Responsive Search Ads – Machine Learning Meets Advertising

Mobile Responsive Search Ad

Responsive display ads first arrived as a new Beta display ad type in Google PPC display campaigns in 2018. Responsive search ads were then introduced by Google and rolled out by Microsoft in mid-2019.

The ad platforms have been surprisingly tight-lipped about their purpose, and they have been designated “Beta” status for much of their existence. Analytics about the performance internals are limited and a bit sketchy.

We’ve found that responsive search ads work best running alongside multiple extended text ads within ad groups. Responsive search ads are not currently stand-alone replacements to traditional text ads. Rather, they augment them and can be utilized as a sort of ad copy split-testing machine. Display responsive ads are a bit different, providing a more efficient way to deploy display ads, and capable of entirely replacing the previous generation of display ads.

Both search and display responsive ads, when rendered, look the same as other display and text ads. However, the images, text, and video content within responsive ads can be served differently by the ad platform each time an ad is served. Responsive ads can also adjust their form factor and size to fit onto a broader range of devices and screen sizes. These self-train and serve adaptively, with the goal being to deliver better performance.

Display responsive ads can serve in “native format” on web pages across the display network, fitting into the style of web pages better than previous generations of ads.

Why were responsive ads introduced?

Of all the ads running in PPC campaigns across Google and Microsoft, many campaigns have poorly written and/or low-performing ad copy. This is not surprising as ad copy development is a learned skill, and many advertisers don’t have the experience or knowledge needed to consistently develop strong ad copy.

Responsive ads are an attempt to at least in part, improve the performance of campaigns by bringing some automation into the testing of ad copy. This makes sense when you consider the immense scale of ads running globally on the two largest ad platforms today. When run alongside extended text ads, they should be populated with about one-third the strongest copy used in extended text ads, with the rest of the copy being new and perhaps untested. The Responsive ad will serve as a sort of split-testing machine, discovering the best of the new copy to serve with the known strong copy.

The Chameleon-Class Ad Type

This new variant of search and display ad is “responsive” in several ways. They can adapt their size and form factor to display on more device types and search engine result pages. Responsive display ads can also render in native styles on web pages across display networks.

Available ad display space varies by device type, size of the screen and is most significant for display ads. Responsive ads can adjust their size and form factor to fit into the “screen real estate” available. Responsive ads do this by adjusting their ad unit size, aspect ratio, and the number of fields of ad copy displayed.

This allows them to have a greater reach, serving onto more device screens when compared to standard ads. Said another way, for a given ad auction and search engine results page (SERP), responsive ads can reach more device types and screen sizes than fixed format ads.

How are responsive ads different?

Compared to standard ads, responsive ads must be pre-configured with 2 to 5x the quantity of ad copy and media as compared to standard ads. As the ad platform serves different permutations of the ad copy and media, a machine learning process records the outcomes (click-through rates). This facilitates the automated discovery of which variants of ad copy which is generating the strongest click-thru-rate for a set of ad auction conditions and metadata.

Responsive ads can select the best performing combinations of images, video, and text ad copy to serve for a particular ad slot. Different ads can potentially be served tailored to the metadata held by the platform.

This is a powerful approach to ad optimization that clearly has potential in large subsets of ad campaigns. Let’s take a look at some of what we’ve learned about this new class of ad, which remains a bit shrouded in mystery. We have learned quite a bit from using them and observing their performance.

How do responsive ads work?

The performance of images, text, and video content in responsive paid search ads are delivered in different combinations while being continuously sampled by a machine learning algorithm. Responsive PPC ads self-train and adapt to serve the best-performing ad copy and media into individual ad slots, tailoring the ad’s content, the sequence of ad copy, and best-performing media for each particular search query (ad auction).

Before the first impression of a responsive ad is served, the ad platform uses heuristics – proprietary rules – to pre-assess which images and ad copy are more likely to generate engagement. So ad copy and images are ranked before ever being served.

In responsive ads, after thousands or even, tens of thousands of ad impressions have been served, we will often see ad copy that has never served a single impression. The rules for ranking ad copy are proprietary, so you may be disappointed to see some great ad copy never served. This is one of many mysteries about responsive ads due to their “black box” or proprietary nature. How and why some strong ad copy remains unserved within responsive ads remains a mystery. We have some theories about that which we’ll leave for a follow-on article after more testing is completed.

As a responsive ad runs, the copy that the platform approves of is served in different combinations. After enough data accumulates, you’ll be able to see the different arrangements that the ad copy was served – per the mobile ad screen diagram below.

163,800 ways to serve a single search ad

When a responsive ad is created, it is populated with additional ad headlines, descriptions, and media (images and video). Refer to the diagram for a comparison of a standard text ad, to a responsive text ad.

Responsive search ads allow up to 15 headline variants and 5 description line variants to be pre-populated. In a responsive search ad, instead of the usual 2 or 3 lines of ad copy, there are 15.  There are 5 descriptions instead of 2.

In a responsive search ad, any of the 15 headlines can be served into one of 2 or 3 available headline fields, and any of the 5 headlines can be served into one of the 2 available description fields. This diagram shows how ad headline and description ad copy is introduced into a responsive search ad.

The math of responsive

By doing a little arithmetic we can see how many combinations of ad copy may be served by a responsive search ad. The ad shown in the accompanying diagram can be populated with up to 15 headlines and 5 descriptions. As shown, there are three 35-character headlines and two 90 character description slots this copy can be served into (displayed).  There are therefore 163,800 unique ways this ad can be served.

If we “pin” (lock) one ad headline to a position (not a recommended practice) this comes down to 3,640 combinations. Even the most diligent campaign manager is unlikely to test even a small fraction of these possibilities of ad copy during manual split testing.

Responsive ads are served in “learning mode” until an adequate volume of impressions and clicks are accumulated to “learn” how different combinations of their ad copy and media perform. They then begin to serve what the platform deems is the best ad copy after winning an ad slot at auction. The ad platforms do not disclose how closely the ads are tailored to suit different auction conditions.

In theory, responsive ads can serve just the right media, and/or ad copy in the ideal position within the ad to achieve the best result for a given, individual auction. In other words, generate the highest amount of engagement for that particular ad auction and instance of the ad, which is measured in the click-through rate. This is an extended topic that’s beyond the scope of this article.

Semi-automation of manual split testing

Even veteran ad copywriters end up discarding much of what originally seemed like great ad copy. Strong ads are a part marketing-advertising art form and part logical skillset. Great ads rarely come from the “first cut” batch of ads in a campaign.

Campaigns that are actively managed by advertisers and PPC pros and the ad copy split testing process consume considerable time and effort after a campaign is launched. When split testing ads, periods of time must lapse between testing new combinations of ads. Otherwise statistically significant data won’t be accumulated and mistakes will be made.

PPC pros tend to be aggressive with ad split testing because it is central to campaign performance to have strong ads. But for a high percentage of PPC campaigns, ad optimization simply never takes place. Responsive ads were introduced to address that issue. They shift more effort onto the initial ad setup and introduce automation to reduce the amount of split testing required.

How and when are responsive ads superior?

In our experience, and from other industry reports and studies, we know that responsive ads can often pull a higher click-through rate (CTR) and in our experience, increased overall conversion rates & conversion volumes.

CTR is a useful proxy for conversion rate in new campaigns before there is conversion data available. But it is not always an accurate proxy. People can click on ads for a variety of reasons, including misunderstanding the message. In theory, responsive ads can evolve to learn how to deliver better conversion rates and conversion volume, not just click-through rates.

In our experience, responsive ads are a good complement to traditional ads when they are deployed within the same ad group, and used as a new, more automated way to test new ad copy. But because of the limitations on analytics provided, they are somewhat handicapped in that regard.

Can we measure if a responsive ad is outperforming a standard text or display ad?

Yes. The best practice when introducing responsive ads into a campaign is to run one responsive ad alongside 2 or 3 traditional text or display ads within an ad group.  This allows observation of the relative performance – click-through rates, conversion rates, and cost per conversion (CPA). It is beneficial to introduce at least one-third Headline descriptions which are different than the ad copy used in the opposing standard ads.  The more unique, well-written ad copy you can create for responsive ads, the better the chances for successful outcomes.

What are the downsides of responsive ads?

The primary issue that advertisers, marketers, and campaign developers have with responsive ads, is that the ad platforms are not disclosing many of the most important details of how responsive ads actually work. The analytics on the performance of individual lines of ad copy provided is limited.

Responsive ads, therefore, remain largely a “black box” to advertisers and agencies. They are useful, but they also have some major issues and limitations which are difficult to sort out due to their proprietary nature.

Advertisers and agencies in certain markets e.g. regulated markets or those that tightly control their brand messages, ad quality, and campaigns may sometimes not approve of ad copy served by responsive search ads, even if it outperforms. This is often for good business reasons. Some advertisers feel a strong need for the sequence of ad copy.  For example, many advertisers want to serve a unique value proposition in the first headline, followed by features-benefits type information, follows by social proof and a call to action.

Responsive ads put more control of what ad copy a campaign serves and what order the ad copy serves into the hands of the ad platform. They take control away from advertisers and their agents.

Ad platforms hold a dominant market position, so without much direct competition to be concerned about they are free to control their ad platforms as they please. Responsive ads are another front in an ongoing struggle between the ad platforms on one side, and advertisers and their agencies on the other. Many advertisers are pro-automation but also want to retain the flexibility to optimize more detail of how their ad copy serves.

Responsive ads take a lot more time to initially configure, and that time can sometimes be better invested into manually optimizing ad copy.

On balance, responsive ads are proving to be a promising innovation and a useful tool for many campaigns.

The bottom line on responsive ad technology

Responsive display ads quickly became a success because they require fewer images and aspect ratios during setup. They are quicker to setup, and less of a variation versus traditional display ads. They also eliminate the requirement to manually populate display ads with many different sizes of the same image.  So ad setup time for display is decreased, a major benefit. And less ad copy is required.

Responsive search ads require more time to configure and can be used to enhance the performance of some campaigns, but they are more controversial. The key to getting the most out of responsive search ads is to first, put the effort in to populate them with one a mix of known strong search ad copy, to two parts new untested ad copy. The responsive search ad can be thought of as an “ad copy combination split tester”.  It will go to work discovering the best combinations of ad copy to serve, and the sequence to serve it.  This serves to extend the reach and conversion volume of each ad group within a search campaign.

This new type of ad will undoubtedly see continued improvements. We’ll have follow-on articles as they continue to evolve.