Bounce rate

Universal Analytics counted a bounce as a session with a single pageview and no further interaction hit. That definition had a famous and structural flaw: it measured whether a second request was made, not whether the visit was any good.

So a visitor who landed on an article, read it for eleven minutes, got exactly what they came for and closed the tab counted identically to one who landed, saw the wrong page, and hit back within two seconds. One of those is the best possible outcome for a content page and the other is the worst, and the metric could not tell them apart.

This produced a decade of misdirected work. Agencies were asked to fix bounce rate on pages that were performing, and responded with related-post widgets and pagination, which improved the metric by manufacturing a second pageview without improving anything for anybody.

It also made bounce rate uncomparable across page types. A contact page and a glossary entry have completely different natural bounce rates, and a site-wide average of the two is a number with no referent.

GA4 does not measure bounces directly. It measures engagement, and bounce rate is whatever is left over: the proportion of sessions that were not engaged.

A session counts as engaged if any one of three things is true. It lasted longer than ten seconds, it triggered a key event, or it included two or more page views. Failing all three makes it a bounce.

The improvement is that the definition is positive. Time on the page now counts as evidence, so the eleven-minute reader is engaged rather than bounced, and the two-second exit is not. It measures something closer to whether the visit was worth anything.

It is still a threshold rather than a judgement, and ten seconds is a low bar. A session that lasts twelve seconds and achieves nothing is engaged by this definition. The metric got better; it did not become an assessment of quality.

These are different measurements that share a name, so a bounce rate chart spanning July 2023 is two datasets joined by a line rather than a trend.

The direction of the change is usually dramatic and in one direction. Sites that reported sixty or seventy per cent bounce in UA frequently report thirty or forty in GA4, because the time threshold rescues a large number of single-page visits that the old definition condemned.

A client seeing that step without explanation reads it as an improvement, and an agency that lets them is accepting credit for a definition change. It will be reclaimed later, usually in a quarterly review, and the reclamation is more expensive than the explanation would have been.

The handling is the same as for every other metric that changed at the migration: annotate the date, state which definition each side of the chart uses, and avoid year-on-year comparisons that cross it.

Bounce rate is only meaningful against the purpose of the page, and some pages exist to end the visit.

A glossary definition, an opening hours page, a phone number, a page answering one narrow question: the ideal outcome is that somebody arrives, gets the answer and leaves satisfied. Every one of those will show a high bounce rate, and improving it would mean failing to answer the question on the first screen.

This matters commercially because clients see the number before they see the context. An agency that reports a site-wide bounce rate will be asked to fix the pages dragging it up, and those are frequently the pages working best.

Two habits avoid the whole conversation. Report bounce rate by landing page rather than as a site-wide average, and say what the page was for when the number looks bad. A high bounce rate next to a note that the page answers one question is information; the same number in a table of averages is a problem to be solved.