A Case Study in Google’s Use of Visit & Click Data

There’s still a weirdly large percent of the SEO and web marketing worlds that, despite overwhelming evidence, don’t believe that Google is collecting or using visit, engagement, click-through or clickstream data. This week, we stumbled across a superb example of the search giant doing just that, so I had to share.


Something funny happens when you Google from the Mozplex

First off, it’s important to know that at the Mozplex in Seattle (our offices on 2nd & Seneca), we’ve got a sizable team of customer success and help professionals who assist Moz’s subscribers over chat, email, and phone calls. ~25 of the 150 Mozzers in Seattle are on this team. Part of that assistance means our help+success folks are often visiting the help content for Moz’s tools and sending those URLs on to our customers and free trialers.

This means that inside the Mozplex, on our wifi and via our IP addresses, a page like Help Getting Started with Keyword Explorer gets a lot more visits per person than it would in the rest of the world.

But does Google actually see that visit data and use it in their search results? Well… You tell me. Here’s the search results from my home (a 30 minute walk from the Mozplex) in Google.com US, incognito browser on a search for “Keyword Explorer”:

I verified with others and everyone outside our offices seems to have exactly the same results for this query. The Keyword Explorer tool itself ranks #1, which is as you’d expect, and then there’s some sitelinks to other relevant pages about the tool from Moz’s blog and the help hub. All good so far. But what happens if we perform this same search from the Seattle Moz offices?

Whoa whoa! That’s not even close to the same thing. The Keyword Explorer tool is gone. The help hub page is the only one visible, and the sitelinks have disappeared. You can run this same query on any machine (desktop, laptop, mobile phone) connected to Moz’s wifi and you’ll see the same thing.

Here’s the Slack thread between one of our engineering managers, Jeremy Modjeska, and our head of SEO, Britney Muller (whose keen detective skills quickly uncovered the cause):

Note that our VPN uses a different connection than our in-office wifi, which is why Britney’s VPN didn’t show the same results.

There are not too many explanations for this anomaly. The only one that reasonably fits the facts, IMO, is: Google’s using some form of clickstream, visit, or engagement data based on the Moz wifi IP connections, and biasing to show the URL that’s uniquely popular from our office. This isn’t unusual on individual devices that are logged-in to Google accounts, but it is the first time I’ve seen it done on a wholesale IP-connection basis for hundreds of connected devices.

There will be those who argue that even if Google is using some form of click/visit data in this case, it doesn’t mean they collect or use that data broadly. Maybe the Mozplex is a totally unique case. Or maybe this is just a weird bug in Google, and not indicative of the data they collect and use in the rankings. Those arguments have been trotted out to explain away the many previous query+click tests I’ve done on stages at events or via Twitter, but I think it’s an increasingly challenging position to take given all the repeatable experiments, the patent applications on engagement data, and the presentations from Google engineers all suggesting this is the case.