There is an excellent post over at Search Engine called “9 Reasons Your Website Can Have a High Bounce Rate
“. There’s a lot more to say about “Bounce rate” which I might launch into in another post, but for now we’ll focus on this one.
The author of the post does not suggest at any point that bounce rate is a factor in search engine rankings. Unfortunately a few commenters do. I say unfortunately because they have probably missed out on all of David Harry’s elaborate and excellent posts about the issue ( see here and here – there’s also a long list of relevant papers for you to read).
“Yes, my friends, search engines continue to evolve. This we do know, but there is little in the way of serious information out there to prove bounce rates are any type of serious ranking signal in the regular index results. My goal is simply to highlight not only some alternate concepts but also impart that we shouldn’t believe everything we read…. This happens too much in the SEO industry..” (David Harry)
To clarify:
I honestly doubt that this is a factor. This is no evidence of this in any of the Google published material, and there has been research carried out on this which concluded that this metric was useless because of it being so noisy. “Noisy” means that it is very imprecise and that this ruins the value of it completely. The research was carried out by extremely senior and well-known researchers. We don’t have rock stars in research but if we did they would be.
“Predicting Bounce Rates in Sponsored Search Advertisements” gives some insight into how it can be useful for website owners, but doesn’t say it can be useful for search engines.
About click-through and personalisation:
“Clickthrough data refers to the links that a user clicks on when presented the result of a web search. This data can be recorded in log files, for instance by using a transparent proxy server which records the clicks, or by asynchronously sending the information about a click to the search engine.” (Zimmerli)
“A number of strategies to extract preference constraints from the observed clicks were proposed and tested. The results were confirmed in a user study that incorporated eye tracking. To get even more information out of the logs, the concept of query chains was successfully established. Using a generated document collection and simulated user behaviour, the presented approach proved robust against noise in the training data and different levels of term ambiguity. And last but not least, the introduction of FairPairs made it possible to collect implicit relevance judgements without presentation bias.” (Zimmerli)
He also mentions that there are issues with this approach such as click-through spam to name but one example.
Susan Dumais (responsible for the “awesome” (for comp. Sci.) but “infamous and useless” (for SEO) LSI algorithm)
If you want to look at how a search engine can use log files and other information rather than that available on web resources, look at those 2 topics in particular. It is far more probable that these would be used or are used more or less. They are not based simply on gathering the data though but rather as a corpus for machine learning.
I won’t elaborate further as there is enough research on this freely available to you and also because I don’t see it as a very important topic for SEO. Beware of misreading research papers as well. When some say that bounce rate was useful they don’t mean for search engines but rather for user preference modelling, and this would usually be but one feature of such a system as Zimmerli illustrates above.
