31 October 2026 · 9 min read

GA4 Data Retention: What the Timer Deletes and What It Spares

Open Admin, then Data Settings, then Data Retention. Most properties have never been there. The setting usually sits at two months, it runs without any visible signal, and the first symptom tends to arrive a year later as an exploration with a hole in it, or an Ads revenue chart reading zero for a quarter the client remembers as strong.

The control is not complicated. Its edges are, and a few of them are not where people look. Here is what the setting governs, what it leaves alone, and the places where something else overrules it without asking.

Two controls, not one

Google splits the timer in two. The first governs user-level data, and a standard property gets a choice of 2 months or 14 months. The second governs event-level data, where the options widen to 2 and 14 months, plus 26, 38 and 50 months on 360 (GA4 Data retention).

Key events follow the user-level window rather than the event window. If you have only ever looked at one dropdown, check both, because they sit on the same page and they are not the same choice.

One row ignores your selection entirely. Age, gender and interest data is always kept for two months, whatever period you pick. Worth knowing before someone builds a demographic view on a 14-month window and expects the first half of it to be populated.

Google signals data is capped at 26 months regardless of the property setting, and if the property is set to 14 months the shorter period applies.

What it does not touch

The standard reports keep their aggregates. Google states that the retention setting does not affect aggregated reports, including primary and secondary dimensions, even when you build comparisons inside them. What it does affect is explorations and funnel reports.

That split causes the most common misreading of the feature. A team raises retention to 14 months, runs a year-over-year comparison in Reports, sees a full year, and concludes the setting is working everywhere. The same team opens an exploration on the same date range and finds ten months missing, because the exploration reads event-level rows while the report reads tables that were aggregated before deletion.

If a client question needs a 14-month breakdown and the property has been sitting on two months, no report will answer it. The rows are gone. What is left is deciding whether to change the setting before the next question arrives.

Reset on new activity

Below the retention dropdown there is a switch labelled Reset user data on new activity. Turned on, every new event from a user restarts their clock, so the expiry date becomes the current time plus the retention period. Google's own example: at 14 months, a user who starts a new session every month has an identifier that never expires.

Turned off, the identifier is deleted at the end of the period whatever the user does. That is the stricter setting, and it is the one to reach for when a privacy position says the clock starts at first collection rather than at last contact.

The switch applies to user-level data only. It does not extend event-level rows.

The deletion calendar and the grace window

Deletion is neither instant nor continuous. Data that reaches the end of the period is removed on a monthly cycle. Shorten the window and everything now out of bounds goes at the next monthly pass, not the moment you save.

Lengthening works retroactively, within a limit. The new period applies to data already collected and not yet deleted. That is the one generous edge of the control. A property that spent three years on two months will never recover year one, but it can recover the recent past still inside the old window.

Analytics waits 24 hours before applying a change, which gives you a revert window during which nothing is affected. It also does not pad a date range for you. Ask for 14 months plus a day on a 14-month property and the extra day is simply absent, which reads as a broken exploration more often than as a boundary.

The Ads window that is not your setting

Google Ads metrics inside GA4 carry a rolling 36-month limit of their own, separate from anything in Data Retention. Past that line, affected reports show a banner and affected cards show a data quality indicator.

Chart behaviour is worth knowing before anyone files a bug. At daily, weekly or hourly granularity, line charts show the last 36 months and then zero for Ads metrics on older dates. Switch to monthly granularity on a range that falls on whole calendar months and the history comes back complete. Report tables and pie charts truncate regardless.

So a chart showing zeros for Ads revenue two years back at daily granularity may be behaving exactly as documented, and the same range at monthly granularity may show real numbers. That is a display rule to explain, not a data loss to investigate.

The downgrade nobody approves

A standard property that grows past the event collection limits for its tier becomes Large, and a 360 property can become XL. When that happens the event-level retention drops to two months automatically, and event data older than that becomes inaccessible and is permanently deleted.

Google does not do this blind. It emails every property administrator when the property is on pace to cross the threshold based on weekly event volume, and emails again when the limit is reached and the setting has been reduced. Between those two messages there is a window: cut billable event volume to stay inside the tier and keep the longer retention.

Two practical consequences follow. Anyone who can add tags can push a property into that downgrade, so long debug sessions and duplicated container loads are a retention risk as well as a cost risk. And the warning goes to administrators, who are often not the people building the reports.

Audiences outlive the data behind them

Audience membership and property retention run on separate clocks, and the clocks disagree. Membership duration can be set as high as 540 days from the last time a user qualified. With reset on new activity enabled, Google limits membership duration to 14 months. And once an audience is exported to another product, GA retention settings no longer control it.

The dislocation shows up when someone asks why an audience still holds users whose behaviour the property can no longer display. Membership is a flag attached to a user, not a record you can query. The flag can persist after the events behind it have been deleted.

On a two-month property, an audience with 540-day membership is a list you cannot audit and cannot reconstruct from the data available. That is a governance problem before it becomes a technical one.

A worked example

An agency takes over an ecommerce account in October and starts with a year-over-year comparison for the Christmas period. The comparison in Reports returns twelve months without complaint. The exploration built to break the same revenue down by landing page returns October onward, with everything earlier blank.

Nothing is broken. The property has been on two months since launch, the aggregates cover the full history, and the row-level detail behind the exploration covers the last two months. Raising the setting to 14 months recovers nothing before August and starts the clock properly for next year.

The honest version of that report says so, and dates the coverage. The version that causes trouble presents the aggregate comparison as if it were the same analysis, which is how a client ends up believing the property holds three years of detail it will never hold.

The UK side: this is a privacy decision, not a reporting preference

UK GDPR's storage limitation principle asks for personal data to be kept no longer than necessary. Fourteen months of user-level history has to be justified against that, and the justification belongs in a document rather than in someone's memory.

Three things have to agree. The retention setting itself. The privacy notice, which should state the period the property actually uses. And the consent position, because on a consent-managed property analytics storage is denied for anyone who has not agreed, so the retained rows describe a subset of visitors rather than everyone who arrived.

A longer window is not free and it is not automatically wrong. It is defensible for a property that needs seasonality comparisons and hard to defend for a brochure site with no analysis behind it. What is never defensible is a gap between the setting, the notice and what the business believes is true.

The control needs the Editor role at property level. On an agency account that means someone outside your team can change it, so record the current value and the date you last checked. A retention change is one of the few configuration edits that destroys history with no undo.

Six checks, ten minutes

  1. Read both dropdowns at Admin, Data Settings, Data Retention, and note the user-level and event-level values separately.
  2. List the reports the team actually relies on. Explorations and funnels are the ones the timer reaches.
  3. Decide the reset on new activity switch deliberately and write down the reason.
  4. Check the property tier against the event limits, and ask administrators whether a warning email has arrived.
  5. Compare the configured period to the privacy notice and to any internal policy that mentions analytics history.
  6. If the business needs row-level history beyond the window, confirm a BigQuery export exists and check its table expiration once.

None of this is difficult. It gets skipped because the control is buried, it does nothing visible for a year, and the cost arrives as an empty exploration nobody can refill.

Not sure how much history your property actually holds? Book a free tracking audit and we will check the retention settings, the Ads limits and the export coverage against the questions your reporting has to answer.