Article5 min de lecture

A good bounce rate, by page type and by tool

There is no single good bounce rate. The number depends on what the page is for and on which tool counted it, and the only benchmark worth using is your own site's same page type last quarter.

Par Glyphex ·

Two pages on one site report the same bounce rate this morning, and one of them is working while the other is broken. Whether a bounce rate is good depends on which tool produced it and on what the page was built to do. A glossary entry that answers a question in one screen should bounce most of its visitors. A checkout page should bounce almost none. The same figure is a good result on one and a failure on the other. The second answer, the one the reader came for, is a method: compare the page against your own site's pages of the same type, over the same weeks, in the same tool, and read a change in that line as the signal.

The record, and who wrote it

A bounce rate is a share of sessions that did nothing after the first request, and every tool decides "nothing" differently. Universal Analytics, retired by Google in 2023, counted a bounce as a session with a single hit. Google Analytics 4 inverts it: a session is engaged if it lasted longer than ten seconds, produced a key event, or had two or more pageviews, and bounce rate is the share of sessions that were not engaged, per Google's Analytics Help documentation of engaged sessions. The ten seconds is a setting an admin can raise to sixty in the data stream.

A cookieless tool without a session timer counts a bounce as a visit with a single pageview and no event. Three definitions, three numbers for the same week on the same page. A "good" figure quoted without the tool named beside it is a figure from somebody else's definition.

The site-wide average

A site-wide bounce rate is an average across page types that behave in opposite ways. Add a batch of well-ranking reference pages and the site figure rises while the site is doing better. Remove them and it falls. The site number moves with the mix, and the mix is not a thing anyone is trying to improve.

The useful unit is the page type.

Reference and answer pages

A glossary entry, a documentation page or a how-to that answers its query in the first screen has done its job when the reader leaves. High bounce here is the shape of success. What tells you whether the page worked is not the bounce but scroll depth and whether an outbound link or a copy action fired.

Articles and guides

A long article read to the end and then closed is one pageview in every tool. GA4 rescues it with the ten-second timer; a single-pageview definition does not. On an article, the number worth watching is time on page or the share of visits that reached the halfway scroll mark, and the bounce rate is a by-product of the tool's rule.

Home and hub pages

A home page exists to send people somewhere. A visit that stops there has, in most cases, failed to find the door, so the bounce rate on a hub is one of the few places where a low figure is the goal in itself. The exception is a hub that carries the answer in its own body, which is a design choice with a measurement cost.

Landing pages with one job

A campaign landing page with a single form is measured by its conversion rate, and its bounce rate is the mirror image: every visit that neither submitted nor clicked through. Here bounce and conversion describe the same visits from two sides, and improving one is improving the other.

Product and checkout pages

A product page bounce is a visitor who arrived from search or an ad and left without opening a second page or adding to a cart. A checkout page bounce is an abandoned purchase. Both are worth chasing, and both are dominated by entry page context: the same product page bounces differently when reached from an internal link than from a comparison-shopping ad.

Industry benchmark tables

A table of bounce rates by industry is an average over the customers of one analytics vendor, in one year, using that vendor's definition of a bounce, across whatever page-type mix those customers happen to run. Two vendors publish two tables and the tables disagree, because the inputs differ on all four counts. Such a table is usable only when the sentence quoting it names the publisher, the year and the definition, and even then it describes that vendor's customers. Your site's own last quarter is a closer benchmark than any of them.

The control line

Put two lines on one chart: the page you are worried about, and the pages of the same type across your site, both from the same tool and the same weeks. If both rose together, the tool's definition or a traffic-mix change moved them, and the page is fine. If the page rose alone, something on that page changed: a slower load, a template change, a new referrer sending the wrong readers. Only the second line is evidence about the page.

The ideal, stated per page

The ideal bounce rate for a page is the one its job produces when the page works. For an answer page that is high, for a hub it is low, for a landing page it is one minus the conversion rate you would be content with. Write the job down before you read the number, or the number will supply a job of its own.

The changes that move a page's own line

Load time is the first suspect, then the match between the query that brought the visitor and the first screen they saw, then whether the page offers a next step at all. The guide on this site to reducing bounce rate walks through each.

The visits outside the number

Bounce rate cannot say whether the visitor found what they wanted; a satisfied reader and a disappointed one leave the same single-pageview record. A browser that blocks the tag, or a visitor who declined consent, leaves no record at all, so the bounce rate is a share of the visits the tool could see. Whenever the tool, the timer setting or the page template changes, write the date next to the chart, because a step in the line on that day is a step in the definition and not in the visitors.