Guide 66 · Reports · GA4 + Tag Manager
Build a cohort exploration: retention by the week users first arrived
A cohort exploration groups users by when they first visited and tracks how many return in later weeks. It answers whether new users stay or vanish.
After step 6, the cohort grid shows each acquisition week's users and how many returned in weeks one to four, with retention computed per cohort.
Download the report reference: the Cohort exploration reference (JSON) (1261 bytes, MD5 5726300ffff7f40344a0dcd7cf2ace44). It holds the report or exploration definition and the worked example with the recomputed figures shown on this page.
What you need
- Viewer access or above on the property to build.
- The entry point: Explore in the left navigation, then the cohort exploration template.
- Enough history for the cohorts to mean something — at least four weeks of data. A cohort grid over five days of data shows one column and no pattern.
The trace
- Open Explore and pick the cohort template. Click Explore in the left navigation, then create a blank exploration and change the technique to Cohort exploration, or pick the template. The canvas shows a cohort grid.
- Set the cohort and the date range. The cohort is the group of users who share a trait — usually the date of their first visit. Set the date range to cover at least four full weeks, so each cohort has weeks to return in.
- Pick the metric the grid tracks. In Tab Settings, choose the metric to follow across the weeks — for example, active users, or sessions. The grid shows that metric for each cohort in each week after acquisition.
- Read down the columns, not across one row. Each row is one cohort; each column is a week after first visit. Read down a column to compare cohorts at the same age — that is the comparison that says whether retention is improving. Reading across one row only describes one cohort.
- Check the retention period before you trust an old cohort. By default a property retains 2 months of data. A cohort from six months ago cannot be read if retention was never raised; Admin, then Data collection and modification, then Data Retention is where the period is set. A cohort grid that ends early is usually a retention limit, not a retention failure.
- Save it under the question it answered. Save the exploration with a name that states the cohorts, for example, Weekly acquisition cohorts — retention to week 4.
Dead ends
DEAD END — READING A COHORT GRID OVER FIVE DAYS. Retention needs weeks. A grid with one column of data cannot say whether users return.
DEAD END — BLAMING A COHORT FOR A RETENTION LIMIT. A cohort that disappears after two months is the default retention setting, not user behaviour. Check Data Retention before you conclude the users left.
DEAD END — COMPARING COHORTS AT DIFFERENT AGES. A four-week-old cohort and a one-week-old cohort cannot be compared on return counts. Compare cohorts in the same week column.
End state
A cohort exploration turns first visits into a retention story. The worked example below recomputes week-4 retention for four weekly cohorts, so the grid's columns can be checked against their own division.
Worked example — stated inputs, recomputed results. The cohort sizes below are stated inputs for the example. Retention is returned users divided by cohort users.
| Cohort (first visit) | Users | Returned in week 4 | Retention |
|---|---|---|---|
| Week of 12 Oct | 840 | 252 | 30.0% |
| Week of 19 Oct | 910 | 291 | 32.0% |
| Week of 26 Oct | 760 | 190 | 25.0% |
| Week of 2 Nov | 880 | 308 | 35.0% |
| Total | 3,390 | 1,041 | 30.7% |
Sources
- Analytics Help — Get started with Explorations — cohort exploration gains insights from the behaviour of groups of users related by common attributes; default retention is 2 months
- Analytics Help — Confirm that you are collecting data — many reports and explorations can take 24 to 48 hours to process data
Checked 11 November 2026. The click paths above are the ones in Google's help pages, opened or read 11 November 2026 (the Explorations and detail-report pages were opened in full; the report, library, navigation and thresholds pages were read from Google's help results the same day, after the browser hit a rate limit). No step in this guide was run inside a live Analytics property, because that needs a Google sign-in and this site does not sign in as anyone. The worked-example figures on this page were recomputed in Python with decimal arithmetic, and the downloadable JSON file carries the same recomputed figures. Google moves menu items without warning; if a path on this page no longer matches, the linked help page is the current map.