Hi, Internet. There's discourse on Twitter again. It's about bounce rate.
I want to preface this whole post by saying I do not believe that Bounce Rate is a Ranking Factor (unless John Mueller comes out by the time I’ve finished this post with a “hell yeah”, in which case, I stand corrected and hand in my SEO gun and badge.)
This isn’t out of any “in Google we Trust” attitude, really-- it’s just because I can’t think of a best case use case that wouldn’t be extremely noisy or unreliable. So this post is me trying to figure that out; if I were Google, how would I use bounce rate? Would I be able to reliably use it as a ranking factor? Would I be able to stop people from manipulating it?
TL;DR: Bounce rate is probably not a ranking factor, but some version of it is likely used to test and feed the algorithm. But it's not as simple as "high bounce rate bad".
Now. Let’s define our terms
What is bounce rate? No, actually.
Well, let’s define what a bounce is first. There seem to be 3 primary responses to this question:
All of these definitions have different problems when it comes to using them to try and analyze user actions en masse. Let’s go through these problems one by one.
Someone enters your site and immediately leaves
This seems like a pretty good definition of a bounce-- the web equivalent of walking into the wrong classroom (or joining the wrong zoom meeting?)
But this could be for reasons that aren’t specifically to do with the site itself-- or are a testament to the site’s speed and ability to get across information. Is “this person does not exist” a bad site because it gives people exactly what they want? If a weather site tells you the temperature quickly, is that a bad thing? If someone accidentally clicks into your site that they visit often, but don’t want to today, is that bad?
Someone only visits one page of your site, takes an action, leaves
Man, it’d be pretty bad for one page sites if this was all it took for a bounce, huh. And Single Page Apps, which are, and this is shocking, a single page (unless you set it up so that different scrolls and clicks count in a good way, but that's site by site again.) If Google, in ranking, said "if someone visits your site and don't do anything but scroll, we are going to penalize you" that would not be good.
Google has a way to set up non-interaction events for pages to separate out true bounces and people visiting your page and doing nothing, but in a good way. But this has to be implemented, and it is often custom, which means it’s not likely to be something Google can use consistently en masse. If Google can separate "good" and "bad" bounces I'd love to know how.
Someone visits one page but takes no action, leaves/times out
This is the same as the above, but more this time. A user visiting one page and Doing An Activity on that page is less bouncy than a user visiting one page and doing Nothing. But in this case, reading counts as doing nothing.
For the record, in Google Analytics, Google defines a bounce like this:
A bounce is a single-page session on your site. In Analytics, a bounce is calculated specifically as a session that triggers only a single request to the Analytics server, such as when a user opens a single page on your site and then exits without triggering any other requests to the Analytics server during that session. (Analytics help) https://support.google.com/analytics/answer/1009409?hl=en
And FYI-- a session lasts until there’s 30 minutes of inactivity.
When a user, say Bob, arrives on your site, Analytics starts counting from that moment. If 30 minutes pass without any kind of interaction from Bob, the session ends. However, every time Bob interacts with an element (like an event, social interaction, or a new page), Analytics resets the expiration time by adding on an additional 30 minutes from the time of that interaction.
Likely the bounce rate on an article like this would be pretty high-- but I can’t be sure whether that’s because someone saw the length of the article and immediately left, or spent time reading the whole thing.
So what is bounce rate?
Again, according to Google:
“Bounce rate is single-page sessions divided by all sessions, or the percentage of all sessions on your site in which users viewed only a single page and triggered only a single request to the Analytics server.
These single-page sessions have a session duration of 0 seconds since there are no subsequent hits after the first one that would let Analytics calculate the length of the session. “
Here’s the thing: Goog probably doesn’t use Google Analytics data in ranking. We have no idea if they use bounce rate, if this is the way they measure that metric, what data they have, if they look at different information, and so on. So we can’t say that “Google uses bounce rate, as defined by Google Analytics, in ranking” because that’s…. Probably not true.
For the purposes of this long rant/blogpost I’ll be using the “enters and immediately leaves” definition, because the other two definitions outlined above seem to have pretty clear biases against information sites.
“Enters and doesn’t click” leads to the kind of thinking that has clickbait list sites convert their listicles into slideshows. This is the most annoying thing to me, and I’ll probably click through two slides before leaving in a huff. According to all three of the above definitions of bounce, this would not count as a bounce. But trust me, I was bouncing.
The other reason I want to use the “enters and immediately leaves” definition is because I like to use the most generous possible definition in a debate; I don’t want to be accused of being ungenerous or not debating the strongest version of an argument.
I also like this definition from the GA tagging concepts page:
Bounce rate is referred to as a single-page session to your site, but is strictly defined as a single interaction request during a user session. For this reason, a bounce rate for a page is also affected by ecommerce transactions and event tracking requests. This is because these features co-exist with page tracking and, when they are triggered, they result in additional interaction requests to the Analytics servers
So now we have our definition of bounce (user enters the page and immediately leaves), for the purposes of this blogpost.
Another question is weight. Both how much weight would you give those signals in an algorithm-- and how much do the bounces of logged in users count against logged out users? I’m not going to necessarily go into this here, but those are all factors that would need to go into this… ranking… factor.. I would guess that logged in users would weigh more than logged out users. I’m also going to assume this is a light and gentle ranking factor (if it is a ranking factor)
The usual reason folks think Goog would use bounce rate is to measure user satisfaction. The thought goes that of course Google would want to know when someone visits a page, thinks it’s garbage, and immediately leaves. If everyone walked into the party and immediately left that would be pretty bad, right?
I think my response to this is just, like, the general concept of Featured Snippets-- SERP features that are designed to give quick responses to users who just want quick answers. Sometimes immediately leaving is a great sign of user satisfaction-- imagine a DMV where people get in, get what they want, and leave… in minutes...
Doesn’t that sound nice?
So then, a thoughtful SEO might think, maybe it changes based on intent or vertical, using some machine learning magic. So someone looking for “vanessa hudgens age” is going to get pages with high bounce rates, and someone looking for “in depth analysis of war and peace” gets a low bounce rate.
But this is still messy! For example, recipe blogs put a ton of irrelevant stuff at the top of their posts-- but the fact I have to scroll through 7 paragraphs about Hayley M’s kids and dissatisfaction with her husband tells me nothing about whether the recipe at the end of the page is good or bad. Or if I end up on a BBC recipe page to find out if figgy puddings actually have figs in-- I find out the answer (yes) and I leave. That does not a bad experience make.
To go back to my two earlier examples-- maybe I don’t just want “vanessa hudgens age”-- maybe I’m looking for articles about how she’s been treated due to her age. And maybe the result I want for “in depth analysis of war and peace” is a joke website that just says “it’s about war and peace.” People are messy and unpredictable-- except in the ways that are like, say, hiring people to “pogo stick” off competitor sites, creating bots to inflate dwell time metrics, or using bad UX to keep people looking at your site far longer than necessary. (I swear to god stop putting your articles in slideshow format.)
So, uh, I don’t think this is it, cap.
Featured Snippet Generation
Now looking at the above, talking about featured snippets, I got the thought-- maybe bounce rate could be used as part of the determination of whether a featured snippet shows up on a SERP? That makes sense, right-- at least from an analysis POV, en masse, if you have a query or semantically similar groups of queries that have a high bounce rate, that seems like a good candidate for a SERP to get a featured snippet of some sort. This could also feed into those shopping comparison SERP features-- if a user is clicking on and off a lot of pages on a SERP, they could be comparison shopping… or they could be finding that those pages contain something that isn’t what they want.
This is where testing comes in-- there are probably times where the results are perfect Featured Snippet fodder-- and some, where the results are just hot garbage and the algorithm needs to be evaluated in bulk. This is where iterative design, A/B testing, and user experience testing all come in handy. High bounce rate-- try adding a featured snippet and see if the CTR plunges. Pogo sticking? Let's come up with a product comparison SERP feature to stop people from having to click so much.
SEOs may not necessarily be fond of Featured Snippets (depends on the person), but users usually are.
Paul Haahr, in his talk at SMX West 2016 “How Google Works: A Google Ranking Engineer’s Story”, talks about naive interpretations of data. In his example they tested a page with high click through rates through one that had low CTR-- because the answer was pulled into the snippet. Both pages were good, and considered Good by Google. A quick glance, however, would say the one with higher CTR was “good” because clicks are “good.” For Google, this is not necessarily true!
Not to go all Bill Slawski here but there are a few Google Patents about Bounce Rate and Advertising, that at least mention using bounce rate as a way to measure advertising effectiveness.
This paper about predicting bounce rates in sponsored search advertisements is also worth a read.
This isn’t a Google Patent, but a Yahoo! One, about Pogosticking! And Google can use it if they want! (not to be confused with this pogosticking patent) Again; it’s kinda 6 degrees separated from Google itself-- Google definitely doesn’t use every patent that it comes up with or can use, because it just kinda owns a lot of ideas and throws them at the wall like spaghetti.
I really like this article from Dan Taylor about whether CTR in general has a place in ranking, and there are some more links over here.
Another way Google might use Bounce Rate is looking at topicality and volatility-- if a SERP suddenly starts getting a lot of bounces, maybe something is happening there. People are looking for something and not finding it, so that in conjunction with a ton of new searches for a query could lead to more top stories SERP features or news related content for a query.
Detecting Anomalies, Spam, and other non Google things
A lot of Google patents and papers wrt Bounce Rate are more about detecting things like “analytics” and “data anomalies” and “spam.” I think it could be reasonable to say that Google would happily use these metrics for certain purposes-- it makes sense to look at email bounce rate for individuals when trying to decide if something is Spam, for example. But that’s a pretty specific use case. (I think this patent about detecting data anomalies is really interesting, by the by)Now let us run through some common arguments for considering bounce rate as a ranking factor:
But You can Also Manipulate Links
One common refrain from pro-Bounce Rate folks is “you can also manipulate links.” I would contend 3 things:
Let’s go through these one at a time:
It’s easier to manipulate bounce rate
There are a lot of things that go into a good link scheme, but you only need one or two things to manipulate bounce rate. A person and some clicks. Also; obviously links are manipulatable and manipulated! How many @s on twitter to various Googlers are there that a spammy site is now getting all the traffic because they bought 20k links or whatever.
It’s harder to manipulate links and not get caught
The thing about link building schemes is they’re obvious. They’re really, really obvious. Also there are tons of SEOs online ready to throw each other under the bus/disavow 3 million links/generally tattle to Google. I think Google is willing to deal with some link fraud to use links in the calculation of page quality; I honestly think Google doesn’t want to use links in ranking in general. Speculation, but I think Google wants to take as much human intervention out of search as they possibly can-- because then they can just blame the robot when ranking goes wrong.
If Google has the data, surely they’d use the data
At least when you're making correlations and algorithms in good faith!
Today I saw this graph:
This is what you want to avoid when looking at trend lines, data, and noise.
(side note if something is a correlation "if you squint" it's not.)
If someone clicks onto your site and doesn’t scroll or interact with it they’re probably unhappy
If I have to click through 8 slides with 3 adverts to get to the 8th best weird food from Italy or the 3rd largest dog I’m going to be unhappy and I will avoid your site.
People like to see information quickly and then like to move on with their lives. Humans in general like to be safe and relatively lazy. If you have a good user experience there will be times when that “increases” bounce rate. That’s fine.
Google has said-- pretty repeatedly-- that they don’t use pogo-sticking (definition: when a user visits different search results to find a page that satisfies their query) for ranking. I asked around my roommates in a very important, completely unbiased study and found that all of them “pogo stick” as their default browsing experience. Maybe that’s just my house, though.
Bad actor zone
So if I’m a bad actor and I want to manipulate my site’s rankings with clicks and bounces-- and manipulate my competitors, how would I do it?
I would basically want to use clicks to inform Google that sites that look like mine are good in my vertical, for the queries that I’d want to rank for. So I’d want to create some fake users-- logged out and in-- to neatly click on my pages and a few competitors to mess with the metrics. It might make more sense to just hire people to do it. You’d want to do it naturally, not en masse, and you’d want to make some robots that act like people. This seems more complicated
I think a lot of ranking factor studies that look at CTR are pretty limited. They’re limited because CTR and bounce rate etc, are all noisy signals that naturally positively correlate to the ranking that’s already in place. Looking at bounce rate can be interesting and informative when met with caution and like, other data.
Machine learning is fundamentally the study of figuring out why computers are so stupid.
It’s extremely messy, and adding more mess to the pile doesn’t make it easier to figure out why Algorithm A is doing one thing and Algorithm B is doing another.
I think it’s fair to say that Google uses bounce rate and user behavior to evaluate algorithms to some extent. But it’s, shall we say, a naive interpretation (heyo) to say that this kind of click behavior directly impacts ranking.
Google doesn’t care about your vanity site metrics. They care about serving the best results to users so users use Google so they can sell ads and eat data. Bounce rate might be bad for your ecommerce site but it might be good for a user.
I think one major thing that seems to have shifted is the language Goog uses around Bounce Rate. Google Analytics used to describe Bounce Rate as:
Bounce rate for a page is the number of people who entered the site on a page and left right away. They came, they said yuk and they were on their way_
But now, the answer to the question “is a high bounce rate bad” is;
One of my biggest problems with bounce rate as a possible ranking factor is that fundamentally I don’t think it aligns with how a lot of people search. I think a significant number of people “bounce” as the main way they navigate websites.
I don’t use a website’s native search engine to search the site. I don’t have time. I usually use Google or DuckDuckGo (depending on how I’m vibing) as a website search engine (especially for Stack Overflow). Every page I visit gets counted as a bounce-- either by analytics standards (30 mins on the page, remember) or I pogo stick around looking for a precise answer. Or I stick it in an open tag, never to be looked at again. I personally am responsible for 80% of the bounces online, and for that I’m sorry.
Basically; Bounce Rate is a hot vanity metric that can either be real good or real bad or real nothing based on what your site is like. If you have an area of your site that has a high bounce rate… is that bad or good? Looking at user experience flows is more interesting and informative than saying flat out an aggregate number is good or bad.
(Argue with me on twitter)