Website Analytics Metrics Explained: What Each One Really Means

Why definitions matter more than numbers

Two analytics tools pointed at the same website will report different numbers for almost every metric. Neither is broken. They are counting different things and calling them the same name.

Before comparing anything, know exactly what each metric counts and where it breaks down.

Pageviews

Definition: one row per page load recorded by the tracking script.

Sounds simple, and is the metric most distorted by modern web architecture. In a single-page app, client-side navigation fires no page load, so unless the script hooks the History API you record one pageview for an entire visit. Conversely, a script that re-fires on every route change and on browser back can double-count.

Also affected by ad blockers (typically 10–30% of technical audiences), prerendering, and bots.

Use it for: relative popularity of pages.

Do not use it for: absolute traffic claims, or comparison across tools.

Sessions (visits)

Definition: a group of a single visitor's activity, ended by a period of inactivity — conventionally 30 minutes.

The 30-minute rule is convention, not physics, and the edges are arbitrary in ways that matter:

• Someone reading for 45 minutes without clicking gets split into two sessions.
• Some tools also break sessions at midnight, or when a new campaign parameter appears mid-visit.
• A returning visitor after 31 minutes is a new session; after 29, the same one.

Use it for: measuring visit volume and comparing it over time on one tool.

Do not use it for: comparing across tools without checking both timeout rules.

Users / visitors

Definition: distinct identifiers seen in a period.

This is the metric most people trust most and should trust least. It counts identifiers, not people. One person on a phone, a laptop, and a work machine is three visitors. Cleared storage is a new visitor. A shared family device is one visitor for several people.

In privacy-first tools using rotating-salt IDs, the identifier deliberately resets on a schedule, so "unique visitors this month" counts each returning person multiple times. That is a real trade for unlinkability, and the number should be labelled accordingly.

Use it for: trend direction on a consistent methodology.

Do not use it for: headcount, or anything requiring cross-device identity.

Bounce rate

Definition: sessions with a single interaction, divided by total sessions.

Definitions diverge badly here. Classic analytics used single-pageview. GA4 reports the inverse of engagement rate, where an engaged session is 10+ seconds, or a conversion, or 2+ pageviews. These produce very different numbers for the same site.

Critically, it does not measure satisfaction. A visitor who read your whole article and left happy is a bounce.

Use it for: comparing a page to itself over time; spotting regressions after a deploy.

Do not use it for: cross-tool comparison, or as a quality score for content pages.

Exit rate

Definition: for a given page, the share of views of that page that were the last in their session.

Constantly confused with bounce rate. Bounce rate is a property of the session's entry; exit rate is a property of a page. A page can have a high exit rate and be doing its job perfectly — an order confirmation page should have a near-100% exit rate.

Use it for: finding unexpected drop-off points in a multi-step flow.

Do not use it for: judging pages that are supposed to be the last step.

Session duration and time on page

Definition: time between the first and last recorded event in a session (or on a page).

The structural flaw: the last page in a session has no subsequent event, so its duration is unmeasurable. Different tools handle this differently — some record zero, some exclude it, some use engagement pings. A tool that records zero systematically understates time on page for exactly the pages where people stopped to read.

Always read the median, never the mean. Abandoned tabs produce multi-hour outliers that make averages meaningless.

Use it for: relative engagement comparison between similar pages.

Do not use it for: claims about how long people actually read.

Pages per session

Definition: pageviews divided by sessions.

Interpretation is genuinely ambiguous. High can mean engaged exploration — or that people cannot find what they need and are clicking around in frustration. Read it alongside conversion rate: high pages-per-session with low conversion is usually a navigation problem, not engagement.

Conversion rate

Definition: sessions (or visitors) completing a goal, divided by sessions (or visitors).

The denominator is where this metric goes wrong. Session-based and visitor-based conversion rates differ substantially, and neither is "correct" — but comparing one to the other, or switching silently, produces changes that look like performance shifts and are not.

Segment before you conclude anything. A blended 2.4% conversion rate across paid, organic, and direct is an average of three unrelated numbers.

Scroll depth

Definition: the furthest point down the page a visitor reached, usually at 25/50/75/90/100% milestones.

One of the most honest engagement signals available, because it is hard to fake and directly observable. Especially valuable paired with bounce rate: a "bounce" that reached 90% scroll depth is a satisfied reader, not a failure.

Core Web Vitals (LCP, CLS, INP)

Definition: real-user performance measurements — largest content paint, layout shift, interaction responsiveness.

Always read at the 75th percentile, which is what Google uses. Averages hide the slowest quarter of your users, which is disproportionately mobile traffic on poor networks — the segment most likely to leave.

Field data (real users) and lab data (Lighthouse) will disagree. Field data is what ranks.

Frequently asked questions

Why do two analytics tools report different numbers?

Because they define the metrics differently and see different traffic. Session timeouts, engagement thresholds, bot filtering, and ad-blocker reach all vary, so a gap between tools is expected rather than a sign one is broken.

What is the difference between users and sessions?

A session is one visit; a user is the person making visits. One person visiting three times is three sessions and one user — and cookieless tools deliberately weaken the user count, so trust sessions more.

Is a pageview the same as a visit?

No. A pageview is one page load; a visit groups the pageviews from one continuous period of activity. A single visit reading four pages is four pageviews.

Which metric should I actually optimize?

The one closest to the outcome you care about — usually conversions or a well-defined engagement event. Traffic metrics are diagnostic context, not goals, and optimizing them directly rewards the wrong behaviour.

Why does session duration look implausibly low?

Because the final pageview of a session has no successor to measure against and is typically counted as zero. Single-page sessions therefore record no duration at all, which drags the average down.

The metrics worth watching

If you track only five things, track these:

1. Sessions over time, segmented by channel — volume and where it comes from.

2. Conversion rate per channel — whether that volume is worth anything.

3. Scroll depth on key pages — whether content is being read.

4. p75 LCP on mobile — the performance number most tied to abandonment.

5. Week-4 retention by cohort — whether any of it compounds.

Everything else is diagnostic: useful when one of these five moves and you need to find out why. Go deeper on bounce rate, funnels, cohort retention, and Core Web Vitals.