Twitter Carousels & Google Search: An Experiment (Real Data), How Measurement Works & More

By Brodie Clark

May 28, 2021
twitter carousel google experiment and data

The presence of a Twitter carousel within Google Search presents an interesting opportunity for publishers. Under the right parameters, content can gain high visibility in Google’s SERPs, but often goes under the radar.

The discussion around the influence of website URLs appearing in Twitter carousels dates back to at least 2018, but I still see confusion around how this SERP feature can influence our data.

With the help of 100+ SEO professionals, I was able to create an experiment to build clarity around this situation. To easily share data around the test, I put my personal Twitter account and website to work.

This post includes an outline of the experiment I ran, what data looked like within Google Analytics and Google Search Console (screenshots included), along with an explanation of the findings with key takeaways.

As a start, lets first take a look at the experiment setup that I had in place.

The Experiment Outline

For the experiment, I needed a Twitter account that had the ability to yield a tweet carousel on Google, a website URL, along with having Google Analytics installed and Google Search Console verified for the website.

To try help create some awareness around the experiment, I tweeted about it a couple of days beforehand to let SEOs know that I’d appreciate their involvement. This then followed with the creation of the experiment itself.

For the experiment URL, I created a blank page on my site with some text and a GIF which can be found here: https://brodieclark.com/twitter-test/. As mentioned on the page, I had added a noindex meta tag to prevent it from surfacing on Google.

By having the noindex meta tag in place, it meant that any data within Google Search Console for this page would be solely from the Twitter carousel. The would allow for clean attribution within my tracking tools.

The next step after the ‘test subject‘ tweet was to attempt to send traffic to my twitter carousel. I kept the instructions simple and made the search term clickable, and gave the option to search a related query (which a handful of folks did do).

Here’s what the tweet looked like, which generated ~40K impressions and resulted in ~400 clicks on the search term URL that was added toward the top of the tweet:

It became a bit tricky constructing this tweet, because some Twitter inception needed to take place to make the instructions as clear as possible. Step 3 in the above has a screenshot of a tweet that was actually Photoshopped.

This was the basis for the experiment. All quite simple, but I tried to be meticulous with the instructions (i.e. ensuring the screenshot reflected what the search results looked like) to ensure there was as much activity as possible.

Now lets get stuck into some of the data that was collected within Google Analytics and Google Search Console, and how it was presented in each tool from clicking the URL from Google’s search results.

Data Collected (Google Analytics)

When the tweet was first sent out, I had a look at my realtime traffic coming through Google Analytics and filtered for my experiment page. Thankfully, the data was coming through as I hoped (categorised as ‘organic’).

I managed to get a screenshot after publishing the tweet, showing 7 active users on the site located in various parts of the world. Here’s what this looked like at the time:

By the end of the testing period (2 days), the page had received 199 ‘sessions’, from various data sources. This included ‘google’ traffic (what we’re most interested in seeing), along with traffic from Twitter when the ‘test subject’ tweet was published, and also some direct traffic.

From the outside, this traffic is indistinguishable from other clicks that have come through from Google’s search results. If my test URL was indexed immediately by Google and started to rank once published, this data would have been mixed in with the tweet carousel clicks, with the separation being unclear.

That’s where Google Search Console comes into play. Although we don’t have specific filters in place for a lot of SERP features (like Twitter carousels), we can at least use the data collected as hints for where the traffic may have come from.

Data Collected (Google Search Console)

Google Search Console is the dataset that I’m most interested in for this experiment. Because GSC measures impressions (unlike GA), this is the primary metric that can skew data and can become confusing.

At the time of writing this post, there was a total of 103 clicks and 367 impressions recorded in GSC for my page that appeared in the Twitter carousel (see the feature image for this post). Because of the data lag, the total amount came in at slightly higher than this.

Within GSC, the data that I’m most interested in relates to the location and device used. Aside from the query, these are the two tabs that you should be digging into when seeing an uptick in impressions that don’t make sense.

For my page, here’s what the county data looked like in GSC for the top 5 locations (shoutout to the SEO community in India for being the top referrer for my test!):

And also data for the device used:

The data being captured within Google Search Console was fascinating to see. Because of the influence of location and device type, results can often appear in a different positions for users on Google.

For instance, the average position for the tweet carousel was quite high in the US at 1.2, whereas more broadly on mobile the average position was lower overall at 3.5. A number of factors will come into play for why this is happening. And the carousel might not show for all users, such as for Mikael Araújo who is located in Brazil.

Another aspect that was interesting to see, which I had confirmation of in the comments of my instructions, was that some mobile devices allowed you to directly click the website URL from Google’s search results (also confirmed in the data) but other mobile devices didn’t – such as for my iPhone 11 Pro.

Aside from the scenario that I created for this experiment, there can also be instances where your website URLs could surface for generic queries. One example I came across recently was shared by Mordy Oberstein.

The data within GSC showed a URL with a spike in impressions, even though it hadn’t been indexed by Google yet (common for newly published content on most sites). The impression spike was for the query “seo”, a highly competitive term.

Entering my time machine with the SEMrush SERP screenshots tool, I could see that Google was displaying a tweet carousel on the day that the spike happened, which also matched up well with the country/device/position data.

Similar to the experiment I ran, the data for this example using the “seo” query was a clean dataset, in that the URL hadn’t appeared on Google yet, but not on purpose with a noindex meta tag like I was using.

From what I could see, it looks like either a tweet from Mordy’s personal Twitter account or the SEORant account was what surfaced in the Tweet carousel on Google for “seo”, for a short period of time judging by the estimated search volume for the query and the impressions that the page received.

An interesting aspect of tweet carousels is how they are measured within GSC. The entire carousel takes the spot of a single position on Google, even if there’s three tweets appearing at the same time within the unit in search results.

Glenn Gabe did a deep dive on similar situations in a post from 2019 which I’d recommendation checking out. The same goes for another post specifically related to Image Packs and images appearing in Knowledge Panels – which I come across regularly, and I suspect Google will look to address in the future.

The tweet carousel is another that can be added on top of the information shared in Glenn’s posts that are useful for SEO professionals to be aware of. When there’s a sudden spike in impressions, dig into the data, use the SEMrush SERP screenshots tool, then try to match up with some of the SERP features mentioned in this post.

Key Takeaways

According to Sensor, tweet carousels appear on ~6% of all Google SERPs, with similar data being visible across mobile and desktop search results. Here’s what this currently looks like for Google’s desktop search results based on the SEMrush dataset:

When triggered, it can mean that website URLs could be instantly catapulted in front of searchers for competitive queries. Based on the ~6% of Google’s search results that have the feature, the chances of the Twitter carousel being the reason is reasonably low in comparison to Knowledge Panels (which often have images included) and take up ~27% of all search results.

For the experiment that I ran, I show how data is collected in Google Analytics and Google Search Console when a link is clicked or scrolled into the view of searchers. This is for a “branded query”, but the same can be seen in the example I shared for a non-branded query.

Here’s some of the key points covered in my post:

  • When a URL appears within a tweet carousel, the data within Google Search Console can become confusing, causing a spike in impressions in most cases (with few or no clicks at all).
  • My experiment showed a clean dataset for how the data is collected in GSC. Showing how the average position can be influenced by device type and location – two important clues.
  • Tweet carousels can appear for both branded and non-branded queries. For the non-branded queries that have high search volume, visibility can be short-lived and difficult to catch in action.
  • Don’t forget to use the SEMrush SERP screenshots tool when investigating similar situations. Pairing your GSC data up with the tool can help build clarity around sudden spikes in impressions.

Another topic of interest with Twitter and SEOs is the influence of tweeting a URL that speeds up indexing of content on a website. Personally, I believe time should be invested elsewhere (improve site-wide content quality, review sitemaps, links etc.), but most folk still think this is worth trying.

Overall, this was a fun and easy experiment to run. I wanted to put this write-up together because I couldn’t find anything published on the topic. Hopefully, this saves some time when investigating unusual activity within GSC for a page. And who knows, you may discover that your tweet has reached a new audience in Google Search.

Recent Posts