Exploring URL Volatility in Google’s AI Overviews

Google’s AI Overviews (AIOs) are clearly here and are clearly here to stay. While what they eventually become and how they evolve are still unknown we can identify how often the AIOs appear, for what keywords, etc.

Still, even with this data, there was something still nagging me about them and their importance to search marketers. Understanding AIOs and their tendencies in how and when they appear is vital but doesn’t fully indicate the importance of an AIO from an SEO perspective. 

Which is why I want to understand:

How consistent are URL placements within the AIO? 

How consistently is Google showing the same URL within the same AIO for the same keyword? If URL placement is incredibly inconsistent what does that mean for websites?

Should much effort still be placed on targeting AIOs the same way it feels like it is right now? If URL placement is incredibly volatile what would that mean for the value of appearing in an AIO?

Simply, without understanding URL consistency in the AIOs it’s hard to understand just how important securing your place within them is for websites, SEOs, and marketing teams of all types. 

Let’s then have a look at URL consistency in the AIOs. 

Key Stats

For keywords where the AI Overview appeared for at least 20 out 31 days:

• 0 keywords saw 100% URL consistency 
• 96% of the URL changes were also domain changes 
• 91% of URLs studied were removed from the AIOs at some point
• Just 43% of removed URLs returned to the AIO later in the month

Methodology

We studied a sample of 1,500 keywords from a US mobile database and 1,500 keywords from a US desktop database. We chose only English keywords that triggered an AIO at least 20 out of the 31 days of October 2024. Keywords were chosen randomly without any filter on length, volume, intent etc. 

On average, we saw about 5 unique URLs present in an AIO for desktop and 4 unique URLs for mobile. As for domains, we saw about 5 unique domains present in an AIO for desktop and 3 domains for mobile.

Is Google Changing URLs Inside AI Overviews? 

The short answer is, yes. 

As mentioned in the methodology section earlier, we looked at 1,500 keywords that produced an AIO for at least 20 out of the 31 days analyzed. Of those keywords, Google changed a URL within its corresponding AIO at least once in all of them. 

Meaning, 0% of the keywords we analyzed retained the same URLs throughout. 

There was at least one URL change throughout the data collection period for all 1,500 keywords. And that is for both mobile and desktop.

That 0% is an important point. We’re talking here about the most consistently appearing AI Overviews. It would be one thing if Google was swapping URLs within an AIO that appeared for a keyword infrequently. However, here, the appearance of the AIOs for the keywords analyzed is quite consistent.

Meaning, Google is changing the URLs within the AIOs that matter to SEOs the most. 

The question is how often is it swapping those URLs? 

URL Changes Inside of AI Overviews

Understanding that Google is swapping URLs in what I will term “consistent AI Overviews” (i.e., for keywords that produced an AI Overview for at least 20 of the 31 days tracked) is one thing. Understanding how consistently Google is doing so is another. 

Looking at the total number of days a keyword produced an AIO within the data period we saw that Google made URL changes on 55.99% of those days (54.77% on mobile). 

Think of that like, there is URL stability within the AIOs on just 4 of every 10 days. Not very stable.

But what is the average number of URL changes Google made throughout the reporting period?

On average, we tracked that Google changes the URLs within said AIOs 12.32 times on desktop and 12.49 times on mobile. 

That’s pretty volatile when you think about it. The maximum number of days the AI Overview could appear here was 31 days and the minimum 20.

12 URL changes over the course of 20 days? That is a lot. Of course, it’s less noticeable if the AIO appeared for 31 out of 31 days. However, the average number of appearances of an AIO for a particular keyword within the data set was 22 days on desktop and 23 on mobile.

It also all depends on how many URLs are within the AIO you are trying to capture. 

If there are 100 URLs present (there aren’t) then 12 of them changing over the course of time might not be a massive concern. 

There were, however, on average just 5 unique URLs on desktop (4 on mobile) within the AIOs that appeared within our data set (to remind you, these are the AIOs that were on the MORE consistent side having appeared for at least 20 out of the 31 days we tracked). 

In the end, we’re talking about an average of 22 AIO instances for a keyword, with an average of 5 URLs per instance (total of 110 URL appearances on average), and 12 swappings of the URLs, all in the span of one month.

Comparing what we saw in the beginning of the month to the end, only 6.7% of the desktop URLs (4.7% on mobile) seen during the first week of October were still there during the final week. 

That seems pretty volatile but maybe Google is only changing one URL each time? 12 URL changes out of 110 average URL instances isn’t terrible at all. 

The question is how often is that actually happening? 

Frequency of URL Replacement Inside of AI Overviews

Is Google swapping URLs within a given AIO every day, every two days? 

For these “consistent” AIOs Google keeps the same URL within them an average of just 3.87 consecutive days on desktop and 3.33 days on mobile.

That’s incredibly frequent. Google was constantly making changes to the URLs within these AIOs. 

To complicate matters even further, 91% of all URLs inside of the AI Overviews studied were removed at some point.

Almost all of the URLs we saw inside these AIOs were at some point replaced with another URL. 

That would seemingly mean that there is essentially no stability nor URL consistency within the AIOs studied. 

However, we still need to take into account the URL’s tendency to make a reappearance. Meaning, is Google returning the URLs it removed from the AIOs at some point? 

If so, then having your URL removed from the AIO might not be a big deal. It might be back the next day and remain there. 

Here’s the real kicker, just 43% of these URLs reappeared on desktop within the data period (41% on mobile). 

That’s concerning. Of the 91% of URLs that were removed, only 43% ever came back. 

That means that just over 50% (52% to be specific) of the URLs studied were not “banished” from the AIOs (at least for the reporting period that is). 

So basically what you have is Google swapping out nearly all of the URLs within these AIOs and returning less than half! 

However, perhaps there’s hope. Perhaps Google is swapping one URL from one domain with another URL from the same domain.

Is Google simply swapping one URL from your site with another URL from your site? If the swap doesn’t result in a new domain entirely, then change in URL is less significant, if not entirely insignificant (depending on if the swapped URL serves the site’s purpose in the same way). 

What the data found was that of these 1,500 keywords the corresponding AI Overviews saw a domain change 96% of the time on desktop and 96% on mobile. So it’s not simply a matter of Google swapping URLs from one domain. Google is going with an entirely different domain with some sort of regularity. 

Putting It All Together 

The data points to there being some serious concerns around URL placement within AIOs. With 91% of URLs falling out and just 43% returning, the AIOs we analyzed seem far from consistent. That could call the value of focusing on AIOs as part of your strategy into question. How much time and how many resources do you want to devote to something as inconsistent as the particular AIOs we looked at? 

At the same time, there is also tremendous potential. And a lot of it will come down to how the AIOs that matter most to you behave. To me, this makes tracking URL behavior within the AIOs quite important. Understanding how the AIOs that you are targeting behave in regards to URL placement might pay dividends if you can spot where there is and is not stability.