How the report library is organised
The left navigation shows a reports snapshot, a realtime report, and then collections. A collection is a group of topics, and a topic is a group of reports: the Life cycle collection holds the acquisition, engagement, monetisation and retention topics; the User collection holds user attributes and technology. Properties that chose business objectives at setup may show a Business objectives collection instead, with topics named for goals such as generating leads or driving online sales. The content is the same data arranged differently.
Within a topic there are two kinds of report. An overview report is a page of summary cards, each showing one dimension against one or two metrics with a link through to a detail report. A detail report is a table with a primary dimension, an optional secondary dimension, a set of metrics and a chart above, which is where real reading happens. Every detail report follows the same pattern, so once Traffic acquisition makes sense, Pages and screens and Ecommerce purchases make sense too. The monetisation topic is also where GA4's average purchase revenue appears, and the glossary entry on AOV in GA4 traces that figure back to the purchase event.
Two reports are worth naming because people look for them in the wrong place. The realtime report shows activity over the last few minutes and is the right place to check that a tag fires, not to judge performance. The Google Ads and Search Console reports appear only once the respective links exist and an editor has added them to a collection; a property where they are missing is usually not linked rather than broken. The GA4 hub lists the setup pages that make those reports appear.
Dimensions and metrics: scope and compatibility
A dimension is a descriptive attribute (source, page path, device category, item name) and a metric is a count or a sum (sessions, views, revenue). Each belongs to a scope, which says what thing it describes: a user, a session, an event, or an item in an e-commerce event. GA4 only produces a sensible number when a dimension and metric describe the same thing, and both the standard report pickers and the exploration builder enforce that, the first by hiding incompatible choices and the second by flagging them.
| Scope | What it describes | Example dimensions | Example metrics |
|---|---|---|---|
| User | A person, across all their sessions | First user source, First user medium, user properties, Audience name | Active users, New users, Total users |
| Session | One visit | Session source, Session campaign, Session default channel group, Landing page | Sessions, Engaged sessions, Engagement rate, Session key event rate |
| Event | One recorded interaction | Event name, Page path, Source (attributed), custom event dimensions | Event count, Views, Key events, Total revenue |
| Item | One product line inside an e-commerce event | Item name, Item brand, Item category, Item list name | Items viewed, Items added to basket, Items purchased, Item revenue |
The traffic-source family is the clearest illustration, because the same words exist at three scopes. "First user source" is the source that first brought the person. "Session source" is the source that brought each session. Plain "Source" is event-scoped and attributed, which means it is filled in by the property's attribution model and only means something next to key events or revenue. Pick the scope that matches the question: how we acquire people, how we acquire visits, or which touchpoints earn credit for outcomes.
Custom dimensionsA GA4 reporting field created by registering an event, user or item parameter so it appears in reports and explorations. follow the same rule. An event parameter the site sends becomes a reportable dimension only after it is registered, at a chosen scope, and it fills from the registration date onward; the events and parameters guide covers the registration step and the documented limits on how many can be registered. A parameter registered at event scope and then paired with Sessions produces a count that is technically correct and usually not what the reader wanted.
Sessions, users, engagement: what the numbers mean
Users. The Users column in most standard reports is active users: people who had an engaged session or whose first visit fell in the period. Total users, which counts anyone who sent any event, is a different metric available in explorations and some views, and New users counts first visits. Which people count as one user depends on the property's reporting identity setting, a view-time choice between blending signed-in identifiers, Google signals, device identifiers and modelled data, or relying on device identifiers alone. Switching it changes user counts without changing the stored data. What a site's own User-ID adds to that blend, and how to choose and set it, is covered in the guide to User-ID and signed-in users.
Sessions. A session starts with a session_start event, carries an identifier and a session number as
parameters, and ends after a documented period of inactivity that an editor can adjust in the data stream
settings. Unlike the previous generation of Google Analytics, a session does not restart at midnight or
when the campaign source changes mid-visit, which is one reason session counts fell for many sites after
migration. Google also documents that user and session counts in reports are estimated with an
approximation technique rather than counted exactly, so small differences between reports, and between
reports and an export, are expected.
Engagement. An engaged session is one that lasted longer than the documented engagement threshold, or
included a key event, or included at least two page or screen views. Engagement rateShare of sessions that were engaged; on social, interactions relative to reach. is engaged sessions
divided by sessions, and bounce rateShare of sessions that were not engaged (GA4); historically single-page sessions. is simply its inverse. Average engagement time is the time the page
was in the foreground, divided by active users, so it is not comparable with the old average session
duration. Views counts page_view and screen_view events together; event count counts everything.
Key events and rates. Key events counts the events marked as key events, and the session key event rate is the share of sessions that included one, which is the metric most teams mean when they say CVRShare of visits or clicks that convert.. The user key event rate is the share of active users who converted. Choose one of the two as the KPIA metric tied directly to an objective. and keep using it; the KPI framework explains why a definition that changes month to month is worse than an imperfect one that stays put.
Which of the two counts should sit under a given rate, and how to explain the mismatch to stakeholders, is covered in the sessions versus users comparison.
Comparisons, filters and the date range
Three controls change what a report shows without changing the report. A comparison overlays a segment built from dimension conditions, such as sessions from mobile devices or users from one country, and the report then shows each comparison side by side, up to a documented maximum per report. A filter restricts the whole report to rows matching a condition, and when an editor saves the customised report the filter becomes part of it, which is how a report for one site section or one market is built. The date range picker sets the period and can add a comparison period, previous period or previous year, aligned by day of week or by date.
Three habits keep these honest. Read partial days as partial: today, and in practice the most recent day or two, are still being processed, so a week-on-week comparison that ends on today understates the current week. Keep comparisons to a readable number, because a table with several comparisons and a secondary dimension says nothing at a glance. And remember that a viewer's comparisons and filters are temporary; if a stakeholder needs the view every week, an editor should save it or build it into a dashboard using the data visualisation guide.
The secondary dimension deserves its own mention. Adding one to a detail report splits each row by a second attribute, which is the quickest way to answer questions such as which campaign brought the paid social sessions or which device category converts worst on a landing page. It also multiplies the number of rows, which is where the cardinality limit below starts to bite.
Reading one Traffic acquisition report properly
A homeware retailer's marketing lead opens Traffic acquisition for the last 28 days and sees 182,400
sessions, an engagement rate of 54% and 2,930 key events for purchase, a session key event rate of
1.6%. She adds the secondary dimension Session campaign to the Paid Social row and finds that a single
campaign, tagged spring_sale_video, brought 21,000 sessions at a 31% engagement rate and 0.4% session
key event rate, while the other paid social campaigns averaged 58% and 1.9%. A comparison for mobile
sessions shows the gap is almost entirely mobile. She checks the landing page on a phone, finds a video
that autoplays above a broken size selector, and files the bug before touching the campaign. The report
answered the question without an exploration; it only needed the right scope and one secondary dimension.
Customising reports and the library
The library, reached from the bottom of the reports navigation by editors and administrators, is where the standard reports are managed. A collection can be published, which makes it appear in the left navigation for everyone, or left unpublished as a working set. Topics inside a collection can be reordered and renamed, and reports can be added to or removed from a topic. Google documents limits on the number of collections, topics and custom reports a property can hold, which is a nudge toward curation rather than accumulation.
Customising a detail report means changing its primary dimension and the list of available dimensions, choosing its metrics and their order, picking the chart types, adding a report-level filter, and then either saving over the existing report or saving as a new one with a different name. Overview reports are customised by adding, removing and reordering summary cards. A sensible first customisation for most properties is a detail report built around the campaign dimensions, with the key events that matter as metrics, filtered to the medium values the team uses; it only works if the campaign tagging follows the naming conventions and UTM guide, because a dimension can only group what the UTMURL tags identifying traffic source. parameters spelled consistently.
Keep a short written note of what was changed and why. A customised report that replaces a default one silently confuses a new team member who follows a help article and finds different columns, and the library does not show history.
Thresholds, sampling and (other)
Three mechanisms hide data in standard reports, and each has a different cause and fix.
Data thresholds. When the reporting identity includes Google signals or modelled data, GA4 withholds rows that could identify individuals, typically small segments over short periods, and shows a warning icon on the report. The documented mitigations are to widen the date range, to remove demographic or interest dimensions, or to switch the reporting identity to device-based, which is a view-time setting that does not alter stored data. A report that looks suspiciously low with the icon showing is not low; it is withheld.
Sampling. Standard reports are not sampledWhen an analytics tool calculates a report from a subset of the data and scales the result up, rather than processing every event.. Explorations can be, when a query covers more events than the documented quota, and the exploration shows an icon when it happens. If a figure from a standard report and an exploration disagree, check the sampling indicator before anything else.
The (other) row and (not set). Each dimension has a documented daily cardinality limit; once a day's distinct values exceed it, the remainder is aggregated into a single "(other)" row that then follows the data into every report using that dimension. Page paths carrying identifiers, search terms and custom dimensions with unique values are the usual causes. "(not set)" is different: it means the dimension had no value for those events, for instance a session-scoped dimension on events that arrived without a session, or a custom dimension registered after the events were collected. The two look alike and have unrelated fixes, and the common GA4 mistakes guide walks through both.
What standard reports cannot answer
Standard reports are tables of one dimension, one optional secondary dimension and a set of metrics, summarised per row. That rules out several common questions. They cannot show a funnel with step-by-step drop-off, the sequence of pages or events people took, the overlap between audiences, the behaviour of a single user, or a cohort defined by anything other than first visit. For those, the explorations tutorial covers the funnel, path, segment overlap, cohort and user explorer techniques, with the sampling caveat above. Questions that need joins to other data, exact counts, or dimensions beyond the cardinality limit go to the BigQuery export, introduced in the BigQuery export basics guide. And a report that several people need every week belongs in a dashboard rather than in a saved GA4 view, which is where the data visualisation guide takes over.
The division of labour is simple to remember: standard reports for the recurring questions with a known shape, explorations for one-off investigation, the export for anything the interface refuses to do.
Common mistakes
- Pairing scopes carelessly. Session metrics against event-scoped custom dimensions, or user metrics against item dimensions, produce numbers that are technically defined and practically wrong.
- Reading Users as everyone. It is active users in most reports; total users is a different metric.
- Comparing engagement time with the old session duration. They measure different things.
- Trusting a report with the threshold icon. Widen the range or change the reporting identity first.
- Ignoring the (other) row. A report with a large (other) row is a cardinality problem, not a traffic pattern.
- Letting viewers' filters stand in for saved reports. If it is needed weekly, an editor saves it.
- Customising defaults without a note. The next person will follow the help centre and find a different report.
- Using a standard report for a funnel question. That is an exploration, and the answer will be wrong if forced into a table.
