In this article
Someone searches for when your shop opens, lands on the page that says 10 to 18, reads it, and closes the tab. That visit is a bounce. The page did its entire job in four seconds, and the number that describes it is the one people treat as a grade.
Bounce rate is the share of visits that saw one page and nothing else. That's the whole definition. Nothing in it is about reading, understanding, buying or coming back, only about whether a second page was ever requested.
Which makes it one of the most misread figures on a dashboard. Here's what it counts, why two tools give you different numbers for the same week, and the handful of cases where a high rate really is telling you something.
What a bounce actually is
Analytics tools group a person's pageviews into a visit, also called a session, and start a new one once they've been gone a while: thirty minutes of inactivity, in most tools. A visit with exactly one pageview in it is a bounce. Bounce rate is those visits divided by all visits.
Two things follow from that, and both surprise people.
- There's no clock in it. “Bounced straight off” is how everyone says it, but the plain definition has nothing to do with speed. Three seconds and eleven minutes are the same bounce, as long as no second page was opened.
- It's a property of the visit, not of the page, even though reports pin it to the page someone arrived on. A site that is one page by design has a bounce rate near 100% and always will.
What the number can't see
Why bounce rate is so thin is worth knowing, because it explains the numbers around it. A page-based analytics tool learns about a visit from requests. Two pageviews give it two timestamps, so it can tell you the visit lasted six minutes. One pageview gives it one timestamp and nothing to compare it with, and browsers don't reliably send an “I'm leaving now” signal when a tab closes.
So the time someone spent on the only page of a bounced visit isn't missing from the report because the tool was careless. It was never measurable. Most tools therefore count a bounce as zero seconds, which quietly drags your average visit length down as well. A site made of good single-page answers looks bad twice.
A bounce is a visit your tool knows almost nothing about. That isn't the same as a visit that went badly.
Where each tool draws the line
If you have ever compared your bounce rate with someone else's and come away either smug or alarmed, this is the part that undoes it. Tools don't agree on what counts.
| Tool | A bounce is | What that does to the number |
|---|---|---|
| Universal Analytics (retired 2023) | A session with a single hit: one pageview and no events | Adding any event at all, a scroll or a video play, stopped visits counting as bounces with nothing changed for the visitor |
| GA4 | A session that wasn't “engaged”. Engaged means over 10 seconds, or two or more pageviews, or a conversion | Lower than anywhere else. GA4 leads with engagement rate and reports bounce rate as 100% minus it |
| Cookie-free tools (Plausible, Fathom, Umami, Lomen) | A visit with one pageview | The plain definition, and higher than GA4 on the same site in the same week |
| Server logs, CDN reports | No such thing: there are requests, not visits | Whatever you compute yourself from one-request sessions, which is why log tools rarely show it at all |
That 10-second rule in GA4 is the big one. A visitor who reads your opening hours for twelve seconds is engaged in GA4 and a bounce in almost every cookie-free tool. Neither is wrong. They answer different questions, in the same way that no two analytics tools agree on your visitor count.
One page, one week, four bounce rates
Illustration, not real data
One pageview and nothing else
cookie-free tools, old Universal Analytics
74%
…unless the visit lasted over 10 seconds
GA4's engagement rule
41%
…unless they also scrolled halfway down
a scroll event added to any tool
23%
…unless they clicked a link off the site
an outbound-click event as well
16%
There's a trap hiding in that chart. In any event-based tool you can lower your bounce rate this afternoon by adding a scroll-depth event, and nobody's experience of the site improves by a hair. When a bounce rate falls in the same week somebody touched the tracking code, that is usually what happened. The benchmarks in “average bounce rate by industry” articles have the same problem: they are numbers from some tool with some definition, and the definition is almost never stated.
What a good bounce rate is
The honest answer is that it depends almost entirely on what the page is for. A rate that would be a crisis on one page is the expected result on another.
| Page | What you'll usually see | How to read it |
|---|---|---|
| Homepage | The lowest on the site | Most people click something. A high rate here usually means the page doesn't say plainly what you do, or the navigation is hard to find |
| Blog post, docs page, answer page | High: 70–90% is ordinary | The page is the destination. Someone who got their answer has no reason to open a second one |
| Contact, opening hours, phone number | High, and fine | The job finishes on the page. Count calls, map clicks and mailto clicks instead, if you count anything |
| Pricing | Middling | People who are seriously interested usually look at something else next, so a very high rate here is worth a closer look |
| One-page landing page | Roughly “didn't click the button” | The one place the rate is close to a real result, because there is exactly one intended next step |
| Category or search results | Should be low | A high rate means people couldn't find what they came for. This is the page type where bounce rate is genuinely diagnostic |
Past that, the only comparison worth making is the page against itself. The same page last month. The same page split by source: visits from search behave nothing like visits from social or from an AI assistant, where people are browsing rather than hunting. Start from your entry pages, the same way you would read Direct traffic by where it lands.
When it's a real signal
Five patterns where a high or rising rate is worth acting on, and what to check first.
The page answers a different question
People arrive expecting one thing and find another. Search Console will show you which queries land on that page; if they don't match what's on it, the fix is the page or its title, not the metric.
It's slow where your visitors are
A rate much higher on mobile than on desktop usually means the page is slow on a phone connection. People who leave before the page paints are bounces with nothing to read.
Something stands in the way
A consent wall, a newsletter popup, an age gate, a login. The tell is that first-time visits bounce hard while returning ones don't, because the returning ones already got past it.
It isn't people
One referrer, a bounce rate near 100%, one pageview each, never a second visit. That's a scraper or an uptime check, and it's worth excluding before you read anything else on the dashboard.
The measuring changed
A jump that lands on every page on the same day is almost always the script rather than the audience: added to some pages and not others, firing twice, or a new definition after a tool upgrade.
What to read next to it
Bounce rate on its own cannot tell a satisfied reader from a disappointed one. Two habits fix most of that.
Decide what you wanted the page to cause, then count that. For plenty of pages it is the next page. For others it's a call, a form, a download, or a click on a link off the site, which you can see if you tag the links you control. A page whose one job is to answer a question has no next step, and that is allowed.
Read bounce rate, pages per visit and average visit length together. They move as a set. Bounces count as zero seconds in that average, so a rising bounce rate always pulls the average down with it: seeing both move is one fact, not two. And when a lot of visits end on a page, check how long those visits were first. A page where long, multi-page visits finish is a page where people got what they came for.
Mostly, though, don't manage the number. Lowering it is easy and nearly always pointless. Split the article across three pages, add a related-posts carousel, track scroll depth. Pages per visit goes up, bounce rate goes down, and not one reader is better served.
Lomen counts a bounce the plain way: a visit with one pageview, visits split on a thirty-minute gap. The card on the overview says as much in so many words, “left after one page”, and it sits beside pages per visit and average visit length, because those three only read properly together. When one of them moves, you can ask Lomen AI what is behind the change, and it answers from that site's own figures.