Website bounce rate explained

Website bounce rate measures the share of a site's sessions that end after one request, with no second page view and no click afterward. Google Analytics 4 decides that number with a countdown clock that an admin can move. Analytics Mania's January 2025 write-up of GA4 bounce rate records the engaged-session timer as a 10-second default that is configurable in the data stream settings up to a maximum of 60 seconds. Push it to 60 and every visit that lasted 11 to 59 seconds without a second pageview or a key event stops counting as engaged. Your bounce rate rises the next morning, and not one visitor behaved differently.

The meaning of a bounce

A bounce is a session that produced one request and nothing after it. The visitor arrived on a page, the tag fired once, and the next record in the log belongs to somebody else. Bounce rate is the share of sessions that ended that way, reported for a site, a landing page, a source or a device.

Two definitions are in circulation, and they do not measure the same event. The classic one counts single-page sessions. Google Analytics 4 counts sessions that failed an engagement test, which is a different question with a different answer on the same week of traffic.

The bounce rate formula

The classic bounce rate formula is bounces divided by total sessions, times 100.

GA4 computes it in reverse. Google's help page on engagement rate and bounce rate defines bounce rate as the percentage of sessions that were not engaged, the exact inverse of engagement rate, so bounce rate equals 100 percent minus engagement rate. Nothing in that arithmetic asks how many pages were viewed.

Bounce rate, exit rate and engagement rate

Metric	Denominator	What it answers
Bounce rate	Sessions that started on the page	Did anything follow the first request?
Exit rate	All sessions that included the page	Was this the last page they saw?
Engagement rate	Sessions	Did the session pass GA4's engagement test?

Bounce rate versus exit rate is a question about position. Every bounce is an exit, and most exits are not bounces: a checkout confirmation can carry a 90 percent exit rate and a bounce rate near zero, because almost nobody starts there. Bounce rate versus engagement rate, in GA4, is not a comparison at all. One is the other subtracted from 100. A related figure, continuation rate, asks the same thing from the other side, how often a page led somewhere.

GA4's engaged-session test

GA4 marks a session as engaged if it lasted longer than the engaged-session timer, or fired a key event, or produced at least two pageviews or screenviews. Any one is enough, and a session that meets none of the three is scored as a bounce.

Two things break this number quietly. The first is a timer that somebody raised: a property at 30 seconds and a property at the 10-second default cannot be compared, and neither can the same property before and after the change, so note the date the setting moved. The second is a key event fired on page load. Mark a scroll or a load-time event as a key event and every session passes the test, engagement rate goes to nearly 100 percent, and bounce rate collapses to a number that means only that you instrumented it away.

To read your own figure, pull it by landing page, then by source, then by device, in that order. A site-wide bounce rate is an average of pages that were never trying to do the same job.

A good bounce rate

Aggregations are the only cross-site numbers on offer, and they are aggregations of other people's settings.

Bounce rate benchmarks by industry are more useful as a ranking than as a target, because traffic mix moves them more than design does. Contentsquare's 2024 Digital Experience Benchmark, covering more than 43 billion sessions across 3,590 websites, found visits from social carried a 41 percent higher bounce rate than visits from paid search. Same sites, same year, one variable.

A good bounce rate is the one your own landing page had last month, measured with the same timer and the same key events, compared with this month. Everything else is somebody else's configuration.

Causes of a rising bounce rate

Load time is the cause with the best evidence behind it. Google and SOASTA trained a deep neural network to 90 percent prediction accuracy on bounce and conversion data, published through Think with Google in 2017, and found bounce probability climbing steadily with mobile page load time.

Consent interfaces move it too. An academic analysis of GDPR's effect on user behaviour measured bounce rate up 2.51 percent eighteen months after the regulation took effect. Treat that as general information about measurement rather than legal guidance; consent questions belong with counsel or your supervisory authority.

The rest are ordinary. The snippet promised a comparison and the page opens with a founder story. The page answers the question completely, which is what a glossary entry or an opening-hours page is supposed to do. Or the tag never fired on that template, so the only sessions recorded there are the ones that left it, and the page shows a bounce rate no design change will fix. Check the server or CDN log for that URL before you redesign anything: if the log holds ten times the requests the tag reported, the problem is instrumentation.

Lowering your bounce rate

Fix the mismatch first, because it costs nothing. Make the title and meta description describe what the page delivers, put the answer in the first screen, and give the page one obvious next step that a reader who got what they came for would still want. Speed work matters as well. Practical detail on these steps sits in the guide on how to reduce your bounce rate, and the vocabulary around sessions and visitors is set out in website metrics explained.

Leave the timer alone while you do it. A configuration change and a content change in the same week produce one line you cannot read.

AI-assistant referrals are arriving with lower bounce rates

Contentsquare's 2026 Digital Experience Benchmark product data, comparing Q4 2025 against Q4 2024, reports bounce rates falling year over year by source: AI-referred traffic down 5 percent, organic search down 4 percent, paid search down 3 percent. AI-referred is the steepest of the three.

That fits what the click is. Someone asked ChatGPT, Claude, Perplexity or Gemini a question, read an answer, and clicked through to check a specific claim, which is a narrower intent than a search result scanned from a list.

The measurement has a hard edge. Visits that carry a referrer from an assistant are a sample of your real AI traffic, because assistant interfaces vary in what they send and some send nothing, and those sessions land in direct. A bounce rate computed on the AI segment describes the visits that kept their referrer. It is not the total, and anyone quoting it as one is quoting a subset.

Measuring bounce rate without cookies

Bounce rate survives cookieless measurement better than most metrics, because the judgement is made inside a single session. Deciding whether a second request followed the first needs a session-scoped identifier, not a persistent one, so a tool that stores no cookies and no IP addresses can still count bounces on a landing page, a source and a device.

What it cannot do is link this session to the one that visitor had on a phone in March. Cross-device journeys stay dark, so "bounce rate for returning visitors" is the segment to distrust, and a returning-visitor share is not recoverable once the identifier is gone. Consent banners shrink the sample further in Europe, which changes the denominator without changing anyone's behaviour.

Contentsquare's 2026 benchmark draws on 99 billion sessions across more than 6,500 websites in nine industries, measured Q4 2024 against Q4 2025, and not one of those properties was running your engaged-session timer.