Quick definition: Bounce rate is the share of visits that end without the next interaction that an analytics system defines as meaningful. Its exact formula depends on the platform, event configuration, and reporting scope, so a bounce rate is interpretable only when those rules are stated.
What is bounce rate?
Bounce rate is a session-level engagement metric. In its traditional web-analytics meaning, a bounce is a single-page session: a visitor lands on a page and leaves without requesting another page. Modern analytics products often use a broader definition. For example, a session may count as engaged when it lasts beyond a time threshold, contains a conversion event, or includes two page or screen views; bounce rate is then the complement of the engaged-session rate.
Neither version measures satisfaction directly. A visitor who reads an answer, copies a phone number, or completes a task on one page may leave immediately and still have a successful experience. Conversely, a visitor can click through several pages because navigation is confusing. Bounce rate is therefore a diagnostic, not a universal success metric.
The page, traffic source, device, consent state, and observation window matter. A support article, login page, one-page campaign, product-detail page, and multi-step checkout naturally have different expected session patterns. Comparing their raw bounce rates without context produces misleading conclusions.
Bounce-rate formula
The general calculation is:
bounce rate = bounced sessions / all eligible sessions × 100%
Under the single-page definition, the numerator is sessions with one recorded page view. Under an engagement definition, it is sessions that fail all configured engagement conditions. For example, if 2,100 of 8,000 eligible landing sessions contain no second page view and no qualifying event, the bounce rate is 2,100 / 8,000 = 26.25%.
engagement rate = 1 − bounce rate, but only when both metrics use the same eligible-session population and the same event rules.
Exclude sessions that could not have been measured under the chosen definition only through a documented, consistent rule. Bot filtering, consent restrictions, redirects, app webviews, internal traffic, and tag failures can change the denominator. A sudden movement in bounce rate can be a measurement change rather than a behavior change.
How measurement choices change the metric
| Configuration choice | Effect on reported bounce rate | Question to document |
|---|---|---|
| Single-page session rule | Counts any one-page visit as a bounce | Does a one-page task count as success? |
| Engagement timer | Long reads may stop counting as bounces | Is the threshold meaningful for this page? |
| Scroll or video event | Can reduce bounces without a business outcome | Does the event represent genuine engagement? |
| Consent or blocked scripts | May remove or truncate observable sessions | Is missing measurement balanced across variants? |
Do not add a low-value event simply to improve the metric. Firing an event after two seconds or on an automatic animation can turn the dashboard greener while preserving the same visitor experience. An event should represent an action that is plausibly related to the page’s purpose: a search, meaningful scroll depth, successful form submission, product comparison, or completed purchase.
Bounce rate in A/B testing
Bounce rate can help evaluate changes intended to orient visitors after landing: a clearer headline, faster rendering, better search, a more relevant campaign page, or a new navigation pattern. In an A/B test, compare the proportion of eligible randomized sessions or users that satisfy the predeclared bounce rule. The primary question should remain tied to the product decision, such as completed registration, qualified lead creation, purchase, or successful support resolution.
The randomization unit matters. If people are assigned by user but bounce rate is calculated per session, frequent visitors can contribute more observations and may violate the analysis assumption of independent sessions. A user-level indicator such as “had at least one bounced eligible visit during the window” answers a different question. Match the analysis unit to assignment, or use a method that accounts for repeated sessions.
For a binary session outcome, estimate the absolute difference in rates: Δ = bounce rate treatment − bounce rate control. A negative value means fewer bounced sessions under treatment. Report counts, rates, the difference in percentage points, and a confidence interval. A relative reduction can aid communication, but “a 20% reduction” can conceal whether the movement was from 50% to 40% or from 0.5% to 0.4%.
Pair bounce rate with downstream outcomes. A variant that forces a click to reveal information may reduce single-page bounces while increasing frustration. A faster page may improve bounce rate but have no effect on conversion because the offer is weak. Use a decision metric and guardrails for harms such as errors, abandonment, complaints, or slower page load; see the guide to primary and guardrail metrics.
Worked example: a landing-page test
A software company tests a shorter landing-page introduction against its existing page. It randomly assigns 20,000 paid-search visitors to each version. Before launch, it defines a bounced session as one with no lead submission, no product-demo click, and less than 30 seconds of active engagement. The primary metric is qualified demo requests per assigned visitor; bounce rate is diagnostic.
Control records 9,600 bounced sessions, or 48.0%. Treatment records 8,800, or 44.0%. The estimated difference is 44.0% − 48.0% = −4.0 percentage points. The relative reduction is 8.3%. Demo requests rise from 3.0% to 3.3%, while form errors and page-load time stay within prespecified limits.
The result suggests that the shorter introduction helps visitors continue. It does not prove that every non-bounce is valuable. The team checks lead quality and sales follow-up before rolling out. It also validates that the engagement timer, campaign tagging, and consent distribution work the same way in both variants. If the page had gained clicks but lost qualified leads, the lower bounce rate would not justify shipping it.
Interpretation and edge cases
A low bounce rate is not automatically good, and a high rate is not automatically bad. A calculator page can solve a visitor’s need in one view. A recipe page may serve the answer before any click. A page with a click-to-call button can be effective even when analytics cannot observe the phone conversation. Conversely, a low rate caused by accidental clicks, auto-playing elements, or a forced interstitial is not evidence of engagement.
Changes in traffic mix frequently explain movements. Brand-search visitors often behave differently from new paid-social visitors; mobile visitors may have different loading conditions and intent than desktop visitors. Random allocation within a concurrent experiment reduces this problem, but a pre/post comparison does not. Segment only when the segment is planned or when the finding is treated as exploratory.
Do not compare bounce rates across analytics implementations. A migration from pageview-based tracking to event-based engagement can create a large discontinuity even if visitor behavior is unchanged. Preserve the definition in the metric name and annotate measurement changes.
Common mistakes
- Treating every single-page session as failure: account for pages designed to answer a question immediately.
- Changing event rules during a test: this can change one arm’s measurement rather than its behavior.
- Optimizing a proxy alone: lower bounce rate without downstream value can be a distraction.
- Ignoring consent and tracking gaps: unobserved sessions may differ systematically by browser, region, or variant.
- Using only relative change: always report the rate and percentage-point difference.
Frequently asked questions
What is a good bounce rate?
There is no universal benchmark. Compare similar pages, traffic sources, devices, and tracking definitions, then judge the rate against the page’s intended task.
Is bounce rate the same as exit rate?
No. Exit rate is the share of pageviews that were the last pageview in a session. Bounce rate is defined at the session level and applies to sessions that meet the platform’s bounce condition.
Can a high bounce rate be acceptable?
Yes. It can be normal for a concise information page, a store-locator page, or a page where visitors call, copy information, or complete an offline task.
Should bounce rate be a primary A/B-test metric?
Usually not. Use it as a diagnostic or guardrail unless leaving without another tracked action is itself the direct business outcome.
Why did bounce rate fall after adding an event?
The new event may have changed the analytics definition. Verify that it represents meaningful user behavior before interpreting the movement as an experience improvement.
Summary
Bounce rate measures sessions that do not meet a stated continuation or engagement rule. Its formula is simple, but its meaning depends on instrumentation, population, and page intent. In A/B testing, use it to diagnose landing-page behavior, analyze it at a unit consistent with randomization, and interpret it alongside a decision metric and meaningful guardrails.
Sources
- Google Analytics Help: Engagement rate and bounce rate
- NIST/SEMATECH e-Handbook: Hypothesis testing
- Kohavi, Tang, and Xu: Trustworthy Online Controlled Experiments