What average engagement time measures
GA4 runs a timer while your page or app screen is in the foreground and stops it when it is not. Switch to another tab, minimise the window, lock the phone or take a call, and the clock pauses. Come back and it resumes. Total that time and divide it by the sessions, or by the active users, and you have the average engagement time reported in that particular table.
The distinction between those denominators matters more than people expect. Average engagement time per session answers how long a visit held attention; per active user it answers how much attention a person gave across everything they did in the period. They are different numbers, they sit in different reports, and quoting one while meaning the other is a common source of confused meetings.
Why average engagement time matters
The metric it replaced was worse. The old session duration was measured from the gap between the first and last recorded hit, so a visitor who read one page carefully and left registered no time at all, and a tab left open over lunch registered an hour of imaginary reading. Foreground timing removes both distortions and, importantly, does capture the last page of a visit.
That makes it a genuinely useful test of whether a page is being read. Long-form guides, service pages, comparison pages and articles all live or die on whether anyone gets past the opening, and this is the cleanest signal you have of that short of scroll tracking. It is also the metric that gives an engaged session its most common qualification.
Common mistakes with average engagement time
The first is treating longer as better. Time on a checkout page, a pricing table or a contact form is usually friction, not interest, and the right direction there is down. Judge the number against the job the page is meant to do.
The second is comparing pages of different lengths or purposes, or comparing your figure with someone else’s account. There is no benchmark worth repeating; averages are also dragged around by a handful of very long visits, so a median from an exploration often describes the page better.
The third is forgetting what slows the clock down. Where much of the audience arrives on a phone over an uneven connection — normal across Nepal and plenty of other markets — a heavy page spends its first seconds rendering rather than being read, and the figure drops for reasons that have nothing to do with the writing.
How to act on it
Use it to sort pages into three piles: pages where attention is falling and shouldn’t be, pages where it is climbing and the change is worth repeating, and pages where time is a cost rather than a signal. Then pair it with something behavioural — key events, scroll depth, form starts — because time alone never tells you whether the reader got what they came for.
Where the numbers look impossible rather than merely disappointing, check the measurement before the content. Duplicate tags, a single-page site sending views incorrectly, or a consent banner blocking collection will all distort it, and that belongs in an analytics and tracking review.
