Session duration

Web, Conversion & AI Searchalso: Average session durationalso: Visit durationalso: Time on site

In one sentence

Session duration is how long an entire visit lasts across all pages, before the visitor leaves or the session times out, and it is a weak quality proxy because efficient answer-finding produces short sessions while confusion produces long ones.

68.01% SparkToro with Similarweb data, published June 2026Last reviewed 4 August 2026

Not to be confused with Dwell time.

Definition

Session duration is how long a whole visit lasts, across every page, before the visitor leaves or the session times out.

It is tempting to read longer as better. It is not. The customer who found what they needed in twenty seconds is your best outcome and your shortest session.

Session duration aggregates page-level dwell time, so it inherits all of the measurement problems dwell time has and adds a few of its own. A session begins on arrival and ends after a period of inactivity, conventionally thirty minutes in many analytics platforms.

The measurement mechanics

  • Time on the final page is difficult to measure, because there is no subsequent event to close it.
  • Some platforms therefore exclude the last page from the calculation, which systematically understates duration.
  • Different platforms handle this differently, so figures are not comparable across tools.

Why it is a weak metric

  • It carries the same ambiguity as dwell time: long can mean engaged or lost.
  • It is heavily skewed by outliers, particularly abandoned tabs.
  • Median is more informative than mean, and is reported far less often.

Where it has some value

  • Comparing traffic sources, where visitors from one channel behave differently from another.
  • Detecting sharp changes, which usually indicate something broke.
  • Segmenting converters against non-converters, which is genuinely informative.

The relationship with voice interactions

  • A voice conversation extends session duration mechanically.
  • A visitor who gets an answer in thirty seconds and leaves has a short session and a successful outcome.
  • A visitor who has a long conversation because the agent could not answer has a long session and a poor outcome.
  • Session duration therefore does not measure voice agent effectiveness in either direction, and should not be used for it.

Common misconception

That longer sessions indicate better engagement. Efficient answer-finding produces short sessions and confusion produces long ones, so the metric frequently rewards the failure and penalizes the win.

Why it matters commercially

Session duration appears in every analytics report and gets used constantly as a quality proxy. It is weak in that role, and its use should be limited to comparison and change detection rather than judging whether a page or a tool is working.

In voice specifically

A voice conversation extends session duration mechanically, so the metric rises whether the conversation helped or not. A thirty-second answer that solved the problem shortens the session; a long conversation that failed lengthens it. Session duration therefore measures nothing about whether the spoken exchange worked, in either direction.

Where AsqVox fits

Session duration should not be used to evaluate a voice agent, in either direction. Task completion, and whether the visitor actually got an answer, are the measures that matter. AsqVox reports on outcomes rather than on how long the Orb kept someone talking.

Visual

Long visit, short visit, and what neither tells you

Long visit, short visit, and what neither tells youSession length (seconds)Outcome30 sec, found the answer8 min, browsed several products12 min, question unanswered15 sec, wrong pageclustersConvertedLeft withoutconverting

Duration and outcome are not correlated in either direction.

Duration runs along the horizontal, outcome up the vertical. The two converted sessions sit at opposite ends of the duration axis, and so do the two that failed, which is the whole point: short and successful, long and failed, long and successful, short and failed all occur. The last page of a session has no closing event, and some platforms exclude it, which systematically understates the figure. A voice conversation extends session duration mechanically, so this measures nothing about whether it helped. Efficient success looks like disengagement. Confusion looks like interest.

Statistics

Every figure carries its source and year. Vendor numbers are labelled as vendor numbers, and where no reliable figure exists this page says so rather than borrowing one.

Zero-click search reached 68.01 percent of US Google searches in early 2026, up from 60.45 percent in 2024.

68.01%independent

SparkToro with Similarweb data, published June 2026, 2026 - Independent third-party measurement rather than a vendor claim, which makes it the firmest figure on this page. It is US Google searches at a point in time, so quote it with the geography and date attached. Its relevance here: later-stage arriving visitors may have shorter, more decisive sessions, which further weakens the longer-is-better reading.

There is no reliable cross-industry session duration benchmark. Figures vary by industry, content type, traffic source and analytics platform methodology.

-no reliable figure

A session duration benchmark quoted without an industry, a traffic source and the platform it came from is not a number you can act on. Compare a site against itself, not against a headline figure.

Analytics platforms differ in whether they include time on the final page, which systematically affects reported figures and prevents cross-platform comparison.

-no reliable figure

Some tools exclude the last page because it has no closing event, which understates duration. Two platforms can report different numbers for the same traffic, so never compare a figure from one against a figure from another.

Session duration distributions are heavily skewed by abandoned tabs, which makes median substantially more informative than mean. Mean is reported more often.

-no reliable figure

The abandoned-tab outliers drag the mean upward. If a report shows only an average, ask for the median before reading anything into a change.

Examples

In practice

A team celebrates rising average session duration following a site redesign. Segmenting by conversion shows non-converting sessions lengthening while converting sessions stayed constant. Visitors were taking longer to fail. The mean moved favorably and the underlying behavior worsened.

The everyday version

Session duration is how long a visit lasts. It is tempting to treat longer as better, but the customer who found what they wanted in twenty seconds is your best outcome and your shortest session. The one who spent ten minutes hunting and gave up looks like your most engaged visitor.

Usage

Who says it

Who uses the term

  • Marketing and analytics teams, in standard reporting, usually as an average alongside pageviews and bounce rate.

Where it turns up

In a reporting pack

  • In analytics dashboards and traffic source comparison, where its comparison and change-detection uses are the ones that hold up.

Common misuse

What it gets used for that it should not

  • Treating longer as better, when efficient success produces short sessions and confusion produces long ones.
  • Reporting the mean rather than the median, when abandoned-tab outliers skew the average upward.
  • Using it to evaluate an on-site tool, such as a voice agent, whose effect on the metric is purely mechanical.

Questions people ask

Does a longer session mean a more engaged visitor?

No. Efficient answer-finding produces short sessions and confusion produces long ones, so session duration frequently rewards the failure and penalizes the win. The customer who found what they needed in twenty seconds is your best outcome and your shortest session, while the one who hunted for ten minutes and gave up looks like your most engaged visitor.

Why report median session duration instead of mean?

Because the distribution is heavily skewed by abandoned tabs. A handful of sessions where someone left a tab open for hours drag the mean upward and make it read higher than the typical visit. The median is substantially more informative and is reported far less often, so if a report shows only an average, ask for the median before drawing a conclusion.

Can I use session duration to judge a voice agent?

No, in either direction. A voice conversation extends session duration mechanically, so the number rises whether the exchange helped or not. A thirty-second answer that solved the problem shortens the session; a long conversation that failed lengthens it. Task completion and whether the visitor got an answer are the measures that actually reflect whether the agent worked.

What is the difference between session duration and dwell time?

Dwell time is time spent on a single page. Session duration aggregates that across the whole visit. Session duration inherits every measurement problem dwell time has, including the unmeasurable last page, and adds the thirty-minute inactivity timeout and cross-platform inconsistency, which is why the two are conflated despite measuring different scopes.

Share this definition

Last reviewed 4 August 2026. Written and reviewed by Dhruv Dholakia, founder of AsqVox.