Bounce rate is the percentage of visits to a website that end without the visitor doing anything the analytics tool counts as engagement. In its oldest and simplest form, a bounce is a visit that lands on one page and leaves without viewing a second. It answers a narrow question: did the landing page persuade the visitor to stay? Advertisers use it to judge landing pages and traffic sources; publishers use it to judge content and referral partners. The number looks universal. It is not. Google Analytics 4 (GA4), the retired Universal Analytics (UA), Adobe Analytics and Plausible each count a bounce differently, so two tools on the same site can report figures far apart.

How the calculation works

Universal Analytics, Google's tool from 2012 until its shutdown, treated a bounce as "a single-page session on your site", meaning a session that "triggers only a single request to the Analytics server", according to Google's legacy help page. Bounce rate was single-page sessions divided by all sessions. Because UA measured time as the gap between two hits, a bounced session lasted zero seconds, so a reader who spent ten minutes on an article and closed the tab was a bounce. Any interactive event, such as a video play, also ended a bounce, while events flagged as non-interaction did not, so implementation choices moved the figure as much as behaviour did.

Google Analytics 4 inverted the logic. GA4 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", according to Google's help centre. Engagement rate is the share of sessions that are engaged, and bounce rate is "the percentage of sessions that were not engaged". The two always add to 100%.

The 10-second timer is a default. Property administrators can raise it to 60 seconds, as PPC Land's guide to GA4 session timeouts set out in May 2024. A site that moves the timer from 10 to 60 seconds will see its bounce rate rise without any change in traffic.

Adobe Analytics keeps the hit-based model. Its bounce rate is "the ratio of visits that contained exactly one hit compared to the number of visits entering on that page", or bounces divided by entries, according to Adobe Experience League documentation updated April 24, 2026. Any second hit, including a link-tracking call, ends the bounce. Plausible, a privacy-focused tool, stays with "the percentage of visitors with a single page view", according to its documentation.

A worked example shows the divergence. Take 1,000 sessions landing on a product page. Six hundred leave without a second page view. Of those, 250 stay longer than 10 seconds and 50 trigger an "add to cart" key event without loading a new page. UA would report a 60% bounce rate. GA4 with the default timer would count the 250 and the 50 as engaged, leaving 300 bounces, or 30%. Raise the timer to 60 seconds and, if only 100 of the 250 pass it, GA4 reports 45%.

Where it sits for buyers and sellers

Bounce rate is a post-click metric with no role in an auction or bid request. It is computed from tags on the advertiser's or publisher's own pages and segmented by landing page, channel, campaign or source. On the buy side, performance marketers read it by campaign and landing page to find mismatches between an ad's promise and the page that follows. On the sell side, publishers watch it by referrer, where a spike from one source can indicate low-quality or automated traffic. In a May 2026 dataset, DataBeat found publisher bounce rate stable at 29% from January to April 2026 even as engagement rate fell from 57.6% to 53.5% and referral engagement collapsed from 42.0% to 19.1%.

Origin and evolution

The term grew out of log-file and tag-based web analytics in the early 2000s. Standardisation came in August 2007, when the Web Analytics Association (WAA, now the Digital Analytics Association) published 26 metric definitions. These defined "Single Page View Visits (Bounces)" as "visits that consist of one page-view" and bounce rate as "single page view visits divided by entry pages", according to Avinash Kaushik, then Google's analytics evangelist, writing on August 23, 2007.

Kaushik had popularised the metric weeks earlier. In a post dated August 6, 2007, he summed up the bouncing visitor's verdict as "I came, I saw, Yuck, I am out of here" and said he preferred a time-based definition, a stay of "usually five seconds or less", to the single-page one. His 2007 rule of thumb was that it was "hard to get a bounce rate under 20%", that anything over 35% was cause for concern and anything above 50% was worrying.

Google moved toward his preference in stages. On July 25, 2012, Alexey Petrov of the Google Analytics Insights Team published a method for adjusted bounce rate, in which a timed event, 15 seconds in his example, fired so that readers who stayed were no longer counted as bounces.

GA4, launched in October 2020, initially dropped bounce rate in favour of engagement rate. Google added bounce rate back in 2022, redefined as the complement of engagement rate, according to Search Engine Land. Standard UA properties stopped processing data on July 1, 2023, and UA 360 properties were shut and their data deleted on July 1, 2024. Ginny Marvin, Google's ads product liaison, later called the GA4 transition "a shock to the system".

Why the marketing community watches it

AI search has given the metric a new job: vendors comparing chatbot and AI-answer traffic with other channels lean on it heavily. On March 17, 2025, Adobe reported that visitors reaching US retail sites from generative AI sources had "a 23 percent lower bounce rate" than those from paid search, affiliates, email, organic search and social, based on more than one trillion visits; travel showed a 45% lower rate. Yet the same visitors were 9% less likely to convert.

More recent figures point the same way. An Adobe and J.P. Morgan chart dated October 27, 2026, cited in an essay by two Andreessen Horowitz partners, showed AI referral bounce rates 32% to 36% below non-AI traffic between March and July 2026. Contentsquare's 2026 Digital Experience Benchmark (99 billion sessions, 6,500-plus sites) reported bounce rates improving, with AI-influenced traffic down 5%, according to the company.

The metric also appears in the dispute over Google's AI Overviews. A randomised experiment with 1,065 Chrome users found AI Overviews cut outbound organic clicks by 39.8%, and the remaining visits showed no meaningful difference in bounce, back-button returns or time on page.

Limitations and disputes

The central weakness is that a bounce does not mean failure. A visitor who finds a phone number, reads an entire article or gets an answer on one page bounces under single-page definitions. GA4's timer replaces one arbitrary threshold with another, and a background tab can pass 10 seconds.

Comparability is the second problem: tags, event design, timer settings and bot filtering all move the figure, so benchmarks rarely compare like with like. Kaushik's 20% to 50% bands were written for single-page definitions in 2007 and do not translate to GA4 percentages. GA4's session logic is also widely misread: 59% of 624 practitioners answered a session attribution question incorrectly in one poll.

The longest-running dispute is over search rankings. Google has repeatedly said it does not use Analytics bounce rate. Gary Illyes said in May 2015 that "we don't use analytics/bounce rate in search ranking", according to Search Engine Journal, echoing Matt Cutts in 2010. Sceptics point to evidence that Google does use its own click data. Pandu Nayak, a Google search executive, testified in October 2023 in the US Justice Department's antitrust trial about NavBoost, a system that "memorizes all the clicks on queries from the past 13 months", according to Search Engine Land's account of the evidence. The May 2024 leak of Google's Content Warehouse API documentation revived the debate. None of this shows that a site's analytics bounce rate feeds rankings; it shows that Google-side click behaviour does.

What bounce rate is not

Exit rate. Exit rate is the share of all pageviews of a page that were the last in a session. Bounce rate, in Google's legacy definition, applies only to sessions that started on that page.

Engagement rate. In GA4 the two are mathematical complements, so a 70% engagement rate is a 30% bounce rate.

Hard bounce. In email marketing a bounce is a delivery failure, and a hard bounce is a permanent one, such as a deleted address.

Bounce tracking. In browser privacy, bounce tracking describes domains that briefly redirect a visitor through themselves to read or set cookies. WebKit, Apple's browser engine, added defences against it to Safari's Intelligent Tracking Prevention. The shared word is coincidence.

Recent developments

The newest argument over bounce rate concerns who, or what, is visiting. The Andreessen Horowitz essay published on November 11, 2026, argued that AI agents are becoming major users of the web and that AI traffic shows "a lower bounce rate". PPC Land's examination of that essay yesterday noted that year-over-year AI traffic growth in the cited chart had slowed from 269% to 62%. Bounce rate was designed for humans loading pages in browsers. As agents fetch pages and complete purchases, the open question is whether an agent session can bounce at all, and whether a falling rate reflects better visitors or merely different ones.

Timeline

  • August 6, 2007 - Avinash Kaushik publishes "Standard Metrics Revisited: #3: Bounce Rate", preferring a five-second definition.
  • August 2007 - The Web Analytics Association publishes 26 standard definitions, including bounce rate as single-page-view visits divided by entry pages.
  • February 2, 2010 - Matt Cutts says Google Analytics is not used in search rankings.
  • July 25, 2012 - Google Analytics blog publishes the adjusted bounce rate method using timed events.
  • May 13, 2015 - Gary Illyes says Google does not use analytics bounce rate in ranking.
  • October 2020 - Google launches GA4 with engagement rate and no bounce rate.
  • 2022 - Bounce rate returns to GA4, redefined as the share of sessions that were not engaged.
  • July 1, 2023 - Standard Universal Analytics properties stop processing data.
  • October 2023 - Pandu Nayak testifies about NavBoost's use of click data in the US v. Google trial.
  • May 2024 - Google Content Warehouse API documentation leaks.
  • July 1, 2024 - Universal Analytics 360 shuts down and UA data is deleted.
  • March 17, 2025 - Adobe reports a 23% lower bounce rate for generative AI retail referrals.
  • February 25, 2026 - Contentsquare's 2026 benchmark reports bounce rates improving, with AI-influenced traffic down 5%.
  • April 24, 2026 - Adobe updates its bounce rate documentation.
  • October 27, 2026 - Adobe and J.P. Morgan chart shows AI referral bounce rates 32% to 36% below non-AI traffic.
  • November 11, 2026 - Andreessen Horowitz essay cites lower AI-traffic bounce rates.

Summary

Who: Advertisers, agencies and publishers who read the metric; analytics vendors including Google, Adobe, Matomo and Plausible that define it; the Web Analytics Association, which standardised it in 2007; and Avinash Kaushik, who popularised it.

What: The percentage of sessions that end without engagement. Universal Analytics and Plausible count single-page sessions, Adobe counts single-hit visits against entries, and GA4 counts sessions under 10 seconds with one page and no key event.

When: Standardised in August 2007, adjusted with timed events from July 2012, dropped and then restored in GA4 in 2020 and 2022, and redefined for Google users when Universal Analytics shut down between July 2023 and July 2024.

Where: On advertiser and publisher websites, measured by tags after the click and reported by landing page, channel and source in analytics tools.

Why: It is a quick test of whether a landing page and its traffic match. Its value depends on the definition in use, it is not a Google ranking signal according to Google, and it is now central to arguments over the quality of AI referral and agent traffic.