Any digital marketer worth their salt knows that testing, measuring and iterating are the three pillars of a successful strategy.
Google Shopping, however, has proved somewhat resistant to this strategy. Unlike other methods of PPC, it’s technically not possible to properly split test alternative titles by showing two different versions of the same ad simultaneously to different people.
The obvious workaround is to compare two different time periods, and while that does come with some challenges, it’s still possible to draw significant conclusions. To help ensure the different time periods aren’t the reason for any uplift/downswing, we also analyze the total performance of all the products over the whole time period.
In their simplest form, Shopping ads are made up of a Product title, description, category, and image. We’ve had significant success in testing the first three over the years – you can read about those findings here – but image true testing has remained somewhat elusive.
In this post, we’ll discuss the current challenges associated with image testing, what testing we’ve done thus far and the insights we’ve derived.
Image testing challenges
An image can tell you far more about a product than a title or a description and a good one is key to catching the attention of would-be customers. Images are widely believed to be the most important factor in shoppers choosing to click on an ad.
Image testing looks to be the next game-changer for click-through and conversion rates, but it’s still a bit of a black box. This is largely due to three factors that make images difficult to test: categorization, image displays, and time lag.
The first issue comes down to how you categorize your images, to begin with. In fashion, for example, a basic image test would be to see whether images with a model perform better than images with just the product in them. However, if you haven’t already categorized your images in this way already, it will be difficult to know which images to change.
Ideally, you’d change all the images in an entire category. But this also means you need to have an image of both types for every product in that category. Which you may not have, especially if you sell products from other brands.
Image categorization is time-consuming and largely manual. However, without accurate categorization, you won’t be able to run a conclusive test.
Conclusion: Spend time categorizing your images along all the lines you think you’d want to test them
While image categorization is something completely within your control, whether or not your image is even shown by Google is not.
On the main SERP (Search Engine Results Page) each of the (up to 8) products shown has an image. However, in the Shopping tab products get clustered into just one offer list. This usually happens when multiple retailers sell the same product.
Effectively, what this means is that you cannot be sure your image is being shown – making accurate testing very difficult.
Conclusion: Unfortunately there’s no real workaround for this as you don’t have any control over which image is shown and where. For more general tests you could test only exclusive products not sold by anyone else, in which case you can be sure that your image is displayed
Another factor out of your control which makes image testing difficult is the time lag. In this instance, time lag refers to the amount of time after you change the image on your website for Google to change the image on your Shopping ad.
It can take up to 72 hours for Google to index a new image when the accompanying URL is changed. If you don’t change the URL and instead do a server-side image swap, re-indexing the image can take up to 6 weeks!
This time lag makes image testing take a while because, in order to accurately measure the effect, you need to wait for all the images to be indexed and then run your test, change them, wait for the re-indexing and run the test again. The whole process could potentially take months.
Conclusion: Make sure your images have changed before you start collecting data for your test – wait at least 72 hours.
Image testing and insights
Despite the challenges accurate image testing presented, we were able to run a test that gave us an interesting insight into an image’s potential to boost a campaign.
The data we used came from two retailers in the fashion industry. For this first iteration, we looked at the effects of using images that featured a modeled product and a product in isolation.
In total, we changed around 1800 images. Up to 40 million impressions were taken into account, and we overlooked any products that were not consistently available. To help measure the true impact of the new image, we also ran a control stream as a baseline.
Interestingly, we found that while in some cases changing the image had a significant impact on the CTR, in other cases it had very little effect.
When we dug a little deeper we found that whether or not an image change had a significant effect depended on the other images it was displayed beside in Shopping. If the image types in Google Shopping varied regularly (ie there was a roughly equal amount of images with and without a model), there was no significant benefit in having the image feature a model.
On the other hand, if most of the other Shopping images featured products on their own, it was beneficial to have an image that stood out (ie one with a model). We observed a 27% increase in CTR when our image stood out.
This presents another major issue with image optimization: knowing what sort of image will stand out. Figuring out what the most commonly displayed images look like for the top 1,000 products you sell presents plenty of its own challenges.
So close, and yet, so far
As Google gets better and better at figuring out synonyms and directing traffic to the right products, optimizing your product titles will get more and more difficult. Images have the potential to fill that gap.
The potential for image testing goes far beyond whether or not the image contains a model. Perhaps certain product colors stand out more than others. Or image background (white, color, scene) could be important.
For now, it appears that simply having an image that stands out from the competition in some way has the greatest effect. But as we pointed out before, even that isn’t an easy thing to figure out and optimize for.
There’s a long way to go before we get the full picture of what an image does to your campaign. And, honestly, it seems like we may need a few changes from Google before true testing and optimization are possible.
Nevertheless, we’ll keep plugging away at it to see what we can find out. We’ll keep you posted!
What do you think of image testing?