Before you start
- Access: Editor or Administrator on the Analytics property (or the right to create one in the account), publish rights on the Tag Manager container (GTMTag management system.), and someone who can edit page templates if no container exists yet. Linking products later needs admin rights on Google Ads and verified ownership in Search Console.
- A measurement plan: the business questions and the actions that answer them, written down. The measurement plan tutorial produces it; this tutorial assumes it exists, because several settings below (key events, custom definitions, retention) depend on it.
- A list of every domain and subdomain the person can move across in one visit: the main site, the checkout or booking subdomain, the payment provider, the help centre, the blog if it lives elsewhere.
- Internal network addresses: the office and VPN ranges, and the addresses of agencies and developers who visit the site daily.
- A consent mechanism already live on the site, and a decision on how analytics storage is handled before consent, because the consent mode guide has to be applied as part of the install rather than after it.
- Campaign tagging conventions agreed, so that UTMURL tags identifying traffic source. parameters arriving in the new property follow one spelling from the first day; the naming conventions guide covers them.
Create the property with the right time zone and currency
In Admin, create a property under the account that the business controls — not an agency's account, and not a personal login, because property ownership is hard to move later. Give it the business name rather than a domain, set the reporting time zone to where the business reads its reports, and set the currency to the one in which revenue is counted. Both are property-level settings and both can be changed, but a change applies going forward only and leaves an odd day in the data, so decide them now. Fill in the business details and objectives screens honestly: they decide which report collections appear by default, and the collections can be adjusted later in the library. Note the property identifier; it appears in Admin and is useful when raising support requests or linking products.
Add a web data stream
Under Data streams, add a web stream with the site's canonical URL and a name people will recognise. The stream receives a measurement identifier beginning with
G-, which the tag will use. One web stream per site is the normal pattern; a second stream is for a genuinely separate site or an app platform, not for a subdomain or a staging environmentA private copy of a website or tool where changes are tested before release, kept behind access control and out of search and production analytics.. Staging is better handled with a separate property so that test events never reach production data. Enhanced measurement is offered at this point; leave the toggle on and review the individual options in a later step. Record the measurement identifier somewhere the Tag Manager implementer can find it, because the next step needs it.Install the tag through Tag Manager
The tag itself is a Google tag configured with the stream's measurement identifier, fired on every page as early as the container allows, and gated by consent state. The install path — container snippet, Google tag, trigger, consent initialisation, preview, publish — is the GA4 via GTM tutorial, which this step defers to rather than repeats. If the site uses a CMS with a native GA4 field, that is acceptable for a page-view-only setup, but it rules out the custom events most of the rest of this tutorial assumes. Whichever path is used, the tag must appear on every template, including error pages, the checkout and any page served from a different system, or users will seem to vanish and reappear as new sessions.
Review enhanced measurement options
Open the stream, then enhanced measurement, and go through the options one by one. Page views include browser history changes, which single-page applications need and which can double-count on sites that push history state on tab changes. Scrolls fire once per page near the bottom. Outbound clicks fire on links to domains not in the cross-domainConfiguring analytics so a visit that moves between a business's own domains stays one session. list, so configure that list before trusting the event. Site search needs the query parameter names the site actually uses; check a real search URL and add the parameter if it is not among the defaults. Form interactions fire on forms the tag can detect, including newsletter boxes and search fields, so sites with a proper lead event often turn this option off to avoid confusing the two. Video engagement applies to embedded YouTube players with the API enabled. File downloads are keyed on extension and are usually fine as they are. Once searches are recorded, the site search analysis tutorial turns them into a report of what visitors asked for and did not find.
Set data retention and confirm the property settings
Under Data collection and modification, open Data retention and choose the longer of the documented event data retention options unless a privacy policy requires the shorter one. The setting governs how long event-level data stays available for explorations and other event-level queries; the standard aggregated reports are not affected by it. Leave the reset-on-new-activity option on so that returning users' retention windows roll forward. While in the property settings, confirm the time zone and currency chosen in the first step, check the industry category, and look at the session timeout and engaged session timer under the stream's tag settings; the defaults suit most sites, and any change should be written down because it alters session counts from that day on.
Define internal and developer traffic, then activate the filters
In the stream's tag settings, define internal trafficVisits from the business's own offices, staff, agencies and testers, labelled by an address rule and filtered so they do not count as customers. with a rule per office, VPN or agency address range; each matching hit is tagged with the
traffic_typeparameter set to internal. Then go to Data filters at property level. GA4 creates an internal traffic filter in a testing state, which means the data is labelled with the test filter's name rather than removed, so you can check the volume in a report before activating. Once the label shows only the expected visits, set the filter to active. Add a developer traffic filter if people will be testing with debug mode on the live site. Filters are not retroactive and an active filter drops data permanently, so activate them deliberately, and revisit the rules when the office moves or a new agency starts.Configure cross-domain measurement and unwanted referrals
In the same tag settings screen, list every domain from the preparation list under configure your domains. With the list in place the tag appends a linker parameter to links and forms between the listed domains, so the same client identifier continues on the other side and the visit stays one session with its original source. Then list unwanted referrals: payment providers, identity providers and any third-party domain the person passes through and returns from, so that the source of the session is not overwritten by the provider's domain. Test by clicking from the main site to the checkout and watching the linker parameter appear in the URL; its absence means the domain is missing from the list or the link is built in a way the tag cannot decorate. Booking widgets in frames and hosted pages that do not carry your tag cannot take part in cross-domain measurement; the guide to embedded widgets and hosted pages covers them.
Decide on Google signals and set consent collection
Under Data collection, Google signals adds demographic and interest reporting and cross-device features for signed-in Google users who have allowed ads personalisation. The cost is that reports containing those dimensions become subject to data thresholdsThe GA4 mechanism that withholds report rows that could identify individuals, usually small segments over short periods, when signals or modelling are used., which hide small rows, and the way signals feed the reporting identity has changed over time, so read the current help page before deciding. Many lead-generation sites leave signals off until an audience use case needs it. On the same screen, review the granular location and device data collection options and the user-provided data setting. The consent behaviour itself is implemented in the container; the consent mode guide covers the implementation and the privacy and consent guide covers the choices, and the admin surface here only reports the resulting consent settings status.
Mark the first key events
Under Key events, create entries for the handful of events that mean business outcomes —
purchase,generate_leador the custom event the measurement plan named — by entering the event name; the entry can be created before the event has been seen, and a toggle on the Events list does the same for events that already flow. Set the counting method per key event (once per event for purchases, usually once per session for leads) and a default value where the event will not carry its own. The choice of which events deserve the mark, and how counting and value work, is the key events guide; the events themselves are designed in the events and parameters guide. This step only switches them on.Link the products the property will exchange data with
Under Product links, link Google Ads first, enabling personalised advertising and auto-tagging in the link so that key events and audiences can flow to Ads and click identifiers flow back; the Google Ads integration tutorial covers what to check after linking. Link Search Console (GSCGoogle's search performance and indexing tool.) by choosing the verified Search Console property and the web stream, then publish the Search Console collection in the report library, or the reports will exist but stay hidden. Link BigQuery if the business will ever ask event-level questions, because the export only contains data from the first full day after it is enabled. Merchant Center and the other advertising products follow the same pattern when they are in use.
Verify in Realtime and DebugView
Open Realtime in one window and the site in another, with Tag Manager preview running so that hits carry the debug flag. Realtime should show your session, the page views and the key events as you trigger them. DebugView shows the same session as a timeline: click each event and confirm the parameters are present and correctly spelled, that user properties appear, and that key events carry the flag icon. Then test the things that break quietly: the cross-domain hop keeps the same client identifier, an internal visit is labelled by the test filter, a consent refusal stops the analytics hits, and a campaign URL arrives with its parameters intact. Come back after a day or two to check the standard reports for self-referrals, "(not set)" landing pages and event counts that look doubled.
Setup checklist
| Item | Where | Done when |
|---|---|---|
| Property time zone and currency | Admin → Property settings | Match where reports are read and how revenue is counted |
| Web data stream | Admin → Data streams | Measurement identifier recorded; one stream per site |
| Tag installed on every template | Tag Manager container | Page views arrive from every page type, including checkout and errors |
| Enhanced measurement reviewed | Stream → Enhanced measurement | Each option deliberately on or off; search parameter confirmed |
| Data retention | Admin → Data retention | Longer documented option chosen unless policy forbids |
| Internal traffic rules and filter | Stream tag settings; Admin → Data filters | Filter tested, then active |
| Cross-domain list and unwanted referrals | Stream tag settings | Linker parameter appears on cross-domain links; providers excluded |
| Google signals and data collection options | Admin → Data collection | Decision recorded; consent status reviewed |
| Key events marked | Admin → Key events | Counting method and value set for each |
| Google Ads and Search Console linked | Admin → Product links | Ads conversions importable; Search Console collection published |
| Verified in DebugView | Explore → DebugView | Parameters, key events and consent behaviour confirmed |
Verifying a holiday-let agency's setup
A holiday-let agency finishes the setup on a Friday and verifies it the following Thursday. The internal
traffic filter, left in testing state, labels 312 sessions from the office and the booking agency over
five days — more than expected, which turns out to be the content team previewing pages — so it is
activated. The booking engine on a separate subdomain was listed under cross-domain, and DebugView shows
the same client identifier on both sides of the hop, yet the referral report still lists the payment
provider as the source of 41 sessions because its domain was not in the unwanted referrals list. Adding
it fixes the attribution for the following week. The generate_lead key event, set to count once per
session, records 56 enquiries in the week against 58 in the agency's inbox, which is close enough to sign
off.
If the property still shows nothing after these steps, the diagnosis page for GA4 showing no data follows one visit from the page to the report to find the break.
Verify it worked
- Browse every template type with Realtime open. Page views should arrive from the checkout and error pages too.
- Click each event in DebugView. Parameters should be present and correctly spelled, and key events should carry the flag icon.
- Click from the main site to the checkout. The linker parameter should appear in the URL, and DebugView should show one client identifier on both sides.
- Visit from an office address while the internal traffic filter is in testing state. The label should appear, only on expected visits.
- Compare a key event's count with a record outside GA4, such as an enquiry inbox. Expect the two to be close, not identical.
If it goes wrong
- Realtime is empty or some page types never appear. The tag may be unpublished, carry the wrong measurement identifier or be missing from that template. Check each.
- The payment provider shows as a traffic source. Its domain is missing from the unwanted referrals list. Add it so the paid channel keeps the credit.
- The linker parameter is absent from a cross-domain link. The domain is not in the list, or the link is built in a way the tag cannot decorate. Check the list first.
- The testing label covers more sessions than expected. Find out who is visiting before you activate the filter, because an over-broad rule removes real visitors permanently.
- Event counts look doubled, or form and search events look odd. Review each enhanced measurement option, especially history-change page views, form interactions and the search parameter name.
Next steps
- Read the events and parameters guide to design custom events beyond page views once the base checks pass.
- Read the key events guide to check your counting method and default values against real data.
- Read the GA4 via GTM tutorial to change tags with preview and versions when a setting here needs revisiting.
Common mistakes
- Creating the property in an agency's or an individual's account. Ownership becomes a dispute later; create it in the business's own account and grant access outward. When an existing property moves between owners, the GA4 property handover checklist checks that access, links and documentation move with it.
- A stream per subdomain. Subdomains belong to one stream with cross-domain configured where needed; separate streams split users and sessions.
- Trusting the defaults on enhanced measurement. Form interactions firing on a search box and a mis-named search parameter are the two most common sources of nonsense events.
- Leaving retention on the shorter option. Explorations silently lose their history; the longer option costs nothing on a standard property.
- Activating the internal traffic filter without testing. An over-broad rule removes real visitors permanently.
- Forgetting the payment provider. Sessions end up sourced to the gateway, and the paid channel that brought the buyer loses the credit.
- Marking key events before the events exist or are stable. A key event on a misfiring event trains bidding on noise once it reaches Google Ads.
- Never opening DebugView. Realtime shows that something arrived; only DebugView shows whether the parameters are right.
The GA4 hub explains the model these settings configure, and the author's Google Analytics for SEO expertise page shows how a property set up this way is read for organic traffic.
