Michael Urquidi II.
Google Analytics 4 // Reference v1 // 2026

The GA4 Field Guide

How GA4 actually thinks, what to configure before you trust a single number, which report answers which question, and the moves that reach the data standard reports will not show. GA4 only. Universal Analytics stopped processing in 2023.

The operating idea

Standard reports tell you what happened. The work that matters is in setup and in Explore.

GA4 measures events, not pageviews, and it ships set to defaults that quietly cost you data. Get the configuration, key events, and data retention right first. After that, most of the questions worth asking are answered two clicks deep in Explore.

00

How GA4 thinks

Everything in GA4 is an event with parameters. There are no goals, no fixed session structure, no pageview-first model. Get these five ideas straight and the rest of the interface stops feeling random.

Data model

Events, not pageviews

A pageview is just an event named page_view. Every interaction is an event, and each event carries parameters (page, value, source). You decide what is worth measuring.

Event types

Four sources of events

Automatically collected, Enhanced measurement (scrolls, outbound clicks, site search, video, downloads, form interactions), Recommended (named by Google), and Custom (yours).

Engagement

Engaged session, not bounce

A session counts as engaged if it lasts longer than 10 seconds, fires a key event, or has 2 or more views. Engagement rate is the positive metric. Bounce rate is simply its inverse.

Key events

You define what matters

Mark an event as a key event and it counts toward your goals. This replaced UA goals and the old GA4 term Conversions. Mark deliberately. Marking everything dilutes the signal.

Identity

How a user is counted

GA4 stitches identity from User-ID, Google signals, device, then modeling. The reporting identity you choose (Blended, Observed, Device-based) changes both your user counts and how much data gets hidden.

Scope

Event, session, user, item

Dimensions and metrics live at a scope. Source has a user scope (first touch) and a session scope (this visit). Mixing scopes in one table is the most common reason a report looks wrong.

The one rename to internalize In March 2024 GA4 renamed Conversions to Key events. Google Ads still calls them Conversions. Same action, two labels: a Key event inside GA4, a Conversion once it is imported into Ads. Nothing reconfigured itself, only the wording changed. When a stakeholder says conversions, they almost always mean key events.
01

Setup that decides data quality

A GA4 property collects data the moment the tag fires, but the defaults are not the settings you want. This is the configuration that separates a property you can trust from one you cannot. Do it once, in this order.

FIRST 30 MINUTES

Foundations that protect your data

do these before reading reports
  • Raise data retention to 14 months. The default deletes granular, event-level data early, which silently cripples Explore. Set both event and user data to the maximum and leave reset on new activity on. Admin › Data collection and modification › Data retention
  • Define internal traffic and filter it. Your own visits and your agency's inflate everything. Create an internal traffic rule by IP, then activate the filter (it starts as Testing). Admin › Data Streams › Configure tag settings › Define internal traffic
  • List unwanted referrals. Stop payment gateways and your own subdomains from showing up as referral traffic and breaking attribution. Configure tag settings › List unwanted referrals
  • Confirm enhanced measurement. It is on by default, but verify scrolls, outbound clicks, site search, and form interactions are enabled, and that site search captures your real query parameter.
Why retention matters more than it looks The retention setting only governs the row-level data behind Explorations. Standard aggregated reports (Traffic acquisition, Pages and screens) keep their totals regardless. So at the 2-month default you can still see old traffic numbers but cannot build a funnel or path for that period. Raise it now, because the change is not retroactive.
MEASUREMENT

Tell GA4 what success looks like

key events and custom data
  • Mark your key events. Find or build the events that represent real outcomes (lead form, signup, purchase, qualified call) and mark them. Keep the list short. Admin › Key events
  • Add custom dimensions for the context you care about. Register the event or user parameters you send (plan tier, content group, logged-in state) so they appear in reports. A standard property allows 50 event-scoped and 25 user-scoped custom dimensions. Admin › Custom definitions
  • Build audiences early. Audiences only collect members from the day you create them, so define cart abandoners, key-event completers, and high-intent visitors before you need them.
  • Decide your reporting identity. Blended maximizes accuracy and cross-device stitching but increases data hiding. Device-based hides far less. Pick based on whether you value precision or completeness more.
CONNECTIONS

Link the rest of your stack

where GA4 gets stronger
  • Google Ads. Import key events as conversions, share audiences for remarketing, and pull Ads cost and click data into GA4. Admin › Product links › Google Ads links
  • Search Console. Once linked and published, GA4 surfaces organic queries and landing pages inside the Reports section, which it cannot show on its own.
  • BigQuery. Free in GA4. Turn on the export to keep raw, unsampled, unthresholded event data with no 14-month ceiling. This is the single highest-value link for serious analysis.
  • Consent and tagging hygiene. Implement Consent Mode where required, and make sure GA4 fires once, from one place (gtag or GTM, not both). Tag your own links with UTMs so they do not collapse into Direct.
Verify before you trust After any tag change, open DebugView and Realtime and watch the events fire with the right parameters. Most bad GA4 data traces back to a tag that double-fires, fires on the wrong page, or drops a parameter. Catch it here, not three weeks later in a report.
02

Reports, what to use when

The left-hand Reports section answers what happened with prebuilt, never-sampled, fast-loading views. Match the question to the report instead of hunting. Most questions map to exactly one.

The question to report map
The question you are actually asking Report Why this one
Is my tag firing, what is happening right now? Realtime Live last-30-minute view. Pair with DebugView to validate events and parameters.
Which channels drive my traffic and key events? Acquisition › Traffic acquisition Session-scoped source, medium, and campaign with engagement and key events side by side.
How did users first discover the brand? Acquisition › User acquisition First-touch (first user source). Use for awareness and new-user origin, not last-click credit.
Which pages get traffic and hold attention? Engagement › Pages and screens Views, average engagement time, and key events by page. The content performance view.
Which landing pages convert? Engagement › Landing page Entry page performance. The honest read on SEO and campaign landing quality.
What events fire, and how often? Engagement › Events Every event by count and users. The inventory check for your measurement plan.
How are my goals trending? Engagement › Key events Key-event counts and rates over time. Formerly the Conversions report.
Revenue, products, and the purchase flow? Monetization Ecommerce purchases, item performance, and the steps of the purchase journey.
Are users coming back? Retention New vs returning, retention curves, and lifetime value at a glance.
Who are my users (age, interests, location)? User › Demographic details Demographics and interests. Requires Google signals on, and rows may be thresholded.
What devices, browsers, and screens? Tech › Tech details Device, OS, browser, app version. Where you find the mobile bug nobody reported.
Cross-channel credit and model comparison? Advertising Attribution, the Model comparison tool, and conversion paths. Lives outside Reports.
User acquisition vs Traffic acquisition These two trip up everyone. User acquisition is first-touch: how a person first arrived, ever. Traffic acquisition is session-scoped: what brought this particular visit. For channel performance and key-event credit, use Traffic acquisition. For where new users originate, use User acquisition.
03

Explore, the power layer

Explorations answer why, who, and which path. This is where you cross dimensions freely, build funnels, trace journeys, and apply segments. If you only ever use standard reports, you are using a fraction of GA4. Reach for the technique that matches the shape of your question.

The seven Explore techniques
Technique What it does Reach for it when
Free form Custom tables, pivots, bar, line, scatter, and geo. The blank canvas. You want to slice a metric by two or more dimensions that no standard report combines.
Funnel exploration Ordered steps with drop-off, open or closed, with elapsed time and breakdowns. You need to see where people fall out of a signup, checkout, or onboarding flow.
Path exploration A tree of the steps users take forward or backward from any node. You want to know what happens after a page or event, including the routes you did not expect.
Segment overlap A Venn of up to three segments and their intersections. You want to find the overlap, for example mobile users who are also converters from organic.
Cohort exploration Behavior and retention by acquisition cohort over time. You are measuring retention, or whether a cohort from one campaign sticks around.
User lifetime Lifetime metrics and predictive values like predicted revenue and LTV. You want to judge an acquisition source by long-term value, not first-session behavior.
User explorer The individual event stream for a single user or anonymous ID. You are debugging, or you need to understand one real journey in full detail.
Segments live in Explore, comparisons live in Reports Standard reports use Comparisons (the chips at the top) to slice on the fly. Explorations use Segments, which are more powerful and reusable, and can be turned into audiences. If you built a segment you love, save it as an audience and send it to Google Ads.
04

High-leverage moves

The handful of techniques that separate people who open GA4 from people who get answers out of it. None of these require code. Two of them include patterns worth copying.

Slicing

Add a secondary dimension

Almost every standard report takes a second lens. Landing page plus session source, or page plus device, turns a flat list into a diagnosis.

Context

Annotate every launch

GA4 brought annotations back in 2025. Drop a note on the day you ship a campaign or redesign so the spike has an explanation later. Reports section only, not Explore or Looker Studio.

Channels

Fix mislabeled traffic

Build a custom channel group to recapture traffic the default rules get wrong, like email that lands in Referral. It applies retroactively in reports.

Attribution

Compare models before trusting credit

GA4 defaults to data-driven attribution. Run the Model comparison tool in Advertising to see how channel credit shifts under last click before you reallocate budget.

Prediction

Use predictive audiences

Once you have volume, GA4 predicts purchase and churn probability. Build a likely-7-day-purchasers audience and push it to Ads for remarketing.

Scale

Leave for BigQuery and Looker Studio

When GA4 thresholds, samples, or caps cardinality, the answer is BigQuery. When stakeholders need a recurring dashboard, the answer is Looker Studio.

PATTERN 01

Regex worth keeping

channel groups, segments, filters
  • Match any paid medium when you build a custom Paid channel or a paid-only segment. Catches the common variants of cpc and paid mediums. 
^(cpc|ppc|cpv|cpm|cpa|paid.*|.*paid)$
  • Keep only your real hostname in an Explore filter to strip out staging, cache, and translation-proxy spam (for example the translate dot goog ghost traffic). Swap in your domain. 
^(www\.)?yourdomain\.com$
  • Split branded vs non-branded organic or paid search. Case-insensitive, matches your brand spelled with or without a space. Replace the brand text. 
(?i)(your ?brand|yourbrand)
PATTERN 02

The BigQuery query you will run first

unsampled, no thresholds, no 14-month limit
  • Top events by count and users over a date range, straight from the GA4 export. Replace the project, dataset, and dates. This is the export equivalent of the Events report, but uncapped and unsampled. 
SELECT
  event_name,
  COUNT(*) AS event_count,
  COUNT(DISTINCT user_pseudo_id) AS users
FROM `your_project.analytics_XXXXXX.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20260101' AND '20260131'
GROUP BY event_name
ORDER BY event_count DESC;
Why this matters BigQuery is the escape hatch for every limitation later in this guide. No sampling, no thresholding, no (other) bucket, and history past 14 months because you own the table. The export is free to switch on. The only cost is BigQuery storage and query, which is negligible for most sites.
05

Data you cannot trust at face value

GA4 will hand you a clean-looking number that is incomplete, bucketed, or measuring something other than what you assume. These are the traps, and the move that gets you past each one.

Thresholding hides rowsWith Google signals on and low counts, GA4 withholds data to protect identity, so totals do not add up. Switch reporting identity to Device-based, or widen the date range.
The (other) rowHigh-cardinality dimensions get bucketed into (other) in standard reports. If a lot of value hides there, move the analysis to BigQuery or lower the cardinality.
Explore can sampleStandard reports never sample, but Explorations do once you pass the event quota. Watch the sampling indicator before you quote an Explore number.
Direct is a junk drawerDirect is where untagged links, app opens, QR codes, and dark social land. A large Direct number usually means missing UTM tags, not loyal visitors.
Two attribution lensesAcquisition reports use last-non-direct click for source and medium. Your property model (data-driven by default) credits key events in Advertising. They will not match. That is expected.
(not set) means missingA (not set) value is a dimension GA4 could not record for that row, often a timing or tagging gap, not a real category.
GA4 is not UADifferent sessions, identity, and attribution. Do not try to reconcile GA4 totals to old Universal Analytics numbers. The methods are not the same.
Double tagging inflatesgtag and GTM both firing, or two containers, double-counts everything. Confirm a single fire in DebugView.
Reports settle slowlyStandard reports can take 24 to 48 hours to finalize. Realtime is instant but limited. Do not call a trend off this morning's data.
Time zone is not retroactiveThe property time zone sets day boundaries. Changing it later does not restate old data, it just creates a seam.
06

Quick reference

The definitions and mappings you will look up again and again. Print this page and the rest can stay closed.

Metrics, in plain terms
Metric What it counts
Users / Active users The headline Users metric in GA4 is Active users: people who had an engaged session in the range.
New users First-ever interaction in the range. New plus returning is not always Total users, because of identity.
Sessions A visit. A new session starts after 30 minutes of inactivity (configurable). Unlike UA, it does not reset at midnight or on a campaign change.
Engaged sessions Sessions longer than 10 seconds, or with a key event, or with 2 or more views.
Engagement rate Engaged sessions divided by total sessions. Bounce rate is one minus this.
Average engagement time Time the page or app was actually in focus. Far stricter than UA's session duration.
Views page_view plus screen_view. This is GA4's replacement for UA Pageviews.
Event count Total times an event fired. Event count per user normalizes it.
Key events Counts of events you marked as important. Formerly Conversions in GA4.
Session / user key event rate Share of sessions or users that completed a key event. The GA4 conversion rate.
Total revenue Purchase revenue plus subscription and ad revenue, where measured.
Campaign tags (UTMs) and where they land
Parameter Purpose GA4 dimension
utm_source Where it came from (google, newsletter, linkedin). Session source
utm_medium The channel type (cpc, email, organic, social). Session medium
utm_campaign The campaign name. Session campaign
utm_term Paid keyword. Session manual term
utm_content Distinguishes ads or links in the same campaign. Session manual ad content
utm_id Campaign ID, ties to imported cost data. Campaign ID
utm_source_platform The platform that sent the traffic (newer, GA4-aware). Source platform
How GA4 buckets channels, and the tagging that keeps you out of Direct
Channel Rule (simplified) Keep it clean
Direct No source, or (direct) / (none). Large Direct usually means missing UTMs. Tag every link you control.
Organic Search medium = organic. Set by search engines. Link Search Console for query data.
Paid Search medium = cpc, ppc, or paid, from a search source. Auto-tagging from Ads handles this. Do not also add conflicting UTMs.
Email medium = email. Tag utm_medium=email or it falls into Referral.
Organic / Paid Social Recognized social source, paid if medium is cpc-like. Tag social posts so organic and paid separate cleanly.
Referral A linking site that fits no other rule. Add payment and auth domains to unwanted referrals.
07

What changed, so your numbers make sense

When a chart shifts and nobody touched the site, it is usually one of these. The platform moved, not your performance.

Jul 2023 Universal Analytics stopped processing. GA4 became the only version. UA report access ended a year later, in July 2024. Historic UA numbers are not comparable to GA4.
Nov 2023 Four attribution models removed. First click, linear, time decay, and position based were deprecated. GA4 now offers data-driven (default), paid and organic last click, and Google paid channels last click.
Mar 2024 Conversions renamed to Key events. Google Ads kept the word Conversions. The migration was automatic and changed labels only, not measurement or history.
Apr 2025 Annotations returned to GA4. Notes on report timelines, with multi-day ranges and color coding. They appear in the Reports section only, not in Explore or Looker Studio.
2025 AI-generated insights and anomaly detection expanded. GA4 now flags unusual movements in reports and suggests likely drivers, which pairs well with annotations for explaining the why.
08

Resources

The official sources worth bookmarking, and a place to practice on real data without a property of your own.

Learn and verify

  • Google Analytics Help Centersupport.google.com/analytics, the current authority on how it behaves
  • Analytics Academy on SkillshopFree official GA4 courses and certification
  • GA4 demo accountThe Google Merchandise Store property, real data to practice on

Build and scale

  • BigQuery export schemaThe field-by-field reference for the raw event export
  • Looker StudioDashboards for stakeholders, blends GA4 with Ads and Sheets
  • Google Tag Manager docsFor event configuration, consent, and clean single-fire tagging

The honest caveat

  • GA4 changes oftenMenu paths and feature names move. When this guide and the interface disagree, the interface wins. Verify in the Help Center.
  • Privacy firstGoogle signals, consent mode, and retention settings carry legal weight in some regions. Configure with that in mind.