In GA4, bounce rate is the percentage of sessions that did not count as engaged. There's no universal good bounce rate, because it describes sessions, not whether visitors got what they came for.
Diagnosis starts with the measurement and the page's purpose. The fixes worth making improve useful actions, with a lower percentage as a side effect, not the goal.
What does bounce rate mean in GA4?
GA4 doesn't define a bounce by pages viewed alone. It asks whether each session met one of three engagement rules. Google's engagement rate and bounce rate page defines an engaged session. It lasts over 10 seconds, has a key event, or has 2 or more screen or page views. Any one of those is enough.
The same page states that the bounce rate is the opposite of the engagement rate. So the two always add up to 100% for the same set of sessions.
In that example the other 70 sessions were engaged, a 70% engagement rate. Count sessions, not users and not events. One person who visits three times contributes three sessions. A session with five ordinary events, one page view and eight seconds in focus is still not engaged.
Each of the three rules has a detail that catches people out. The 10 seconds is measured as time with your page in focus. Google's user engagement documentation defines engagement time that way. For an app, it's time with the screen in the foreground.
The timer stops when someone focuses away from the page or closes the tab. Say a visitor opens your page in a background tab and reads something else for a minute. If they then close it, the session can still be a bounce.
The 10 seconds is also only the default. Google's sessions page explains that the web data stream settings let you change it. So two properties can produce different bounce rates from identical behaviour.
The key event rule is the easiest to misread. Ordinary clicks, scroll events and video plays aren't key events unless someone marks them in the property. A phone click nobody configured as a key event doesn't rescue the session from being a bounce. A scroll event someone did mark as a key event does the reverse. It turns everyone who reaches the bottom of the page into an engaged session.
The two page views rule is the simplest. It's why a duplicated tag can push bounce rate to almost zero overnight. The diagnosis section comes back to that. All three rules depend on settings someone in your business controls. Those are the timer, the key event list and the tag that sends page views.
What changed from Universal Analytics
Google's legacy bounce rate article explains how Universal Analytics counted a bounce. It was a session that sent only a single request to Analytics. So a phone click tracked as an interaction event stopped a one page session counting as a bounce. The rules are different. That's why you shouldn't compare an old Universal Analytics percentage with a GA4 one as if nothing changed.
Bounce is not exit
An exit belongs to a page. A bounce belongs to a session. Google's entrances and exits page defines Exits for a page or screen. It counts how many times a session's last event happened there. Exits and its partner metric, Entrances, are both available in Explore.
A page can be the exit for a session that read six pages and submitted a quote request. A bounce rate of 60% on a page doesn't mean 60% of visitors closed the browser there. Google documents Exits as a count in Explore, not as a rate. So if someone quotes you a GA4 exit rate, ask how they worked it out.
How do you find bounce rate in GA4?
Bounce rate isn't in the default report tables. So you add it yourself, using the short steps on Google's engagement rate and bounce rate page. You need an Editor or Administrator role, because the Customize report button doesn't appear for anyone else.
Open Reports from the left menu. Then open a detail report such as Landing page, Pages and screens or Traffic acquisition.
Click Customize report in the upper right corner.
Under Report data, select Metrics, then click Add metric.
Search for Engagement rate and add it, then search for Bounce rate and add it. If a metric does not appear in the search, it is already in the report.
Click Apply, then save the changes to the current report.
Two things trip people up. Overview reports can't take these metrics. If you only see Add cards and no Metrics option, you're in one and need a detail report. And the new metrics are added as the last two columns. So on a report with many metrics, you may need to scroll right before concluding the save failed.
Bounce rate is also a metric in Explore and in Data Studio reports connected to the property. That's useful when you want it beside outcomes the standard reports don't show together.
Start with the Landing page report
For diagnosis, the first page of the session is the one that matters. That's what Google's Landing page report shows. The dimension is the page path and query string of the first page view in a session. The report is session scoped.
You reach it through Reports, then Engagement, then Landing page. If it isn't in the menu, it may have been removed from the property's navigation. Anyone with edit access can add it back.
Add Session source / medium as a secondary dimension to see which traffic lands on each page. Use the device category comparison to split phone from desktop sessions for the same page.
Pages and screens is a different report answering a different question. It counts every page viewed at any point in the session. So take a service page most people reach from the homepage. Its bounce rate in this report will be very different from its rate as a landing page. Neither number is wrong.
The common mistake is reading Pages and screens, finding 70% for the contact page and rewriting it. But almost nobody landed on it.
Look for a landing page row called (not set). Google's (not set) explanation says it appears when a session has no page_view event. That points to how the tag fires, not to what visitors did. A large (not set) row is a reason to check the setup before reading anything else.
Check the timer before you compare periods
The engagement timer lives in Admin. From there, click through:
Data streams and your web stream
Configure tag settings
Show all
Adjust session timeout
The timer sits beside the ordinary session timeout. By default, that ends a session after 30 minutes of inactivity.
Record both values and the key event list, with the date, before comparing this quarter with last. If someone changed the timer in between, the comparison is meaningless.
Keep that record with the reports. Here's a common mistake. An agency, a plugin and a business owner each change one setting in the same year. Then nobody can say which bounce rate was measured the same way.
Setting up the property so the key events mean something is a separate job. Our guide to GA4 for small business covers it, from choosing key events to testing the tag.
What is a good bounce rate for your page?
There's no useful universal answer. A benchmark is only worth reading if it states its GA4 definition, year, sample and page population. It also needs the engagement timer and key event settings behind it. A figure built on a 30 second timer and generous key events isn't comparable with yours.
The useful comparison is internal:
the same page role
the same channel
the same device
a meaningful window of time
the session counts beside every percentage
Here's what that looks like when every number comes from properties whose settings we know.
SourceOur client GA4 properties, reviewed October 2026
Each bar spans the 25th to the 75th percentile: the middle half of that role's landing pages. The bars overlap and each one is wide. Paid landing pages can run high in industries with a lot of bot and competitor clicks.
A page role tells you where to start looking. It can't tell you whether your number is good. Your own page against its own history, on the same channel and device, can.
The right question for each page is whether the task happened. The table matches the common page roles to the evidence that should sit beside engagement.
Page purpose | What to inspect alongside engagement |
|---|---|
Homepage or service selection | Relevant route clicks, suitable enquiries and whether people understand the offer |
Service or paid landing page | Validated leads, qualification, cost per lead where relevant and the match between promise and page |
Contact or urgent trade page | Successful contact, answered calls and availability. Reading a number and dialling by hand is untracked |
Professional service | Consultation requests, suitability and the delay between research and enquiry |
Product page | Product questions, cart progression, purchases and the purchase conditions |
Article or help page | Whether the question was answered and whether relevant next steps exist. A short visit alone proves nothing |
A phone click and an answered call are different things. Both differ from a recorded bounce. A tradie's emergency page can produce a real job from a session GA4 records as a bounce. The visitor read the number off the screen and dialled it from another phone. No engagement rule was met.
The reverse also holds. A session can qualify as engaged by sitting in focus for 11 seconds while producing nothing of value. Neither direction makes bounce rate useless. It makes it a prompt for a question rather than an answer.
Don't excuse a page too quickly either. A quick exit from an informational article might mean a quick answer. That doesn't make the page automatically healthy. If the page exists to route people toward a service, check whether anyone takes the route.
And do not average the percentages of ten pages into one site figure without their session weights. A page with 40 sessions and a page with 4,000 do not deserve the same vote.
How do you diagnose a high bounce rate?
Diagnose before you fix. The first three steps below can make the problem disappear or move elsewhere before anyone touches the page.
Define the intended outcome
Write down what the page is for and which accepted action proves it worked. For most business pages, that's an accepted enquiry, an answered call, a booking or an order. It's not a scroll or a click.
Confirm collection and settings
Check the tag fires once per page, the timer value, the key event list and the consent setup. A bounce rate from a property with a duplicated tag is fiction.
Choose a comparable cohort
Pick one landing page and one meaningful period. Split it by session source or medium and by device. Compare it with the same page role and channel in an earlier period, not the site average.
Inspect the counts and the splits
Read the session counts beside every percentage. A large swing on a small sample can be pure noise. A sitewide move can reflect a new traffic mix, seasonality, a campaign, consent behaviour or a tagging change.
Test the actual task
Open the page on a phone from the same channel. Then try to do what the visitor came to do. Two minutes doing the task can show you what two hours in reports can't.
Choose one evidenced change
Pick the single change the evidence points to and record it. Leave the measurement settings alone so the before and after can be read.
A page can have a high bounce rate and pass every check. Then you monitor it and spend the effort somewhere else.
When the number itself is the problem
When bounce rate jumps to near 0% or near 100% overnight, suspect the measurement before the page. Visitors rarely change that fast. Tags and settings do.
Broken measurement is common, too. We ran 100 website audits from May 2023 to September 2026. Across them, about 3 in 4 sites had no working GA4 key event for the main enquiry. That figure includes about 1 in 2 sites, where GA4 was missing or not collecting at all.
Look for these before anyone touches the copy.
Duplicated page views. A plugin and a Tag Manager container both sending page_view gives every session two page views. Every session is then engaged and bounce rate collapses toward zero.
Missing page views. A tag can fire late, fire only after consent, or not fire on some templates. That removes sessions or produces (not set) landing pages.
Single page applications and history changes. Enhanced measurement can send page views on browser history events. That means one visit to one screen can register as several pages, or none. It depends on how the site was built.
Key event marking. Marking a common event as a key event makes nearly every session engaged. Examples include scroll, click and session_start.
Timer changes. Someone raised or lowered the engaged session timer. Now the same behaviour lands on the other side of the line.
Consent and blockers. People who decline consent, or whose browsers block the tag, never enter the population. Where behavioural modelling for consent mode is active, part of the reported population is modelled rather than observed.
None of those symptoms proves a particular error on its own. Duplicated page views and a key event on scroll both produce a near zero bounce rate. Only looking at the setup tells you which one you have.
Which fixes are worth making first?
The one the evidence points to, not the one at the top of somebody's generic top ten.
Promise mismatch or irrelevant visits
One channel or campaign bounces heavily, the rest of the site does not
Match the keyword, the ad and the page, and separate poor fit traffic from early research
Unclear offer or missing detail
Short visits from relevant traffic with no route clicks
State service, audience, area, eligibility and next action, and add scope, price guidance and timing
Mobile friction
Phone sessions bounce far more than desktop on the same page
Test the real task on a phone and fix fields, tap targets, call links and sticky elements
Slow or unstable experience
Poor field data in PageSpeed Insights for the page's URL group
Fix the loading, response and layout shift problems the field data names
Weak or misplaced action
People read, scroll and leave without clicking anything
Make the next action specific and put it where the decision happens
Weak proof or a risky offer
Relevant traffic reads the offer and does not act
Put genuine relevant evidence and clear terms near the decision
Clutter or blocked content
High bounce after a popup, slider or chat widget was added
Remove competing interruptions and keep labelled, useful detail
No useful onward route
Articles rank well, read well and send nobody anywhere
Add a relevant service, guide or contact route where it fits
Traffic fit before page fixes
Compare what the search or the ad promised with what the page delivers. Separate poor fit traffic from valuable early research. A visitor who searched a broad term and left in five seconds was never a lead. Rewriting the page for them loses the people it was built for.
Channel can move the number a long way. Someone who searched for your service arrives with a question. Someone scrolling social media wasn't looking for you at all. Say one landing page gets visits from four channels. Its bounce rate might split like this:
Those figures are an example, not a measured result or a benchmark. The point is the split. A page that picks up a burst of traffic from a weaker channel can bounce harder. That can happen without a word on it changing.
Our guide to high intent and low intent keywords covers how to read intent in search terms. For paid traffic, our guide to Google Ads clicks but no leads details the ad to page match.
An end of lease cleaning ad lands on a page headed Complete cleaning solutions, with no tenancy information anywhere on it.
The page names end of lease cleaning and lists what is included and excluded. It offers a quote request for exactly that job.
That's an example, not a measured result. The point is the match: the ad promised one job. The After page names that job before it says anything else.
Clarity, mobile and speed
If the traffic fits and people still leave, check whether a stranger can tell:
what you do
for whom
where
what to do next
Check whether the buying answers are there: scope, price guidance, timing and process.
Try this quick test. Have someone who doesn't know your business read the first screen and say the offer back to you.
Our guide on how to write a homepage that converts covers the first screen's structure. The sentence level work is in our guide to website copy that doesn't sound like a robot.
On phones, test the real task rather than the layout. Can a thumb hit the call link? Does a sticky chat button cover the submit button? Is the form usable with the keyboard open? Can the text be read without zooming?
Use the device mix GA4 actually reports for that page rather than assuming every site is mostly mobile. Some professional service sites are not.
Chart data:
| Sessions | Engaged sessions | |
|---|---|---|
| Desktop | 400 | 300 |
| Phone | 600 | 300 |
The blended figure for that page looks unremarkable. It's 600 engaged sessions out of 1,000: a 60% engagement rate and a 40% bounce rate. Split it and desktop bounces at 25% while phone bounces at 50%. That's a mobile task problem hiding inside an average.
Our guide to getting more phone calls from your website covers the call journey. That includes why a click isn't a call. A difficult or broken form is covered in our guide to enquiry form design.
For speed, run the page through PageSpeed Insights. Read the field data section first, because it reflects real visitors. A Lighthouse performance score is a lab number, as Google's Web Vitals guidance explains. It isn't a pass on the Core Web Vitals thresholds. Nor is it a bounce rate diagnosis.
The check is worth doing. Interaction to Next Paint became a Core Web Vital in March 2024. Since then, about 1 in 2 sites we audited failed at least one Core Web Vital on mobile. Our guide on why your website is slow covers the causes and the order to fix them.
Action, proof, clutter and the onward route
Make the next action specific and relevant to the page. "Request an end of lease quote" tells people what happens next and "Contact us" doesn't. Contact us isn't always wrong, though. And no page needs the same button repeated after every section.
Put genuine evidence near the decision. That means real reviews with their count and source, real credentials and clear terms. Fake reviews, guarantees you won't honour and false urgency are consumer law problems as well as conversion problems.
Remove competing popups and interruptions. Keep the useful detail, labelled sections and navigation for readers who need them. Where an article sends nobody anywhere, add the relevant service, guide or contact route. Don't create extra page views whose only purpose is to lower the bounce rate.
We'll check your GA4 setup and your busiest landing page, then tell you which fix we'd make first.
What should you avoid when lowering bounce rate?
Anything that changes the measurement instead of the experience. The following all lower the percentage and improve nothing.
Marking trivial events such as scroll, click or a timer as key events.
Raising or lowering the engaged session timer to move sessions across the line.
Firing extra page views, or splitting one useful page into three thin ones so people have to click.
Adding popups, interstitials or other interactions just to keep people on the page for 11 seconds.
A lower percentage after any of those is a measurement change, not evidence of a better page. The honest method is slower. Keep the analytics configuration stable. Record the change and its date. Compare suitable periods, or run a controlled test when the page has the volume for one.
Review engagement beside accepted enquiries, answered calls, qualification and cost per lead. Note the sample size, any campaign changes and the sales cycle delay. A simple before and after comparison can suggest the change helped. It cannot prove the page change caused the result, because other things move at the same time.
Page and intended task: [URL and outcome]
Population: [dates, session count, device and channel]
Configuration: [timer and selected key events]
Baseline: [engagement, bounce and accepted outcomes]
Observed problem: [evidence from the actual task]
Change: [one action and owner]
Review: [date, counts, quality and limitations]
Run these checks before you call a page fixed or broken. The broader page by page review lives in our DIY website audit checklist.
The tag fires once and the settings are on record
Duplicated page views and changed timers distort every row
You are reading landing pages, not all page views
Pages and screens answers a different question
Session counts sit beside every percentage
Small samples swing for no reason
Phone and desktop are split for the page
Mobile friction hides inside a blended number
The intended outcome is measured and validated
A click is not an enquiry and a submit is not a receipt
The traffic source matches the page's promise
Poor fit traffic bounces whatever the page says
Someone did the real task on a phone
The reports show where, the task shows why
What we recommend at Elev8d
We start every bounce rate question in the property settings, not on the page. We check the tag fires once and write down the timer and key events. Then we make sure the enquiry form and the phone link are tracked as key events.
Then we pick the landing page that matters most to revenue. That could be the service page your ads or your best rankings send people to. We split it by channel and device before anyone mentions a redesign. If just one channel bounces, we fix the traffic first. That's because rewriting a page for the wrong visitors costs you the right ones.
Only then do we change the page, one thing at a time. We judge it on accepted enquiries and answered calls. If calls are your main enquiry, measure answered calls, not phone clicks. Use call tracking for Google Ads or your phone system.
We'd leave a high bounce rate alone on a page that already brings in the enquiries it should. And we'd skip page by page work when enquiries are missing across the whole site. That's a journey problem, not a page problem.
FAQ
Why did bounce rate change suddenly?
Look for what changed in the property before you look at the page. Property change history, under Admin, shows who changed which setting and when. It covers the last two years. Annotations let you mark releases and campaign launches on the report charts. Google may add its own when a significant data issue hits.
A jump that lines up with a tag release or a settings change is a measurement story. One that lines up with a campaign is a traffic story.
Can a one page website have engaged sessions?
Yes. Staying in focus past the timer makes a session engaged without a second page. So does a key event, such as an enquiry or a phone click you've marked as one. That makes the key events the whole game on a one page site. If they're missing, the engagement rate only tells you who stayed past the timer.
How often should I review bounce rate?
As often as the volume and the business rhythm justify. Always review it after a release, a tracking change or a campaign change. A site with 50 sessions a week can't support a weekly reading. The percentage will move by chance alone. A busy campaign landing page can be read daily while the campaign runs.
Does bounce rate affect Google rankings?
Bounce rate is a figure in your own Analytics property. Google's guide to using Search Console and Analytics data for SEO says engagement rate shows traffic quality. Treat bounce rate the same way: a diagnostic for your own site, not an SEO score.
Next steps
Review one landing page with the worksheet
Pick the page with the most sessions from your most important channel. Add Engagement rate and Bounce rate to the Landing page report. Split it by device and source, do the task on a phone and fill in the worksheet above.
Get help when the problem spans the whole journey
Sometimes enquiries are missing across the whole site rather than on one page. Then the diagnosis covers traffic, pages, forms, calls and follow up together. Our guide to why your website is not converting walks through it. That whole site diagnosis is where our conversion optimisation service starts.
