GA4 APIs Explained: Data, Admin, Realtime and BigQuery

Before automating GA4 reporting, decide whether you need aggregated metrics, configuration information, recent activity or exported events. The Data API, Admin API, Realtime methods and BigQuery serve different purposes. A reliable design records authentication, report conditions, quotas and reconciliation rules before scheduling recurring extracts.

Separate Data, Admin, Realtime and BigQuery export

The Google Analytics Data API retrieves aggregated reports using dimensions and metrics. It is not a way to rewrite stored reporting data. Ordinary methods such as runReport and the runRealtimeReport method answer different time-based questions.

The Admin API handles account, property and stream configuration, including methods that can change settings. BigQuery export is a separate event-export mechanism. It is not a Data API feature for downloading every raw event.

Choose an interface for the reporting question

For a weekly summary, start with the fewest Data API columns that answer the question. If a CRM join is required, establish a permitted join key and data scope before choosing an event-level architecture. More data does not automatically justify using every interface.

Check API access, authentication and property permissions

Follow the official quickstart to confirm the Cloud project, API enablement, authenticated identity and access to the GA4 property. User authorization and a service account are different identities. A service account also needs access to the target property; enabling the API alone is insufficient.

• Use the numeric property ID, not a measurement ID beginning G-.
• For reporting, confirm read access and an appropriate OAuth scope without granting unnecessary administrative changes.
• An API key alone does not grant access to private reporting data.
• Keep credentials out of articles, shared URLs and spreadsheets; assign ownership and a revocation or replacement process.

Define dimensions, metrics, dates and filters

Use the runReport specification to define dimensions, metrics, explicit dates, filters and ordering. Check metadata and compatibility before building a large extract. Begin with a small result you can compare with the interface.

For organic landing sessions, match the landing-page dimension with session channel conditions such as Session default channel group. First-user, session and event traffic scopes are not interchangeable. Record production hostname and language, especially when the same path exists on several hosts. Distinguish an absent row from a returned zero-valued row.

Plan for quotas, pagination and retries

The following main standard-property Core limits were checked on October 12, 2026. Tokens are not request counts.

Consumption depends on rows, columns, filter complexity and date range, among other factors. Inspect returnPropertyQuota, then respond to the specific limit behind an error such as 429 by adjusting complexity, frequency or concurrency. Bound delayed retries rather than resending indefinitely. Check the current categories separately for methods outside Core.

runReport defaults to 10,000 rows and returns at most 250,000 per request. For a fictional 32,000-row result with limit 10,000, use offsets 0, 10,000, 20,000 and 30,000: four requests, with 2,000 rows in the last one. Fix dates, filters and deterministic ordering, and reconcile collected rows with rowCount. Record capture times because updates during pagination are not guaranteed to share one frozen snapshot.

Reconcile UI, API and BigQuery figures

First confirm that the figures answer the same question. Match metric definitions and denominators, dates, time zone, filters, Reporting identity and freshness. Check your own aggregation logic: adding page-level active users does not create a deduplicated site total.

BigQuery event export does not contain all the reporting transformations or modeled values found in the UI. Nevertheless, raw data is not complete ground truth about every person. Check consent, unsent activity, identity, duplicates, latency and query definitions. Streaming is best effort and can contain gaps; include daily-table reconciliation where appropriate. Do not assume every discrepancy is either a documented difference or a product defect.

Design a fictional weekly reporting workflow

This fictional weekly specification compares Japanese article landing sessions over a completed seven-day period and its matched predecessor. Organic Search and all-channel sessions are extracted separately.

Successful receipt proves that a request worked, not that the comparison or business outcome is valid. Reconcile one run manually before scheduling it, and specify who investigates missing rows or changed definitions.

Record ownership, costs and unmeasured outcomes

Check the export start date and available history; do not assume enabling BigQuery will automatically backfill all earlier activity. Standard daily export is limited to one million events per day. Confirm storage, processing and streaming costs through the official export guidance and current linked pricing rather than promising a free or unlimited system.

A BI tool reading BigQuery still faces Cloud costs, quotas and update delays. Assign responsibility for credentials, failures, missing data, spend and metric changes. Make one necessary report reproducible, then expand the architecture when its next decision requires more data.

Continue with the guide that matches your next task: What Is a DMP? Compare CDPs, CRMs, and Audience Activation; How to Build a Web Analytics Report: GA4 Metrics and Action Plans; GA4 Average Engagement Time: Denominators, Examples and Interpretation; How to Research Keywords: Ideas, Search Volume and Search Console.