For many years now SEOs have talked about the various behavioural and query data points from the perspective of being ranking factors with Google or other search engines. You know the ones, behavioural data such as;
- Query history (search history)
- SERP interaction (selections and bounce rates)
- User scrolling behaviour (on selected page);
- User document printing behaviour;
- Adding a document to favourites (bookmarks);
- Data from different application (application focus – IM, Email);
- Surfing habits (frequency and time of day)
- Interactions with advertising
- Demographic and geographic
- Explicit editorial (such as +1 voting)
- Dwell time (time on page)
They are usually quoted by SEO aficionados because, well, they just seem logical.
Sadly our instincts, not matter how well honed, aren’t enough to go by. And testing? I’ve argued many times that this is also fools gold in that we can glean only insights into the nature of the beast that way, nothing definative. Why not you ask? I won’t get into it too much, but just consider the myriad of potential situations, markets and our inability to isolate that which we don’t know, and the viablility of (small scale) tests becomes quite self-evident.
But the question remains; are they worthy of being ranking factors? Is click-through rate a ranking signal?
Not ready for prime time
Ok, first let’s get a little perspective shall we? I have been learning about behavioural data and search engines since my early fascinations in 2006. From 2007 on I started writing about it including the post; Beware: Google is watching you! And What every SEO should know about Personalized Search. At that time I was actually quite bullish on the concepts and had felt they may be playing a part in the search (ranking) algorithms.
Of course, as anyone following along over years would tell ya, I began to doubt this and said as much in the 2009 with (now infamous) The final word on bounce rates as a ranking signal.
Enough with the history lesson, let’s get back to the issue at hand shall we?
Like most behavioural signals, click-through rates are noisy at best in most situations. Let us first consider the obvious; it is easily spammable. Yes, patterns might emerge, but by and large the quelity improvements likely are offset by potential spam issues. Second, we have the whole problem of click bias. That is to say, the habit people have for clicking the first result, second etc (see Understanding search user behaviour for more).
This creates some MASSIVE noise as far as click-through rate metrics are concerned.
Is all lost for CTR and behavioural metrics as ranking factors? Maybe. Maybe not. Last year we had a major infrastructure update from Google dubbed Caffeine. This was an interesting twist because it may mean more processing power which could be used in areas such as this for greater spam detection or a deeper layer of personalization (more on that in a moment). It was a bit of a shining light… a ray of hope if you will.
You see, most testing that has been done with these types of signals showed they can indeed create greater relevance in the results. The glaring issue of course was the spam potential and being able to weave the various metrics into something of value
You can’t simply take one factor (bounce rate, click-through rates) and derive valuable data. One would most certainly need to have them all in play.. which of course means more potential areas to police for spam. Did Caffeine enable that for Google? Hard to say.
A personalized world
The next piece in the puzzle is personalization. What’s interesting here is that one certainly can’t spam themselves, which solves the first problem. Let us remember that personalization was once ‘search history’ and has now evolved (at Google at least) into ‘surfing history‘ as well.
What is still a limitation though is the fact the users are categorized. It is not a pure form of personalization on a user-by-user basis. Meaning, that if you can crack the entity categorizations and spam accordingly, you can indeed spam in a personalized search world. Not ideal at all (for the search engineers)
Another newer element of personalization is, of course, the social graph (which we’re written about a TON here on SNC, most recently with; Google Social Search; seriously, WTF people?). This is yet another layer of personalization that can also help better categorize users and hopefully cut down on the spam issue.
Will personalization make behavioural metrics more valuable in search? Most certainly. I just don’t know if we’re there yet.
Query revisions and recommendation engines
Another element, and general over-sight among SEOs, is that there can be many ‘signals‘ that aren’t used for actual rankings. A few common ones are discovery and indexation. Another, more related to this CTR discussion, is query data.
Have you ever wondered where query refinements, recommendations and suggestions (like Google suggest) come from? Most of them are from query data that they glean from users. It doesn’t play into the rankings of documents per se, but does give insight into how successful/statisfied the user may or may not be with the results. I can see this as one area these signals/metrics can come into play.
So what’s the verdict?
I don’t bloody know ok? Anyone telling you bare-faced, 100% factually about almost anything in the Google black box is a little bonkers.
What we can say at this point though is that it is certainly unlikely that CTR (and other behavioural data points) are being used as ranking signals in any meaningful way in the major search engines at this point.
Should we (as SEOs) ignore it? Certainly not. While the prospect of these playing a larger role is looming, the fact remains that in many cases we’re talking about user-engagement which is ALWAYS a good thing. Great websites, great marketing, great engagement means users are talking about you.
That my friends, is most certainly good SEO.