RSA Optimization Series: Part 1 – RSAs vs ETAs

By Brad

16 comments

Ads, General, PPC Management

In a few months, you will no longer be able to create Expanded Text Ads (ETAs) as Responsive Search Ads (RSAs) will become the default ad type. In getting ready for this transition, we are comparing ETAs and RSAs to better prepare for the upcoming changes.

This comparison will let you see how the performance data in your accounts might change with the transition as well as share strategies on how to best create and manage RSAs. We’ll have some upcoming posts on different ways you can create, manage, and test RSAs so you can hit your account’s goals.

In 2019, we compared RSAs to ETAs and found RSAs often won for CTR while ETAs often won for conversion rates. That was almost 3 years ago, so we expected to see RSAs win more often across the board as Google’s machine learning has had a lot of time to gather data and create a better algorithm.

In this post, we’ll not only look at an overall comparison of ETAs and RSAs, but we’ll also delve in detail into six different performance metrics to assess the severity of wins and losses by each ad type.

How the comparison was conducted

We used data from several thousands of accounts to find ad groups that met the following specific conditions:

  • Ad groups that contained both an ETA and an RSA. The sample size is almost 1 million ad groups.
  • Ad groups that had both ad types running at the same time and for a minimum length of time. Adalysis tracks the activation dates for all ads and this allows us to compare data when ads were active at the exact same time to ensure the integrity of the comparison.
  • All comparison data came only from the ad test results for the above ad groups. The Adalysis ad testing algorithm enforces a minimum impression threshold to ensure low volume ad groups are filtered out and wouldn’t skew the comparison data.
  • We included the comparison data for a metric only when that metric had reached statistical significance, meaning the ad will most likely perform consistently in the future as concluded. Adalysis automatically performs these ad tests on all accounts making it possible for us to analyze data of split tested winners and losers by ad type and metric combination across large sample sizes.

By using the above criteria in selecting the ad groups, we are confident the conclusions reached in this post are as accurate as could possibly be, and the trends will hold across any large enough group of accounts.

Overall comparison results

When we look at the number of times one ad type outperformed the other in six different metrics, we find that ETAs won in every single one of the metrics analyzed.

In addition, we compared the impact of rotating ads evenly in campaigns that use manual CPC bidding versus scenarios where optimized rotation was used. We saw no difference in the data of the two sets as far as how often one ad type outperformed the other.

The above results tell us how often one ad type outperformed the other but don’t tell us by how much. To better understand the severity of the difference in performance between the two ad types, we will delve into each comparison in more detail. You will see charts below, for each metric, exposing a better breakdown of the above data.

How to read the below charts

The way the below charts are plotted are as follows:

  • Each dot represents one ad group. A blue dot represents an ad group where the RSA outperformed the ETA while an orange dot represents an ad group with an outperforming ETA.
  • The horizontal axis is the count of ad groups. Hence, the more space a color occupies horizontally, the more ad groups there are where that ad type performed better.
  • The vertical axis represents how much better the outperforming ad was compared to the other ad type in that ad group e.g. a value of 3 means the winning ad was 3 times better than the losing ad.

CTR comparison – ETAs (51%) vs RSAs (49%)

This was a huge surprise to us. We expected RSAs to at least win by CTR as Google is excellent at optimizing CTR. While ETAs barely won, it was still unexpected.

From the above, we can conclude the following:

  • ETA outperformed RSAs in 51% of the ad groups analyzed.
  • There is a tendency to win by large margins. When either ad type outperformed the other, quite a few of them had a much higher CTR compared to the underperforming ad. It can be seen in the above chart that a multiple of 5 is not that uncommon.
  • The majority of outliers were RSAs meaning that when an ad type won by an overwhelming amount, it was typically an RSA.

Conversion rate comparison – ETAs (67%) vs RSAs (33%)

This result was not surprising as that has long been the main complaint against RSAs i.e. their inability to consistently convert better. However, the percentage of wins of 67% to 33% was surprising.

From the above, we can conclude the following:

  • ETAs outperformed RSAs in 67% of the ad groups analyzed.
  • Outperforming by the large margins we saw in the CTR comparison was not common when it came to conversion rates. However, there are a significant number of ad groups where one ad type has a conversion rate that is 2-3 times better than the other ad type in the ad group.
  • The majority of outliers were ETAs meaning that when an ad type won by an overwhelming amount, it was typically an ETA.

CPA (Cost/conv.) comparison – ETAs (67%) vs RSAs (33%)

 

The CPA data shares a lot of trends with the conversion rate data. From the above, we can conclude the following:

  • ETAs outperformed RSAs in 67% of the ad groups analyzed.
  • Comparing the CPA of the outperforming ad to the underperforming one had similar margins to what we saw in the conversion rate data.
  • The majority of outliers were ETAs meaning that when an ad type won by an overwhelming amount, it was typically an ETA.

ROAS (Return on ad spend) comparison – ETAs (69%) vs RSAs (31%)

ROAS was one of the oddest metrics to work with as there were so many outliers and massive differences in ROAS by ad type. If the outlier was above 100, we lowered it to 100 to make the data easier to read as some of the outliers hit almost 2000, meaning one of the ad types was over 2000 times better than the other ad type.

From the above, we can conclude the following:

  • ETAs outperformed RSAs in 69% of the ad groups analyzed.
  • Relative to the other comparisons above, it was much more common for the ROAS of the better performing ad to be magnitudes better. Multiples of 10+ were not that uncommon.
  • The majority of outliers were ETAs meaning that when an ad type won by an overwhelming amount, it was typically an ETA.

We also analyzed the data using a number of different filters e.g. include only ad groups where there were at least 10 conversions per ad, or at least 500 in conversion value, etc. Irrespective of how we sliced the data, the overall trends of huge differences and massive wins in ROAS remained.

We are confident that the reason ETAs won so overwhelming in ROAS is due to their higher conversion rates.

Conv./impression comparison – ETAs (55%) vs RSAs (45%)

Conversion per impression (CPI) looks at the ratio between how often an ad is displayed (impression) and how often it receives a conversion. CPI is one of our favorite metrics since it combines both the frequency of clicks (CTR) and how often a click converts (Conversion Rate).

From the above, we can conclude the following:

  • ETAs outperformed RSAs in 55% of the ad groups analyzed.
  • There are a significant number of ad groups where one ad type has a CPI that is 2-3 times better than the other ad type in the ad group.
  • The outliers were almost equally split between ETAs and RSAs.

Conv. value/impression comparison – ETAs (63%) vs RSAs (37%)

Conversion value/impression is another favorite metric as it looks at how much revenue an ad brought compared to how often it was displayed. This metric takes into account CTR, conversion rate, and average checkout amount into a single metric. It’s a favorite among e-commerce sites and those that track and bid based upon revenue.

As with ROAS, there were a number of outliers in the 100-250 range, so we capped them all at 100. The outliers were not nearly as extreme as with ROAS, however, a lot of the same types of conclusions can be made as with CPI.

From the above, we can conclude the following:

  • ETAs outperformed RSAs in 63% of the ad groups analyzed.
  • Similar to the ROAS comparison, it was very common for the Conv. value/impression of the better performing ad to be magnitudes better. Multiples of 10+ were not that uncommon.
  • The outliers were almost equally split between ETAs and RSAs.

Overall, ETAs brought in more revenue per impression than RSAs. Much of that is due to the conversion rate differences. We would not want to make this a firm statement, but what we often see is that ETAs are written in specific ways to try and incentivize users to both convert and to have higher checkout amounts through offers like free shipping at certain amounts, discount codes to increase shopping cart basket size, etc.

With ETAs, you know how these ads will be rendered and can make a direct connection from the ad impression to the products. With RSAs, that same control does not always exist. There are still ways of creating RSAs that can make them behave similarly to ETAs which will be covered in an upcoming post.

Wrap-Up

Overall, the win percentages of ETAs are higher than RSAs across all metrics. We feel this is partly due to the RSA format itself which requires more effort from the search marketer to create and manage in more strategic ways than they are used to with other ad types.

Based upon the above data, this is how we recommend most accounts should approach the ETA to RSA transition:

  • Avoid replacing large numbers of ETAs with RSAs within a short period of time. A sudden replacement of multiple ETAs would most likely cause a large drop in conversions.
  • We’d suggest you start adding RSAs to your account and testing them against ETAs. As Google is going to leave ETAs active for a while (after the deadline), we’d suggest leaving ETAs running in your account until the RSAs can consistently beat them. You can keep tweaking the RSAs by changing the pinning, themes, and ad lines until the RSAs at least match or outperform your ETAs.

The transition period right now, where you can still create ETAs, is a good time to gain insights from all your ad types so that you are ready for the day ETAs will be sunsetted.

To help you create better performing RSAs, we’ll go into more detail in future posts about what you need to know during this transition. Keep an eye out for the next parts of this series in which we will cover:

  • How the RSA performance correlates to its ad strength and pinning of its headlines and descriptions.
  • Best practices to follow when creating RSAs.
  • New tools from Adalysis to help our customers manage the transition to RSAs at scale.

Related Post

16 Comments

  • Alex

    Hi Brad,

    Thanks for the analysis and insights it provides. Google says RSA’s participate in more auctions than ETA’s. Therefore a lower CTR, CVR etc. could be due to the fact that you are participating in lower quality auctions (less relevant). Still by participating in more auctions you can get more conversions and conversionvalue at a similar CPA or ROAS. Whats your opinion about this?

    Kind regards,

    Alex

    Reply
    • Brad

      I think the ‘RSAs cause you to get more impressions’ info has been misrepresented and misinterpreted a lot.

      If you add an RSA to an ad group, that ad group does not suddenly get 3.5 more impressions due to a new ad. Many auctions are pretty competitive already, and Google is not taking impressions from one person to give it to someone just because they added an RSA. People are not suddenly searching more and causing your impressions to go up because you added an RSA. There’s a very different answer to how & why RSAs often have more impressions.

      If you are using any automated bidding or optimized ad rotation, then Google has control over the ad serving, which they like. They also like serving RSAs. Therefore, RSAs often get more impressions than ETAs.

      If you are using phrase and exact match, the ad group’s impressions do not change due to the ad types (ignoring quality score for now, which assumes you have well built ads). You can add an RSA and then pause it, and the impressions for the ad group remain static (of course, bids, budgets, quality score, etc might still have them change, but it was due to factors beyond the active ad ty pe). So the RSA is not causing you to get more impressions, Google is just serving the less effective RSA ad more often than the ETA ad.

      If you are using broad match, we have sometimes seen the ad group’s impressions go up. These impressions sometimes are for keywords that are in other ad groups, and this can hurt conversion rates since an incorrect ad is showing for that keyword.

      In other cases, these impressions are new as Google matched you to new search terms. As with broad match, these additional search terms rarely convert, but there are exceptions.

      If you are adding RSAs and getting a lot of new account impressions and conversions, what that really means is the account missed a lot of good keywords in its setup process and should look at those new search terms and create appropriate ad groups as well as do some more keyword research.

      Overall, there is no correlation between RSAs getting more impressions and the account getting more conversions. RSAs are just served more often than ETAs regardless of stats.

      For instance, in these two ad tests, RSAs are not the winner by any metric. In the one case, it is also not the loser.

      However, the RSAs were served much more often than the ETAs. In these examples when the RSAs were paused, these ad groups ended up with significantly more clicks and conversions since the more effective ads were now served more often.

      RSAs do sometimes win, and they do sometimes add new impressions when using broad match. There are even some accounts where RSAs are doing quite well. We’re looking at the overall trends, but there are many exceptions when you get inside specific accounts and ad tests.

      Reply
  • Andrew G

    Hi Brad,

    On a much lower sample size, my impression of the trend is similar to yours.

    A thought: one reason ETA’s perform well, in well-managed accounts (some of the time), is that they’ve been tested and honed by fire over the years. RSA’s, as you mention, “take more effort.” Sometimes, they do better. But oftentimes all they wind up doing is diluting your existing, well-honed ad elements with a jumble of ideas and triggers that aren’t as compelling. Google doesn’t mind: for the purposes of learning and training the machines, “throw anything and everything at the wall” (especially if it doesn’t harm CTR) is fine. But individual firms forgoing profit for a few quarters or even years, in the spirit of excessive experimentation… I think we can see how that’s problematic.

    Reply
    • Brad

      Hi Andrew 🙂

      I totally agree with you. Often when accounts are creating ETAs they are thinking about how the ad fits together. What line is going to make my ad seem relevant? What offer or benefit should I pair this with? Etc.

      With RSAs, many people are throwing everything into the ad and letting Google try and figure it out. Google algo doesn’t seem to understand that relevance & offer combinations trump almost every other combination out there.

      There are exceptions where the ETAs aren’t that relevant and RSAs do well. There are RSAs where each line was thought about well, some light pinning was done, and the end result is a great ad.

      I personally think a lot of the RSA issues stem from how Google positioned them. Toss everything together, the more the better, and we’ll figure it out. With that positioning, we’ve seen a lot of RSAs that will rarely do well as it’s the same RSA in every ad group, or the ad group has a large mixture of keywords as the person thinks Google will figure out the keyword and headline combinations so the organization gets sloppy, etc.

      One of the top RSA issues is just the creation itself and how they have been positioned by Google and used by marketers.

      Reply
  • Odi Caspi

    Hi Brad

    Thank you for sharing this detailed comparison. I think it proves that a lot of advertisers have been seeing in their own accounts.
    I have a question. If CPAs are higher and ROAS is lower, does that mean that Google has just increased the cost of advertising on search globally? Or will the market balance it out?
    With no real competition in their field, I wonder how much incentive Google has to give us a better product that will bring client costs down.
    Thanks
    Odi

    Reply
    • Brad

      CPAs are often higher for RSAs over ETAs, and the same with ROAS. That doesn’t mean that accounts are getting lower returns as the competition level doesn’t change much due to the ad format (some slight exceptions with broad match and RSAs). If CPAs rise to a point that they are unprofitable or most companies, they will lower their bids. If they can’t get enough profitable traffic on Google, they will move budgets to ad roll, tradedesk, Microsoft, facebook, reddit, LinkedIn, Pintrist, etc. So Google does have an incentive to keep budgets on Google and have happy advertisers.

      Reply
  • Ian

    Awesome analysis Brad – thanks! Your conclusions align well with a study we did in December focused on the Rx Pharma Brand Manufacturer vertical, where we simply looked at # of ad headlines served by brand on specific keywords over a ten month period on Google and compared that to each respective brand’s average ad position on the same keywords. # of headlines was used as a proxy to help marketers understand if more modular assets (which RSAs enable) translates to better performance at least in terms of ad position. Limited insight, but adds additional context to the ETA>RSA migration conversation.

    Essentially, we saw that more headlines tested translates to higher ad position. (Yes, this could be due to multiple other factors like bidding and ad copy relevance, but the trend is pretty clear.

    I wonder if your dataset can be segmented to reveal whether ad groups containing RSAs that tested more headlines than the ETA ad groups performed better?

    Reply
    • Brad

      We didn’t segment data that way. However, looking over the data, that does not appear to be a trend. There were plenty of accounts where the RSAs have 10-15 headlines and there were only 1-2 ETAs in them. Even in these ad groups, ETAs held the trend of performing better than RSAs.

      Within any one account or ad group, the results did differ. There are plenty of instances of RSAs doing well. We’re looking at the overall trends.

      We’ll later get into more articles about ways of creating and testing RSAs, and then do some more analysis on how pinned headlines perform.

      Reply
  • John Cammidge

    Fantastic, in-depth article, Brad. We plan on creating RSA adverts across all accounts and running them alongside existing ETA’s.

    However, we will be pinning headlines and descriptions so they mimic ETA’s. After discussing this with my team we have decided to test pinning multiple headline 1’s, and maybe headlines 2’s, providing they are readable from a customer intent point of view. Do you think this is a sensible approach to transitioning to RSA’a?

    Reply
    • Brad

      Yes. I think there are ways to properly pin lines to use a combination of your creativity and machine learning.

      We’re working on that article right now 🙂 and should have it published in the next week or so. Stay tuned 🙂

      Reply
  • Al Mullowney

    Hi Brad, great article. Is one RSA per ad group still the recommendation?

    Reply
    • Brad

      Hi Al,

      If you are testing RSAs, it’s ok to have multiple RSAs per ad group. For the short term, 2 ETAs and 1 RSA are fine. For the long term, you’ll eventually pause your ETAs and run 1-3 RSAs within an ad group.

      We’re going to look at testing and RSA creation in our next article.

      Reply
  • Hans

    Hi Brad!
    Very interesting, thank you. In his excellent book “Join or Die” (publish. 2020) Patrick Gilbert claims that some auctions is reserved for just RSA:s. Can you confirm this?

    Reply
    • Brad

      Hi Hans,

      On occasion, search or display partners serving native inventory reserve spaces for responsive ads. They are usually responsive display ads, but they can be responsive search. These spots are more likely to be across the display network where the ads need an image and text to look like editorial listings so the ads blend in with the content. This is why Google likes you to create responsive display ads over image ads.

      I have yet to see an occasion where adding responsive search ads increased our ad group’s impressions by any measurable amount. However, there could be some exceptions across search partners or in various countries that I’ve not come across. So it’s possible, but it’s probably pretty rare.

      Reply
  • Jos

    Thanks for researching and sharing your insights. Have you made any split up between adgroups using broad match keywords vs adgroups using only exact match or phrase match keywords? In other words, when slicing the data for the six metrics you described, is it possible that RSA’s still beat or equal ETA’s when only looking at broad match ad groups, or at least the difference in percentage is smaller?

    Reply
    • Brad

      No, we haven’t done an RSA to match type analysis yet. I’ll keep that in mind for future data ideas 🙂

      Reply
  • Leave a comment

    How well is your account performing?

    Find out, in minutes, what your account score is with our FREE audit report.

    Get my account score