Google famously says,
“Focus on the user and all else will follow.”
With this in mind, it’s no surprise that the Google Ads auction is set up to reward advertisers for making their ads as relevant and compelling to the user as possible, through the Quality Score system.
Not only would this benefit Google with users trusting their platform to find what they are looking for, but also benefit the advertiser with higher ad rank in competitive auctions, leading to higher intent users clicking on their ads.
In recent years, Google’s machine learning innovations have also been in line with this sentiment. Smart bidding allows advertisers to move away from just setting bids at keyword level, by making real-time adjustments based on user intent predictions. Smart creative ensures that users are served the most relevant and compelling ad in each auction, and smart shopping and display allows for dynamic prospecting of users who are most likely to take action, when they are most likely to do so. These machine learning improvements are made possible due to the abundance of data and exponential processing power that Google Ads have achieved recently
With the changing technology, day to day optimisations are largely taken care of by Google’s automation, with the advertisers focus becoming more on strategic algorithmic-based decision making. As such, advertisers need to be able to adapt in order to gain insights from Google’s machine learning, to improve their overall marketing message.
Follow this guide on how to approach ad copy testing and optimisation driven by machine learning, so that you can serve the right ad to each user in every auction and improve your overall messaging.
A change of approach
When it came to ad copy testing previously, advertisers would A/B test static text ads with equal traffic and refresh the ad copy for the lowest performer based on conversion and engagement metrics.
This required ads to be rotated indefinitely so that each variant received a fair amount of traffic and advertisers would have to wait for significant data to accumulate in order for ad testing to be carried out accurately.
However in 2018, Google introduced responsive search ads (RSA), allowing advertisers to select up to 15 headlines and 4 descriptions, with machine learning selecting the best combination for the user being shown in response to their unique data signals.
To be able to utilise this technology effectively, you were no longer able to rotate indefinitely to test ad copy variants, because the RSA wouldn’t be able to responsively show to the user and instead would be rotated evenly into the mix. Therefore, in the same year, Google introduced ‘Optimise – best performing ads’ ad rotation. This meant that machine learning would show what is deemed to be the most relevant and best performing ad to the user in each auction.
This was great, because it now meant advertisers didn’t have to go through a testing period of potentially showing a weaker ad copy variant an equal amount of times, and account performance improved as a result. However, it now meant that the old way of ad copy testing and optimisation had to change.
With machine learning now powering the ad rotation, the ad that is shown the greater number of times can be considered to be most relevant to a greater number of users. This means that you can gain an idea of the best performing variant a lot sooner than before, and don’t have to risk serving weaker ads during an extensive testing phase.
How to set up your ad groups
Google currently recommends using 2 expanded search ads (ETA) with 1 RSA in each ad group. In this situation, your static ETAs should do most of the heavy lifting in terms of impression share and the RSA should be used as a way of scooping up additional traffic, by increasing your relevance in a greater number of auctions. This leads to a reduction in impression share lost due to rank, and an increased competitiveness in higher intent auctions because your ad is more compelling to the user and quality score is improved, resulting in a higher conversion rate.
How you should set up your RSA
Make use of the RSA’s ad strength indicator to help improve your RSAs as you’re creating them if doing so in the interface. This will give you suggestions of headlines you could include and an indication of how well your RSA has been set up. It is important to remember that there are many different reasons why users would buy your products. For this reason, in order for the RSA to work optimally, you should include many different selling points to appeal to as many users as possible.
Should you pin?
Using headline pins could limit the RSA’s ability to show the best performing combinations.
However, if you need to have an element of control over which headlines appear, e.g. you want to include your brand name in every ad, then pinning can be a good solution whilst still partially maintaining the responsiveness of the RSA. If this is the case, you must pin it to either headline position 1 or 2, and description position 1 to ensure that it is shown in every ad.
You may also wish to test different headlines in specific positions, in which case, pinning multiple headlines to the same position will ensure that these are the only headlines that show there. Again, this could restrict the RSA’s ability to show the best combination.
How you should set up your ETAs
Because your ETAs are supposed to lead the way in terms of impression share, in order to achieve this you should approach them similarly to before, using historical data to inform which variants are most important.
How to set up your keywords
The more data you have to use, the better and quicker machine learning works. In fact, RSAs currently ask for >5000 impressions a month to give a definitive answer of how ‘good’ the assets are.
For this reason, unnecessarily over-segmenting data across multiple ad groups with ‘SKAG-like’ structures is no longer beneficial to the account. Instead, ad groups should be set up with keywords of a common theme that would require the same landing page, with dynamic keyword insertion and the multiple RSA headlines being used to ensure optimal keyword relevance.
Testing and optimising
Because ETAs and RSA play a different role, you should not directly compare them.
You should test your ETAs in an A/B split test like before, but by using Optimise – best performing ad rotation, you will be able to gain an idea of the best performing variant a lot sooner based on the impression share that it receives. This means that for the ETAs in your ad group, you will be able to determine a winning variant even if you are not receiving >5000 impressions a month in your ad group, and will be able to replace this variant to improve your messaging.
Quick tip: Use the best performing RSA assets for the new ETA and replace them in the RSA.
To improve the RSA’s copy, Google has a feature that will tell you which assets are performing best. This can be found on the ‘View assets details’ page.
This feature will compare assets of a similar type and tell you which ones perform the best. You should then replace the low performers, monitor results over time, and if one of the newly added assets is deemed to be better performing, the RSA will tell you.
Quick tip: Combine RSAs with Dynamic Keyword Insertion to improve relevance across multiple keywords and give yourself more space for different headline variants.
What if the ETA with the lowest impression share has a better conversion rate and CTR?
This means that the lower impression share variant is more relevant to users who are higher in intent, but there are less of them.
However, it was considered to be the best performer for fewer searches than the other one, meaning that if it was rotated evenly, the CTR and Conv. rate would likely not be as high as it is for the highest impression share variant.
In this situation, you could try adding the ETA assets to the RSA, so that this combination is still viable to those high intent users. You can then test a new ETA variant against the current highest impression share one to be more relevant to a greater number of users.
What if the RSA has the highest impression share and ETAs are not performing as well?
This can happen when the RSA is considered to be more relevant in a higher proportion of auctions and is not an issue if your audience is fairly diverse.
In this situation, consider using dynamic keyword insertion in your ETAs if you are not doing so already, or swap the ETAs with the highest impression share RSA combinations and replace these assets in the RSA.
How important is campaign data when determining which ad variants are shown? If previous campaign data is used to determine which ads are best performing, will the new variants not be at a disadvantage?
There’s a trend with smart solutions that campaign data is becoming less important for machine learning to make predictions. In this case, Google will show your new variants when they are considered to be more relevant than the current ones, regardless of how well the current ones have performed. Campaign data is just one signal that will be taken into account if it’s relevant to do so.
A lot of accounts we audit are not utilising the machine learning available and there is a lot of scepticism in the community surrounding it, which is largely due to a lack of understanding of the algorithms and the benefits of using them, as well as the slowness of 3rd party optimisation tools being able to adapt to them.
If you’re using Google’s ad rotation algorithms (and you should) to show the best ad to the user in each auction, you can improve the overall relevance of your ads, make them more compelling to the user and improve your overall messaging as a result.