The fastest way to build a useful custom GA4 report is to define the decision first, then specify the property, date range, row grain, metrics, filters, comparison, sort order, and output. AI can translate that brief into a valid GA4 Data API request, check the field combination, run a small sample, and export the complete result.

The important distinction: this workflow creates an on-demand report in your AI workspace. It does not save or publish a report in the GA4 Reports Library.

“Show me traffic performance” is not a report specification. It leaves the AI to guess which property, acquisition scope, dates, dimensions, metrics, and conversion definition you intended.

A reliable brief answers eight questions:

The most important choice is row grain. If one row combines a landing page, campaign, device, country, and date, the result may be too fragmented to guide a decision. Start with the fewest dimensions that answer the question. Add detail only when it changes the action.

Begin with property discovery instead of pasting an ID from memory. Confirm the property name, website, currency, reporting timezone, data streams, key events, Google Ads links, and registered custom definitions.

This prevents three common errors: querying a staging property, comparing dates in the wrong timezone, and asking for a custom field that belongs to another property.

GA4 report requests use API names, not whatever shorthand appears in a dashboard or internal spreadsheet. Search the property’s reporting metadata for the concept you need – such as landing page, session campaign, engaged sessions, or a registered custom dimension – and review the descriptions before choosing.

Do not invent custom field names. Property-specific metadata is the safest source because it includes registered custom dimensions and metrics available to that property.

Dimensions determine the rows; metrics provide the values. Their scopes must make sense together.

For acquisition work, keep these distinctions explicit:

A report can be syntactically valid and still answer the wrong question. For example, a first-user campaign is not a substitute for the campaign that generated the current session.

GA4 does not allow every dimension and metric in the same Core report. Run a compatibility check on the proposed combination before adding filters or exporting thousands of rows.

Compatibility is a schema gate, not a quality guarantee. It does not confirm that your dates are sensible, the filters express the intended audience, the tracking is correct, or the report supports the business conclusion. It also applies to Core reporting; realtime reporting has a different field set and different rules.

Start with a modest row limit and a clear sort order. Inspect:

A five-minute sample review is cheaper than interpreting a polished but incorrectly scoped export.

Use dimension filters for row attributes such as country, device category, session source, page path, or campaign. Use metric filters for aggregated measures such as sessions greater than 100 or revenue above zero.

Keep boolean logic visible. “Google Ads traffic from India or Singapore, excluding brand campaigns” should be represented as explicit AND, OR, and NOT groups – not compressed into an ambiguous sentence.

For trend decisions, compare complete, equal-length periods. Label each date range so the output remains readable. Current-day data is provisional, and recent event attribution can still change, so avoid treating a partial period as final.

Large GA4 results are paginated. A useful response should tell you the total matching row count, how many rows were collected, the starting offset, and whether the result was truncated.

Do not confuse the inline preview with the complete dataset. A workflow can show 200 rows in chat while exporting many more to CSV. If the collected-row cap is below the total matching rows, raise the cap or narrow the question before analysis.

The CSV is only half the deliverable. Save the report brief alongside it:

This makes the report reproducible and gives reviewers enough context to challenge the conclusion.

Decision: Which paid landing pages need a message, speed, or conversion-path review?

Rows: Landing page plus session source/medium.

Measures: Sessions, engaged sessions, engagement rate, key events, session key-event rate, and revenue where relevant.

Filters: Paid sources or campaign rules that match the team’s definition. Avoid assuming every CPC-tagged session is Google Ads if other networks use the same medium.

Review gate: Join CRM or lead-quality evidence before reallocating budget. GA4 behavior alone does not establish downstream lead value.

Decision: Which campaigns changed materially, and what needs investigation?

Rows: Session campaign, optionally adding session source/medium.

Measures: Sessions, engaged sessions, key events, and revenue.

Comparison: Last complete 28 days versus the previous 28 days.

Review gate: Check naming changes and (not set) rows before ranking campaigns. A renamed campaign can appear as two separate rows.

Decision: Which traffic segments reach a key event, and where does the path appear to break?

Rows: Event name plus one diagnostic dimension such as device category, landing page, or session campaign.

Measures: Event count, users, sessions, and the relevant key-event metric.

Review gate: Confirm the event is registered and firing correctly. A reporting query cannot prove the browser event or consent flow works.

Filters are where otherwise good reports often go wrong.

Dimension filters select or exclude row attributes. Common operations include exact match, contains, regular expression, and in-list matching. Metric filters are evaluated against aggregated values, similar to a spreadsheet filter applied after the totals are calculated.

Write the logic in plain language before translating it:

Then ask the AI to restate the final boolean logic before it runs the report. This creates a small but valuable review gate.

Use this as a starting point, then replace the bracketed details:

HireOtto can handle the mechanical reporting sequence inside Claude, ChatGPT, or another compatible AI client: discover accessible GA4 properties, inspect property configuration, search property-specific metadata, check a Core dimension-and-metric combination, run filtered and ordered reports with multiple date ranges, and return an inline sample or CSV export. It can also surface coverage details so the reviewer can distinguish a preview from a complete result.

The current GA4 connection is read-only. HireOtto does not save reports in the GA4 interface, change streams or key events, register custom definitions, edit attribution settings, publish tracking changes, or prove that a browser-side event fired. It also should not be presented as a first-class replacement for every GA4 reporting surface, such as Explorations, funnel reports, pivot reports, or audience exports. The marketer still owns the decision, KPI definition, tracking validation, interpretation, and approval of any downstream action.

See the GA4 tools reference and custom reporting guide for supported inputs and limits. Then use the GA4 quickstart and AI-client connection guide to get started.

Not through this HireOtto workflow. It creates an on-demand, read-only Core report and can export the result. Saving or publishing a report in the GA4 interface remains a separate manual workflow with the appropriate GA4 permissions.

Use Core reporting for historical analysis, comparisons, most acquisition questions, and complete exports. Use realtime only for the recent activity supported by the realtime schema. Do not assume a Core field is available in realtime.

Compatibility only checks whether the selected Core fields can be requested together. It does not validate your dates, filters, tracking implementation, business definition, or interpretation.

The GA4 Data API currently allows up to nine dimensions in a Core runReport request. That is a technical ceiling, not a recommended design. Most decision reports are clearer with far fewer.

Inline display and CSV export are separate output controls. The preview stays compact for review, while the export can contain the larger collected dataset. Always compare matching rows with collected rows and check the truncation indicator.

No. It can reveal patterns, comparisons, and segments worth investigating. Causality may require experiment design, CRM evidence, site diagnostics, or other data outside GA4.