Google is almost certainly using usability signals as a significant factor in the rankings. Google probably doesn't use "bounce rate", at least not as measured by Google Analytics. Instead, Google relies on:
- Click through rate (CTR) - The number of people that click from the SERPs to a site is a good indication of whether the site is relevant for the query or not. When a site gets a worse CTR than it should for the position it is in, its ranking will get worse. When a site gets a higher CTR than other sites would at that position, the ranking will get better.
- Bounce Back Rate (BBR) - The number of people that click the back button from the site back to the SERPs and then hide the site from their results, click another site, or refine their query. Like CTR, Google is likely to make adjustments when the BBR is much better or much worse than expected.
Bounce rate can usually be used as a proxy to measure your BBR, but there are some limitations:
- Bounce rate is measured as the percentage single pageview sessions. Bounce back rate is the number that hit the back button.
- Bounce rate includes people who click on external links on your site (including ads), bounce back rate does not.
- Bounce rate includes people who close the tab or the browser window, bounce back rate does not.
- Some sites provide the full answer that users seek in a single page. Such sites may have high bounce rates, but low bounce back rates.
- Bounce rate can be gamed by dividing articles into multiple pages. That tactic hurts bounce back rate.
Furthermore, as other answers have pointed out, Google's Matt Cutts stated that bounce rate is not used to his knowledge as part of the ranking algorithm. He said nothing about bounce back rate (which is subtly different).
I am convinced that Google uses these signals based on my experience with a site on which I was doing the SEO. It was a type of product site. We noticed that we just couldn't get some products to rank for their targeted keywords, despite pouring massive amounts of internal pagerank into them. One pattern that emerged was that the products that were not ranking had less content than the ones that did rank. Content didn't always mean lots of text, we had several type of content:
- A list of places to buy the product
- Prices from multiple vendors
- Reviews written by users about the product
- Professional pictures of the product
- User submitter pictures of the product
- External links to other sites with articles about the product
- A map of where the product could be found near you
We realized that many of these types of content would be difficult for Google to measure directly. Did it really know there was a map on the page? Was it trying to detect the presence of prices? All the user reviews were on their own pages, could it really measure the amount of text associated with each product by crawling lots of pages and adding the totals? We theorized that it would be much easier for Google to measure how users react to the page and to adjust rankings on that rather than trying to measure amount of content directly.
First, we made some changes to how our bounce rate was measured. We implemented "events" so that when users clicked on the external links, it would be measured in analytics. We also put in "events" for items like moving the map, and scrolling down the page. We figured that when a user interacts with the page, the shouldn't count as a bounce, even if that user didn't view more than one page on the site.
Then we correlated the bounce rate with the amount of content we had for each product. The results were much more dramatic than we were expecting. For products with no content to speak of, the bounce rate was around 90%. For products with lots of every type of content, the bounce rate was under 15%. Products with some content fell in between. We could use this to see which type of content users found most valuable. We could also put a value on soliciting the tenth user review vs digging up the first external link to an article.
Rankings also correlated very closely with this bounce rate. We needed fewer internal links pointing to pages with very low bounce rate to get them to rank #1 than
pages with moderately higher bounce rate.