Does Google use click data to rank pages?
Yesterday week I was watching a Hangout with (among others) Andrey Lipattsev, Search Quality Senior Strategist at Google. At one point Rand, Eric and Ammon were talking to him about some experiences they'd tested with click data, and asked if indeed Google was using it in some form.
Long story short, Andrey couldn't really confirm nor deny the specific instances the boys mentioned, without more information. He did however imply that in some instances (shorter temporal ones) there could indeed be some ranking benefits from said signals.
That of course is interesting as Googlers have generally distanced themselves from that as being too noisy and spam-able. I myself have had that reply in the past when talking with former web spam team head, Matt Cutts.
And as recently as last summer, Google's Gary Illyes said that they use them for “evaluation and for experimentation” not for ranking.
Implicit User Feedback
Regardless of what you want to believe, I felt it worth getting into, if only to add to your education as a search geek. These kinds of signals are what is known as implicit user feedback. To get a sense of what types of signals can fall under that umbrella, consider;
- Query history (search history)
- SERP interaction (clicks, query revisions, selections and bounce rates)
- User document behaviour (time on page/site, scrolling behaviour);
- Surfing habits (frequency and time of day)
- Interactions with advertising
- Demographic and geographic
- Data from different application (application focus — IM, email, reader);
- Closing a window.
This is something that I last wrote about back in 2011 in a post on personalization and user behaviour. At that time I did muse that it “might” make some sense in a personalized environment, not the core search setting.
Patents of interest
Next, let's look at some of the patents on the topic. As recently as this past January (2016) Google was awarded a patent on implicit user feedback with;
Modifying search result ranking based on implicit user feedback (filed in 2013).
Which was a re-filing of the same patent that was awarded back in 2014 as well (filed in 2006). It is important to note that this approach was kicking around Google a decade ago. This is not exactly fresh.
This one involves;
“... the first number corresponding to longer views of the document result, and the second number corresponding to at least shorter views of the document result; and outputting the measure of relevance to a ranking engine for ranking of search results, including the document result, for a new search corresponding to the search query.”
In short... bounce rates (often referred to as pogo-sticking). The reason that this type of click-data can be noisy is that given the context of a query, a shorter view isn't always a bad thing. Consider price comparisons etc.
One of the authors of that one, was also involved in;
And again, not new as it was filed in 2007, though awarded in 2015. Once more, not exactly a fresh idea, which we can infer could be outdated and merely IP protectionism. In this case, they're looking at click-data in context of selections (click bias) and selection history (bias) to refine rankings of the documents returned for a query.
Another one, filed in 2007, awarded in 2015 (covered by Bill Slawski here) is;
Modifying search result ranking based on a temporal element of user feedback
Again, more click data... read Bill's post for more on that one.
Ultimately, I can show you another 1/2 dozen or so patents relating to implicit user feedback. And yes, there has certainly been some interest over the years, but not a lot of them were actually filed over the last 5 years or so. Take that for what you will.
At the end of the day
I really wouldn't start to use clicks as an effective long-term SEO strategy. Even the testing the boys had done only had a short term benefit. It could be that implicit user feedback only affects temporal spaces that are prone to the QDF.
As always, we don't really know.
We have to consider the value. Does the noisy potential and ability to spam these signals outweigh the benefit to Google? It's possible. Given the long history of detecting various forms of click-spam, it seems to me the noise factor would be the more important one. Is the signal clean enough to offer enough benefit?
We can also consider that it could be used in limited query spaces, temporal natures or even personalized settings.
At the very least it seems the boys are on to “something” as intimated by Andrey in the Hangout in question. Exactly what though, is still unknown.