Traffic
Average session duration
Also called visit duration.
Average session duration is the mean elapsed time between the first and last hit of each session. A session with one hit has no span to measure, so how a tool handles it changes the average.
How it is measured
For each session subtract the timestamp of the first hit from the timestamp of the last, then average across sessions in the report. The last page is the blind spot: nothing fires after it, so its reading time is never counted unless a heartbeat or unload event was sent.
Tarsier keeps a random session id in the tab's session storage and sends a ping every 15 seconds while the tab is visible, plus one when the page is hidden. That gives the last page a real value instead of zero, but closing the tab or opening a new one still starts a fresh clock.
Worked example
A recipe blog has 400 sessions in a week. Eighty of them are single-hit and are scored at zero seconds. The other 320 average 3 minutes 20 seconds, so the blended figure lands near 2 minutes 40 seconds. Next week the blog adds a print-recipe button that opens a second page, and the average rises to 3 minutes 05 seconds with no change in reading behaviour.
A reader who spent seven minutes on one long post and left still counts as zero. The mean hides her completely.
How it differs
Average session duration spans a whole sitting, from first hit to last. Time on page measures one pageview against the next hit. Engagement time counts only the seconds the tab was visible. The first can be long while the last is short if people open tabs and walk away.
Common errors
Excluding or including single-hit sessions without saying so. Reading a mean when a few marathon sessions skew it. Treating a rise as better content when a new link just added a second page. Comparing tools with different timeouts. Expecting the last page to count.
In practice
Look at the median or a distribution beside the mean, and split single-hit sessions from the rest. If a number matters, review it by landing page instead of site-wide. Do not set a target for it until you know how single-hit sessions are scored.