Google Analytics (GA4) — Knowledge Base

v1 — verify/expand. GA4 + GTM + UTM essentials for 2026. This is the measurement backbone the other docs reference (ads.md, lead-generation.md, email-marketing.md). For presenting the data, see ../dashboards_decks_playbook.md.

GA4 mental model (it's not Universal Analytics)

Everything is an event. No more pageviews-as-a-separate-thing — page_view is just an event. Sessions are derived, not primary.
Event-based, not session-based. Engagement (engaged sessions, engagement rate, engagement time) replaced bounce rate as the core quality metric.
Data model = events + parameters + user/session scopes. Plan events and the parameters they carry up front.

Events

Four tiers:

Automatically collected — page_view, first_visit, session_start, etc. Free, no setup.
Enhanced measurement (toggle in the data stream) — scrolls, outbound clicks, site search, video, file downloads, form interactions. Turn on; cheap signal.
Recommended events — Google's named events with prescribed parameters (e.g. purchase, generate_lead, sign_up, add_to_cart). Use the official names/params so reports + ecommerce features work.
Custom events — your own, for anything not covered. Name in snake_case, consistently.

Register custom dimensions/metrics for any event parameter you want to report on — unregistered params aren't queryable in standard reports.

Watch limits: 50 custom dimensions / 50 custom metrics (standard), and high-cardinality params bloat reports.

Conversions (now "key events")

Mark meaningful events as key events (GA4 renamed "conversions" → "key events"; "conversion" now leans toward Ads-imported usage).
Define a small set of real key events (purchase, qualified lead, demo request) — not every click. Diluting key events makes reporting meaningless.
For lead gen, fire generate_lead on the real qualifying action (handraise), not every form view — see lead-generation.md.
Import key events to Google Ads for bid optimization; use server-side / Enhanced Conversions for signal durability (see ads.md).

Key reports (where to look)

Reports → Acquisition: traffic + user acquisition by channel/source/medium (UTMs drive this).
Reports → Engagement: events, pages, engagement time/rate.
Reports → Monetization: ecommerce (needs the ecommerce event schema implemented).
Reports → Retention: returning users, cohorts.
Explore (free-form, funnel, path, cohort, segment overlap): the real analysis surface — build funnels and paths here, not in standard reports.
Advertising → attribution / model comparison: how channels get credit (GA4 default = data-driven attribution).
Realtime / DebugView: validation, not analysis (DebugView for confirming events fire correctly during setup).

GTM basics

GTM is the deployment layer; GA4 is the destination. Tags (what fires), triggers (when), variables (data).
Data layer is the source of truth — push structured events/data to dataLayer from the site/app, read them into GTM variables, send to GA4. Don't scrape the DOM when you can use the data layer.
Server-side GTM for first-party, durable tagging (CAPI/Enhanced Conversions, reduced ad-blocker loss). Increasingly standard.
Consent Mode v2 required for EEA/UK ads + modeling — wire consent state into GTM.
Test in Preview mode + DebugView before publishing; version + name everything.

UTM hygiene (the cross-doc rule)

Inconsistent UTMs fragment all your reporting. Lock a convention and never freelance.

Param	Use	Rule
`utm_source`	platform/origin	lowercase, consistent (`facebook` not `FB`/`Facebook`)
`utm_medium`	channel type	use GA4's recognized values: `email`, `cpc`, `social`, `organic`, `referral`, `affiliate`
`utm_campaign`	campaign name	a consistent slug (`spring_sale_2026`), kebab/snake, no spaces
`utm_content`	variant/creative/link	for A/B + which link/ad
`utm_term`	paid keyword	search only

Rules:

Always lowercase, no spaces (GA4 is case-sensitive → Email and email split into two channels).
Use a central UTM builder / sheet so everyone tags identically.
medium must match GA4's default channel groupings or traffic lands in "Unassigned."
Don't UTM internal links (overwrites the original source — breaks attribution). UTMs are for inbound only.
Turno house format: source=email, medium=email, campaign=slug — follow per-client conventions where set.

Common pitfalls

Unassigned / (not set) traffic — usually broken or non-conforming UTMs, or missing channel mapping.
Self-referrals / internal UTMs corrupting source attribution.
Treating GA4 like UA — looking for bounce rate, expecting session-first model.
Unregistered event params → not reportable.
Key-event sprawl → every event marked key → meaningless conversion data.
No DebugView validation → events firing wrong/duplicated, discovered weeks later.
Data retention defaults (2 / 14 months) — set to max if you need history in Explore.
PII in events/params — never send emails/names/phone to GA4 (policy violation; use hashed for Enhanced Conversions only).
Cross-domain / subdomain tracking not configured → broken sessions.
Platform numbers ≠ GA4 numbers — they attribute differently; reconcile, don't expect a match (see ads.md).

Quick checklist (GA4/GTM setup)

Enhanced measurement on; data stream configured
Key events defined (small, real set) + imported to Ads
Custom dimensions/metrics registered for needed params
Recommended event names/params used (snake_case)
GTM via data layer; Preview + DebugView verified before publish
Server-side GTM / Enhanced Conversions for durable signal
Consent Mode v2 if EEA/UK
UTM convention enforced (lowercase, GA4-recognized medium, central builder)
No PII in events; internal links not UTM'd
Cross-domain + data retention configured