First, line the two up like for like
Pick one number that disagrees, such as sessions for last week, and reduce both sides to it. In the GA4Google's event-based analytics platform. interface, open a standard report, not an exploration, set the exact start and end dates, remove comparisons and filters, and note the figure. In Looker Studio, add a plain scorecard on a blank page, using the same data source, the same metric, a fixed custom date range and no filters. Confirm the data source points at the same property: a staging or old property with a similar name is a common culprit, and the source settings, described in the data sources and connectors guide, show which one it reads.
If the bare scorecard matches, the gap is in the original chart: its dimensions, its filters, its calculated fields or its blend. If it does not, the gap is in the dates, the cache or the property. Either way you have halved the search. The two tools are built for different jobs, and the comparison of GA4 explorations and Looker Studio explains which questions belong in each, which is why the like-for-like check uses a standard report rather than an exploration. Disagreements between the property and an ad platform are a different matter, covered by the guide on why platforms and analytics disagree, and a gap against raw exported rows belongs to the sibling page on a BigQuery export that does not match GA4.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Totals differ by roughly a day's traffic, or shift when the report is opened later | The chart's date range control, comparison setting and the time zones of every source | Different date range, comparison or time zone | Use fixed, matching ranges; align time zones before comparing |
| Users or sessions inflated when broken down, or blank against some dimensions | Whether the chart sums users across rows or pairs session metrics with event-scoped dimensions | Mixing dimensions and metrics of different scopes | Pick dimensions of the metric's own scope; report users as a single total |
| Breakdown totals below the scorecard total; an "(other)" row appears | Threshold behaviour and cardinality of the dimension in the chart | Thresholds, sampling or the (other) row in the API | Widen the range, simplify the breakdown, or move to the export |
| Blended totals above or below either source alone | Join key, join type and grain of each blended table | Blends joining on the wrong key or join type | Fix the key and grain; validate blended totals against each source |
| Rates such as engagement or conversion differ from the interface | Whether the field divides sums or averages per-row ratios | Calculated fields re-aggregating ratios | Rewrite as a sum divided by a sum at the source |
| Yesterday or today differ; a refresh changes the number | Data freshness setting and whether the source is an extract | Cached or extracted data out of date | Refresh, shorten freshness where needed, or schedule the extract |
Fix, verify and prevent each source of the gap
Different dates on each side
A date control that defaults to a rolling period, a comparison period left on, a chart with its own date range overriding the page, or a blended non-Google source reporting in another time zone all shift totals by part of a day or more.
Fix: set a fixed custom range on both sides for the comparison, turn off comparisons, and align time zones of any other sources, using the source settings described in the data sources and connectors guide.
Verify: the bare scorecard matches the interface for a closed period.
Prevent: default report date controls to closed periods such as last complete week, and label the time zone on the page.
Scope mixed in the chart
A metric only makes sense with dimensions of its own scope. Session counts split by an event-scoped custom dimension, users summed across a daily table, or "Session source" in one chart and the attributed "Source" in another produce numbers that are defined and still wrong for the question.
Fix: choose dimensions that match the metric's scope, show users as a single total for the period, and read the scope table in the reports and dimensions guide.
Verify: the chart's total equals the bare scorecard for the same range.
Prevent: name fields with their scope in the data source so the next editor sees it.
Rows withheld or rolled up
Looker Studio receives the property's data through the analytics API, which applies thresholds when the reporting identity includes signals or modelled data, and folds high-cardinality values into an "(other)" row. A breakdown by city or page path can then total less than the scorecard, or hide the rows you want.
Fix: widen the date range, drop demographic dimensions, reduce cardinality, or consider the device-based reporting identity, as the reports and dimensions guide explains; for exact detail, move the chart to a warehouse source built from the export.
Verify: the breakdown's total now matches, or the remaining gap equals the withheld rows.
Prevent: design recurring charts at a grain that clears thresholds.
A blend that drops or multiplies rows
Blends join tables inside the report. A key that differs in case or type drops rows; a grain mismatch makes one side's rows repeat for each match on the other, inflating its metrics; an inner join hides whatever did not match.
Fix: choose the spine table, left-join the others on a normalised key at the same grain, and follow the validation step in the blending tutorial.
Verify: the spine's blended totals equal its own totals exactly, and unmatched rows are explained.
Prevent: move any blend used on more than one page into a single upstream table.
Ratios averaged instead of recomputed
Engagement rate, conversion rate and revenue per user are ratios. A calculated field that divides per row and lets the chart average the result weights every row equally, so the figure drifts from the interface as soon as the chart is grouped.
Fix: rewrite the field as an aggregated numerator divided by an aggregated denominator in the data source, as the calculated fields tutorial shows.
Verify: the ratio for one day and one channel matches the interface to rounding, and still matches when regrouped.
Prevent: keep ratio definitions at the source with field editing in reports switched off.
A cache or extract behind the property
Results are cached according to each source's freshness setting, and extract sources serve a stored snapshot until their next scheduled refresh. Recent days in particular can differ from a live interface view.
Fix: refresh the report to bypass the cache, check the extract's schedule, and set freshness to match how the data actually changes, as the performance and caching guide describes.
Verify: after refresh, the closed-period figure matches.
Prevent: label pages fed by extracts with their refresh time and keep the latest day off headline scorecards.
Prevention: one definitions sheet for every dashboard
Most mismatches come from definitions living in charts rather than in one agreed place. Keep a definitions sheet listing, for each headline measure (KPIA metric tied directly to an objective.), the property, metric, scope, filters, date convention, time zone and the data source that implements it, with the test figure that proves it. Build reusable data sources and create calculated fields there, never in individual charts. Default every report to closed periods and show the date range and time zone on the page. When a new chart goes live, compare one figure with the interface and record the result in the sheet. Note which charts use attributed dimensions, because the reporting AttributionAssigning conversion credit to touchpoints. model moves them while session dimensions stay put, and a reader comparing the two will otherwise think the dashboard is broken. The dashboard launch QA checklist makes matching headline totals against the source platform one of the checks every new dashboard passes before stakeholders see it.
A weekly dashboard that overstated users
A marketplace's weekly dashboard shows 61,800 users for the last week while the interface shows 48,300. A bare scorecard on the same source reads 48,300, so the property and dates are fine. The original chart is a table by day with a total row: it sums daily users, counting a visitor who came on 3 days three times. A second gap sits in the engagement rate, 57% on the dashboard against 52% in the interface, from a calculated field averaging daily rates. The analyst replaces the total row with a period scorecard, rewrites engagement rate as engaged sessions over sessions at the source, and records both definitions in the sheet. The next week's dashboard matches the interface on both figures.
Common mistakes
Comparing a dashboard with an exploration instead of a standard report. Summing users across days. Defining a ratio in a chart. Trusting an inner join because the totals look tidy. Reading today's figure from a cached source as final.
