Gabriel Espinheira
A visitor lands on your services page, reads the one line that tells them you fix their exact problem, and taps your phone number nine seconds later. Google Analytics files that visit as a bounce.
Now hold that next to the report your last agency sent: bounce rate down 14%, engagement up, three green arrows in a row. Both things can be true at once — a lower bounce rate and a website that sells less than it did last quarter. In GA4, the number and the outcome are not the same thing, and sometimes they move in opposite directions.
Here is the part most founders never get told. A bounce in GA4 is not "someone hated your page." It is "someone didn't cross an arbitrary line" — a ten-second timer, a second pageview, or one tracked action. Miss all three and you bounced, whether you left happy or furious. So on the pages that actually matter to your business, bounce rate tells you who lingered. It says nothing about who bought. Optimise it directly and you can make your site worse.
What GA4 actually counts as a bounce
Google's own documentation defines an engaged session as one that "lasts longer than 10 seconds, has a key event, or has 2 or more screen or page views." Bounce rate is the mirror of that: "the percentage of sessions that were not engaged." Engagement rate and bounce rate always add up to 100.
Read the criteria again, because they are doing something quiet. Nothing there asks whether the visitor got what they came for. A session clears the bar by sitting open for eleven seconds. No scroll, no click, no reason — just time on the clock.
It was not always this loose. In the old Universal Analytics, which stopped processing data on 1 July 2023, a bounce meant a single-page session with no interaction at all. One and done. Google's former analytics evangelist, Avinash Kaushik, once described that bounce from the visitor's side as "I came, I puked, I left." Crude, but clean: it meant the page failed to earn a second move.
GA4 kept the word and changed the meaning. Bounce rate was cut from GA4 at launch, then brought back in July 2022 after enough people complained — rebuilt as the inverse of that ten-second engagement timer. So if your monthly report runs one "bounce rate" line straight across mid-2023, it is splicing two different metrics into one chart and calling the wiggle a trend. That is a graph pretending its own y-axis never moved.
A bounce is a shrug, not a verdict
The instinct is hard to shake: high bounce bad, low bounce good. It is wrong often enough to be dangerous. Bounce rate measures inaction, not dissatisfaction, and those are not the same event.
Picture two visitors. The first reads your whole pricing explainer, finds the single answer they needed, and closes the tab satisfied. The second lands, sees a stock photo above a wall of jargon, and leaves annoyed. GA4 can log both as bounces. The fastest, cleanest win on your site and its most useless visit look identical in the report.
This is why "what's a good bounce rate?" is the wrong question, even though every benchmark post answers it — usually with a range around 40 to 55% and a shrug about context. A glossary page, a contact page, a blog post that fully answers one question: each can run a bounce rate north of 70% while doing precisely its job. The number carries no meaning until you know what the page was for.
You can lower your bounce rate by making the page worse
Here is the part that should make you suspicious of the green arrow. Because the default engaged-session bar is a ten-second timer, the easiest way to "improve" your bounce rate is to make people stay longer. Staying longer is not the same as buying.
Autoplay a background video and engaged sessions climb, because the tab sits open past ten seconds while the clip runs. Bury your phone number three scrolls down and people spend longer hunting for it, so the timer clears the bar before they give up. Slow your page load and the visitor waits, staring, technically engaged. Every one of those changes lifts your engagement rate. Every one of them is worse for the person trying to hire you.
So when a report says "we reduced your bounce rate," the honest follow-up is one word: how? A clearer headline and a faster path to the enquiry is a real win. A hero video and a longer page is not — you paid to make the metric prettier and the sale harder. This is the report theatre owners describe when they say they "stare at reports full of jargon, green arrows, and charts that don't really mean anything." The green arrow is not lying about the data. It is lying about what the data means.
Measure who acted, not who stayed
Retire the site-wide bounce number as a scoreboard. Put one question in its place: did the people who were supposed to act, act?
That needs a real conversion event on every page that matters — a call placed, a form sent, a booking made — tracked from click to client, not click to dashboard. On a service page the number is enquiries, not engagement. On a checkout it is completed orders. On a booking page it is calls held. If a change lifts that number, it worked. If it doesn't, it didn't, however green the engagement chart looks.
Then read it by intent. Ten seconds means one thing for a buyer on your pricing page and something else for a reader who wandered in from a two-year-old post. Separate the traffic that was meant to convert from the traffic that never would, and judge each page against its own job. One honest conversion number, split by why the visitor came, beats a tidy engagement percentage every time.
When bounce rate is still worth a glance
None of this makes the metric useless. It makes it something to investigate, never something to trust. On a content page where you would expect people to read and click through, an engagement rate scraping the floor can be a genuine smoke alarm — the wrong audience found you, the page loads too slowly, or the headline promised something the page doesn't pay off. That is worth a look.
The rule is the direction of the arrow. Use a bad engagement number to ask "what's broken here?" and go find out. Never use a good one to conclude "this page is working." A page is working when it produces the action it exists for. The rest is weather.
The metric that judges your website should be the one your business actually runs on. Not a dwell-time percentage a slower page can inflate. An enquiry. A booking. A call you can answer. That is the whole test — and it is why "I want results, not reports" is the most reasonable thing a founder can say to anyone running their site. The next time a chart shows a metric improving, ask the one thing that cuts through it: which real outcome moved with it? If nothing did, the metric was never the point.
FAQ
What is a good bounce rate in GA4?
There isn't a universal one. GA4 bounce rate is only the share of sessions that didn't last ten seconds, trigger an action, or reach a second page, so a "good" figure depends entirely on the page's job. A blog post can sit above 70% and still work; a checkout at 70% is a fire. Judge each page against the action it exists to produce, not a benchmark.
Why did my bounce rate change when I moved to GA4?
Because the definition changed. Universal Analytics counted a bounce as a single-page session with no interaction. GA4 counts it as any session that fails its engagement test — under ten seconds, no key event, fewer than two pageviews. They measure different things, so a bounce rate that "improved" across the 2023 switch may reflect the new maths, not a better site.
Is bounce rate a Google ranking factor?
No, and Google has said so for over a decade. Search Advocate John Mueller called it "a bit of a misconception that we're looking at things like the analytics bounce rate when it comes to ranking," and Gary Illyes put it plainly: "we don't use analytics/bounce rate in search ranking." Chasing bounce rate for SEO chases a number Google never sees.
