Google Analytics 4: the complete guide to a smart and privacy-compliant GA4 setup

Google Analytics 4 is now the standard. Universal Analytics was permanently phased out in July 2023, and for anyone who missed that: there’s no going back. GA4 is fundamentally different in design from its predecessor, and you’ll notice that as soon as you start using the platform properly. It no longer operates on the basis of sessions, but on the basis of events. That may sound like a minor detail, but it changes everything: how you measure, how you report and how you draw conclusions.

In this guide, we explain how GA4 works, which settings you mustn’t overlook, what the latest updates mean for your reports, and how to set up the integration with BigQuery and Consent Mode v2 effectively. No introductory guide for beginners telling you that measurement is ‘important’, but practical steps and honest insights from people who work with GA4 every day.

What makes GA4 different from Universal Analytics

The key difference lies in the data model. GA4 uses an event-based model: everything a user does, from visiting a page to scrolling or clicking, is recorded as an event. This offers much greater flexibility than Universal Analytics’ session-based model, but it also requires a different way of thinking about your data.

• Event-based tracking: every interaction is an event, with freely configurable parameters.
• Cross-platform measurement: the web and app are combined into a single property, providing a consistent user journey.
• Privacy-first by design: built-in features for cookie-free measurement and data minimisation.
• AI-driven insights: machine learning automatically detects trends and predicts user behaviour.

GA4 also automatically tracks a number of events: page views, scrolls, clicks through to external sites, site searches, video engagement and file downloads – all without you having to write a single line of code. That’s a major advantage over the manual goal configurations of the past. But automatic doesn’t always mean correct: you’ll still want to verify for yourself exactly what’s being tracked.

Basic settings you really must configure

Setting up a GA4 property is a breeze. The pitfall lies in everything that comes afterwards. Below are the settings we check first for every new or existing GA4 implementation.

1. Extend data retention to 14 months

By default, GA4 only retains user data for two months. This is rarely sufficient to analyse quarterly trends or seasonal patterns. Fortunately, you can change this to 14 months: go to Admin, then Data collection and modification, then Data retention, and set ‘Event data’ to 14 months. Do this immediately after creating your property, as you cannot retrieve data that has already expired. Please note: for long-term storage and advanced analysis, the integration with BigQuery remains the only real solution.

2. Filtering internal traffic

Your own visits and those of your colleagues skew your data, particularly for smaller websites. Via Admin, you can set up a filter for internal traffic based on IP addresses. At the same time, be realistic about it. The tool already filters out bot traffic automatically, and manually tracking all the IP addresses in a hybrid-working organisation takes a disproportionate amount of time for limited benefit. Focus on excluding regular, heavy internal users, not on eliminating every occasional click from a colleague.

3. Configure the report identity correctly

GA4 offers various ways to recognise users: via a User ID (when someone is logged in), via Google Signals, or via a device ID. The ‘Reporting Identity’ setting determines which method takes priority. For e-commerce sites with logged-in users, it is worth implementing a User ID: this allows you to track a single user across multiple sessions and devices, which significantly improves the quality of your attribution.

4. Establish naming conventions for events

GA4 expects event names to be in snake_case, so ‘form_submit’ or ‘video_play’ rather than random strings. Set this out in a shared document before you start creating custom events. Events that are named inconsistently now will be difficult to consolidate later without breaking historical data. And a practical tip: in 2025, Google renamed ‘Conversions’ to ‘Key Events’ within the GA4 interface. Make sure everyone in your team uses the new terminology so that reports don’t become confusing.

Event tracking: what are you measuring, and why?

One of the biggest misconceptions about GA4 is that you have to measure everything. That’s not the case. Too many events lead to noise, higher costs for BigQuery exports and reports that nobody can make sense of anymore. For each event, ask yourself: what decision am I making based on this data?

Define custom events that align with your specific business objectives. If you have a lead generation website, a click on ‘Request a quote’ or a completed contact form are relevant events. Then use these events as Key Events (formerly: conversions) to make them the focus of your reports and link them to Google Ads. You can then use the audiences you build based on these events for remarketing or lifecycle analysis: who visited a product page but didn’t complete a purchase, and what did that person do afterwards?

Consent Mode v2 and privacy: not an option, but a prerequisite

Privacy has been a moving target for years, but in 2024 and 2025 the requirements have become more specific. Consent Mode v2 is Google’s approach to ensuring that tracking complies with European privacy regulations such as the GDPR. Anyone advertising in the EEA via Google Ads or using GA4 must work with a certified Consent Management Platform (CMP) that correctly transmits consent signals to Google tags.

How does this work in practice? When a user refuses to give consent, the Advanced version of Consent Mode still sends anonymous, cookie-free pings to Google. These pings form the basis for conversion modelling: Google estimates, based on the behaviour of users who did give consent, how many conversions are likely to have taken place amongst the rest. It’s not an exact science, but it still provides you with useful insights into campaign performance, even if a large proportion of your visitors refuse cookies.

Server-side tagging is a logical next step in this regard. Whilst client-side tags are vulnerable to ad blockers and browser restrictions (Safari and Firefox already block third-party cookies entirely), server-side calls are routed via a server that you manage. This improves data quality, reduces page load times and makes your setup more resilient to future privacy changes. For advertisers running substantial budgets via Performance Max or other Google Ads campaigns, this can directly improve the quality of optimisation signals. You can find more background information on Consent Mode v2 in the official Google documentation.

Attribution in GA4: data-driven by default, BigQuery as a workaround

GA4 has set data-driven attribution as the default. This means that, using machine learning, the platform assigns credit to the touchpoints that contributed most to a conversion, based on all available conversion and non-conversion paths. That sounds good, and in many cases it is indeed better than last-click, but there are two caveats.

Firstly: data-driven attribution requires sufficient data to work effectively. For small websites with few conversions, the model is less reliable. Secondly: the way in which organic traffic is attributed has changed following the introduction of data-driven attribution as the standard. Branded search traffic that was previously recorded as direct or organic is now sometimes attributed to Display or Video if the user had previously had a touchpoint with that channel. This explains many of the ‘why have my figures changed?’ queries we receive.

Would you still like to work with rule-based attribution models or apply your own custom attribution logic? If so, the BigQuery export is your best option. GA4 exports raw user-level event data to BigQuery, where you can use SQL to build your own session-level attribution models that aren’t available in the GA4 interface. Good news: since the end of 2024, the export delay has been significantly reduced to four to six hours for standard exports, compared with the previous twelve to twenty-four hours.

Connecting GA4 to Looker Studio and BigQuery: recent improvements

Looker Studio (formerly Data Studio) has seen its native support for GA4 improve steadily in recent times. The GA4 connector now makes almost all event-scoped dimensions and metrics available, including the predictive metrics that were previously only accessible via BigQuery. This makes it easier to build rich dashboards without needing any knowledge of SQL.

The real highlight of 2025 is the support for combined data sources in a single Looker Studio report: merging GA4 data with Google Ads, Search Console and YouTube Analytics previously required a workaround via Google Cloud. It is now available via the standard blended data functionality. For SEO teams looking to compare organic traffic against paid and video traffic, this is a huge time-saver. A handy addition is the ‘data current as of’ timestamp, which you can now pin to any GA4 report so that everyone knows how up-to-date the figures are.

For teams that want to go beyond standard reports, the Explore reports in GA4 itself also offer valuable capabilities: funnel analysis, segment comparisons and path analysis help you answer specific questions that aren’t visible in standard dashboards. For example: at what point in the checkout process do users drop out, and does this vary by channel or device?

AI insights in GA4: useful, but not infallible

With each recent update, GA4 has gained more and more AI-driven features. The platform automatically detects anomalies, highlights spikes in traffic and suggests possible causes, directly within the reporting interface. This works surprisingly well for identifying situations that require attention: a sudden rise in the bounce rate on a landing page, or a drop in conversions on a specific device type.

The AI insights are presented in plain language, making them accessible even to non-analysts. This is an advantage when working with a diverse team. However, do exercise critical judgement: the automated suggestions regarding ‘possible causes’ are hypotheses, not facts. Always combine them with your own annotations (which you can now add alongside the automatic flagging) and validate them via Search Console, your campaign data or other data sources before drawing any conclusions.

GA4’s predictive metrics, such as purchase probability and churn probability, are most valuable for e-commerce and subscription services with sufficient historical data. For smaller sites, they are less useful due to insufficient training data. Use them as a supplementary signal, not as a primary performance indicator. Read more about how GA4 builds these predictions in the Google Analytics Help Centre documentation on predictive analytics.

Recommended tool stack for GA4

GA4 is at the heart of it all, but it is not an island. A well-functioning analytics stack for 2025 looks like this:

• Google Tag Manager (web): for managing and deploying all tags without having to rely on developers for every change.
• Server-side GTM: for privacy-compliant data collection, improved data quality and future-proofing as third-party cookies are phased out further.
• GA4: for data storage, basic analysis and building audiences.
• BigQuery: for raw data export, long-term storage and custom attribution modelling.
• Looker Studio: for dashboards and reports aimed at stakeholders who do not have access to GA4.
• Google Search Console: for organic search data, which you can combine with GA4 data in Looker Studio.

Each of these tools has its own specific role. GTM collects data, GA4 stores and analyses it, BigQuery delves deeper, and Looker Studio presents the results. Using the tools interchangeably for tasks for which they were not designed is one of the most common causes of inconsistent data and non-reproducible reports. You can find more background information on the recommended GA4 configuration in the official Google Analytics documentation.

Common mistakes in GA4 implementations

Based on the audits we carry out, we see a number of errors recurring time and time again:

1. Data retention has not been adjusted: it remains at two months, meaning that explorations do not contain any historical data.
2. No internal traffic filter: visits from within the site distort the data, particularly on smaller sites.
3. Inconsistent event names: a mix of camelCase, spaces and snake_case makes analyses unreliable.
4. Consent Mode v2 has not been implemented: the platform lacks modelled conversions and is exposed to legal risk within the EEA.
5. Key Events not defined: GA4 is set up without conversion goals, meaning you don’t know what ‘success’ is.
6. Setting everything as a Key Event: the opposite problem, where every click is counted as a conversion and the data becomes meaningless.
7. Do not revalidate new conversions following attribution changes: scores will be assigned differently from those set for previously configured Key Events.

Conclusion: GA4 is powerful, but requires a deliberate approach

GA4 offers more possibilities than Universal Analytics ever did. This is both its strength and its pitfall: without discipline, the freedom to measure everything leads to a messy implementation that raises more questions than it answers. Start with a solid basic setup, be selective about what you measure, set up your attribution carefully, and ensure that privacy and consent are sorted from day one, rather than as an afterthought.

The latest developments in AI insights, improved Looker Studio integrations and faster BigQuery exports are making GA4 increasingly useful as the central nervous system of your marketing data. But the technical capabilities are only valuable if the foundations are right. Would you like to know how your current GA4 setup is performing? We regularly carry out GA4 audits and help you configure the right settings, from basic configuration to advanced BigQuery integrations.