above are meeting conversion performance benchmarks, while the third one couldn’t win anywhere close to the number of Impressions as the other two.  Clearly it doesn’t belong in the mix.  So, I eliminate that ad and replace it with another one…

Click to enlarge

I monitor these ads for about another month and analyze the results…

Click to enlarge

The new ad got WAY more Impressions and Conversions.  Nice.  But hold up.  We analyzed lead quality and found that this fourth ad didn’t pre-qualify users well enough.  Most of the Conversions were for consumers looking for online workouts, which is not what this company offers.  So that ad’s not going to work.

So I eliminate it and replace it with another one…

Click to enlarge

Notice how I more clearly pre-qualify searchers with this ad.  We learned our lesson on that last one :).

After running this for a few more weeks, here were the results…

Click to enlarge

While the new ad had the least Impressions of the 3, it wasn’t drastically lower.  Plus, look at the Conversion numbers!  Wow, my pre-qualifying those Clicks is working as that ad is crushing it.

Better performance results

I feel like I have a choice now.  I can be satisfied that I have 3 quality ads and use their elements to create my RSA.  Or maybe I want to keep testing.

The point I wanted to show here was the process I go through to make sure I’m creating a quality RSA.

Using this process, I make sure I’m doing what I can do well as a human (creating quality inputs) and then giving those inputs to the machine to do what it does well.  Doing this ensures I get better performance out of my Responsive Search Ads than blindly inputting variables I have no idea are any good into a machine that gives me no data and makes decisions too quickly.