GA4 explorations and Looker Studio at a glance
Explorations sit inside the analytics property, in Explore. Google's help page on explorations describes them as a set of techniques that go beyond the standard reports, and the explorations tutorial builds each one from a question: free-form tables, funnel and path explorations, segments and their overlap, and the user explorer. They read event-level data within the property's retention setting, they can use any registered custom dimensionA GA4 reporting field created by registering an event, user or item parameter so it appears in reports and explorations., and they carry a sampling indicator that says whether the result rests on all the data or on a share of it. An exploration is private to the person who built it until they share it, and sharing makes it visible, read-only, to other users of the property, who can duplicate it to make changes of their own.
Looker Studio sits outside the property. It holds no data of its own: as the data sources and connectors guide explains, a connector fetches rows, a data source shapes and secures them, and a report draws them. The GA4 connector is one of many, so a page can place analytics beside ad platforms, a CRM export or a spreadsheet of targets. Access to the report, the credential mode of each data source, link settings, embedding and scheduled email delivery are all decided in Looker Studio, and the Looker Studio Help centre documents each one.
The standard reports sit between the two. The reports and dimensions guide puts the division of labour in one line: standard reports for recurring questions with a known shape, explorations for one-off investigation, the export for anything the interface refuses to do. Looker Studio is where a recurring question goes once its audience reaches beyond the people who work in the property.
Compare table
GA4 explorations vs Looker Studio reports
| Row label | GA4 explorations | Looker Studio reports |
|---|---|---|
| Built for | Investigating a question: where a journey loses people, how groups differ, what one user did | Reporting an answer: the same pages read by the same people, on a rhythm |
| Who builds it | An analyst or marketer with a role on the property that allows explorations | A report editor, who may never work inside the property |
| Who reads it | The creator, and property users it is shared with, read-only | Anyone given access to the report, with or without access to the property |
| Data sources | One GA4 property, at event level within its retention | GA4 through the connector, plus other connectors, spreadsheets and warehouses, blended or side by side |
| Techniques | Free form, funnel, path, segment overlap, user explorer, cohorts, segment comparisons | Scorecards, time series, tables, charts, controls and calculated fields over whatever the source returns |
| Sampling and thresholds | May be sampled above the documented quota, with an indicator; thresholded rows withheld | Thresholded rows and the (other) row can arrive through the API; caching decides how fresh a figure is |
| Sharing | Private until shared; shared copies are read-only and can be duplicated | Named people, groups and link settings; owner's or viewer's credentials decide whose data appears |
| Scheduling | No native schedule; export by hand as a file, spreadsheet or PDF | Scheduled email delivery of chosen pages as a PDF with a link to the live report |
| Version control | One creator edits; others work on duplicates that drift apart | One shared report and reusable data sources, so a field fixed at the source changes every report built on it |
| Typical failure | A sampled figure quoted as exact, or the same exploration rebuilt every Monday | A dashboard that reads differently from the interface, or a report shared further than its data should go |
The rows that decide most cases are "who reads it" and "scheduling". If the answer must reach someone who does not work in the property, or must arrive without anyone rebuilding it, it belongs in Looker Studio. If the question is new, its shape is unknown and the person asking is the person building, an exploration gets there faster and with techniques Looker Studio does not reproduce, such as path and overlap analysis.
When to choose GA4 explorations
Choose an exploration when the question is new and the answer will shape a decision once. A funnel to find which checkout step loses mobile visitors, a path ending on a key event to see what led to it, an overlap between two segments before an audience is shared to advertising, a user timeline to see what a test session actually sent: each is a one-off investigation, built and read by the same person, and the exploration techniques fit them exactly. The explorations tutorial's decision table matches the question's shape to the technique.
Choose an exploration, too, when the question needs data that a connector would hide. Explorations read event-level data inside the property and can segment on sequences of events; the Looker Studio connector returns rows for the combinations of dimensions and metrics the product will answer, so a behaviour that exists only as a sequence is easier to find in Explore. Where the result might be sampled, the indicator says so, and the tutorial's advice is to narrow the range, simplify the segments or reduce the breakdown and to note any sampling on the exploration itself.
An exploration is also the right first draft of a future report. Building the question as a free-form table, with the dimensions and metrics it needs, settles the definitions before anyone designs a page. The comparison of sessions and users is a useful check at this stage, because a rate built on the wrong denominator in an exploration carries straight into every report built from it.
When to choose Looker Studio
Choose Looker Studio when other people will read the answer, more than once. The dashboard design guide sets the bar: a page earns its place when a person who was not involved in the data can answer a question and act on it without asking anyone. That calls for one question per page, the headline KPIA metric tied directly to an objective. with its target, a trend, the drivers and a breakdown, and none of it is what an exploration is designed to present.
Choose it when the view needs more than one source. Cost from the ad platforms, leads from analytics and targets from a spreadsheet can sit on one page, and the comparison of Google Search Console and GA4 is an example of two Google sources whose figures answer different questions and are often read together in one report. Choose it when the audience has no business in the property: with owner's credentials a sales director can see the cost per lead page without any analytics access, and the guide to sharing and embedding reports explains why that convenience needs the same care as any other form of sharing. And choose it whenever the view should arrive by itself, because scheduled delivery is a Looker Studio feature with no equivalent in explorations.
Looker Studio is not a replacement for investigation. A dashboard answers the questions it was designed for and shows little beyond them. When a monitored figure moves and nobody knows why, the investigation goes back into Explore, or to the export for questions the interface cannot answer.
Decision table
| Situation | Use | Why |
|---|---|---|
| A new question about where a journey loses people | GA4 exploration (funnel) | One-off, built and read by the analyst |
| What visitors did before reaching a key event | GA4 exploration (path) | Paths have no equivalent chart in a report |
| The same figures for a manager every Monday | Looker Studio with scheduled delivery | Explorations cannot be scheduled |
| Analytics beside ad cost and a target sheet | Looker Studio | Several connectors on one page |
| A stakeholder with no property access needs a view | Looker Studio | Report access and credentials are set outside the property |
| Checking whether two audiences overlap before sharing one | GA4 exploration (segment overlap) | A one-off check inside the property |
| Debugging what one test session sent | GA4 exploration (user explorer) | Sensitive, individual-level, never for a report |
| A dashboard disagrees with the interface | The like-for-like check, then fix the report | Most gaps lie in how the question was asked |
| A question that needs joins, exact counts or long history | Neither; the BigQuery export | Beyond what either interface does |
Moving a question from an exploration to a report
The explorations tutorial's own advice is that a view that should arrive weekly belongs in Looker Studio on the GA4 connector, with the exploration kept as the analysis behind it. Moving a question well is more than rebuilding a table.
Start by writing the question as the dashboard design guide asks, as a sentence a non-specialist understands, and place it in the plan it serves. The measurement plan framework runs from objective to goal, KPI, target and segment; a question that cannot be placed there is probably a one-off and can stay in Explore. Then check that every field the exploration used exists through the connector, because the connector decides which fields and combinations it returns. A dimension with many distinct values can also collect an (other) row once it passes its CardinalityThe number of distinct values a dimension or field holds; high-cardinality dimensions, such as page paths with IDs, can exceed report row limits. limit, and the reports and dimensions guide covers where those rows come from.
Build the page with fixed dates and a plain scorecard for the headline figure, and check it against the standard report for the same dates before adding anything else. If the two disagree, the page on Looker Studio numbers not matching GA4 lines them up like for like and names the cause. Only then add the trend, the drivers and the breakdown, set the schedule and decide the credential mode. Record on the page where each figure comes from, so a reader does not compare it with a sampled exploration and conclude that one of them is broken.
Some questions outgrow both tools: several platforms joined on one key, history beyond the retention window, exact counts where the interface samples or thresholds. The guide to reporting and business intelligence describes when that move is worth making.
An invented tea shop moves a weekly question out of explorations
Teffulin Teas, an invented online tea shop, launched a monthly tea subscription and asked its analyst how many visitors who viewed the subscription page went on to sign up, by device. She built a closed funnel exploration over four weeks: of 6,200 users who viewed the page, 1,140 started sign-up and 410 completed, with mobile completing at about half the desktop rate. The finding led to a shorter mobile form. The owner then asked for the same funnel every Monday. For five weeks the analyst duplicated the exploration, changed the dates, exported a PDF and emailed it, spending about 40 minutes each time; twice the export was sampled and once the dates were wrong. She rebuilt the view in Looker Studio as one page: a scorecard of sign-ups against a target of 120 a week, a weekly trend, sign-up rate by device and a table by landing source, with a fixed comparison to the previous week. The first scorecard read 104 for a week the standard report showed at 104, so the page was scheduled to the owner at eight on Monday mornings. The funnel exploration stayed in Explore for the next question about why mobile sign-ups lagged.
Common mistakes
- Rebuilding the same exploration every week. A recurring view belongs in a scheduled report, with the exploration kept for analysis.
- Quoting a sampled exploration as exact. Check the indicator before any figure leaves the screen.
- Using a dashboard to investigate. A report shows what it was designed to show; a new question needs Explore.
- Sharing a report further than its data should go. Owner's credentials plus a permissive link expose the owner's view to everyone who has it.
- Assuming the connector returns everything Explore can see. Sequences, paths and overlaps do not carry across.
- Comparing a report with an exploration without lining them up. Fixed dates, one metric and no filters on both sides first.
- Building a report for a question with no place in the plan. If no decision depends on it, it does not need a page. The comparison library lists other pairs that are easy to confuse.
