GA4 time zone: why yesterday's numbers keep moving
Someone screenshots yesterday's revenue from GA4 at nine in the morning, compares it to the Google Ads console, and gets two different numbers. The instinct is to look for a tracking fault. Usually there is no fault. The two systems put the day boundary in different places, and no amount of tag debugging will reconcile them.
Time zone is the least glamorous setting in analytics and one of the few that quietly decides whether a report is trustworthy. It is also the one setting people change at the wrong moment, for the wrong reason, and then spend a quarter explaining the resulting step in the chart.
One setting decides where your day starts
A GA4 property carries a reporting time zone, chosen when the property is created and editable afterwards in the property settings. That zone sets the day boundary for reports, no matter where the traffic comes from or where the person reading the report sits. Google states the consequence plainly: if a visitor arrives on a Tuesday in their own time zone but it is still Monday in the property's zone, the visit is recorded as having happened on Monday.
That single rule explains most cross-platform arguments. A team in Warsaw and a finance lead in New York looking at the same property see identical daily numbers, because neither of them is reading their own clock. They are reading the property's clock. The moment a second platform enters the comparison, with its own zone, the two clocks separate.
Daylight saving adds a wrinkle worth knowing before it bites. If the chosen zone observes daylight saving, Analytics adjusts with the clock change automatically. Google offers Greenwich Mean Time as the alternative for properties that should not shift twice a year. Properties spanning several countries often pick a fixed offset for exactly this reason, so a European and an American market land in the same buckets all year rather than drifting apart by an hour for part of it.
What actually happens when you change it
Changing the property zone is not a display preference. Google's documentation is explicit that the change only affects data going forward, and warns about two visible effects: shifting the clock forward leaves a flat spot, and shifting it back produces a spike, because one stretch of hours is either skipped or counted twice at the boundary.
Three further details matter in practice.
- Reports may continue to show the old time zone for a short period after the change, until Analytics servers have processed it. Anything you read in the first hours after editing the setting is unreliable.
- Google advises against changing a property's time zone more than once in a day, because the second change may not be processed as expected.
- Editing property settings requires Editor or above, so the change is at least gated behind a permission most stakeholders do not have.
The practical conclusion is that a time zone change is a migration, not a correction. If the zone is wrong by a wide margin, fix it once, note the date, and treat the boundary as a break in the series. An annotation on the date does more for the next analyst than any explanatory email. If the zone is merely inconvenient, leave it and adjust the comparison instead, because a stable series with an awkward boundary beats a clean boundary with a hole in front of it.
Google Ads runs on a clock you cannot change
This is where most daily mismatches come from. A Google Ads account's time zone is fixed when the account is created and cannot be edited afterwards. Google's current documentation is blunt about it: time zone changes are no longer supported in standard serving accounts, and the only route to a different zone is a new account.
Manager accounts get one concession. Google can reset a manager account's time zone once, on request from an Admin user, and only eastward, so a manager on GMT can move to India Standard Time but not the other way. Serving accounts get nothing. Accounts created without a deliberate choice default to Pacific Time, which is the most common accidental setting in the platform.
What the setting controls is reporting. Reports and statistics follow the account time zone, and any report segmented by time is bucketed on that clock. Billing pages are the exception and always display in Pacific Standard Time, which is a separate source of confusion when someone reconciles spend against invoices.
Zones are limited to whole and half hour offsets from GMT. A market on a quarter hour offset cannot be represented, so the account has to pick a neighbouring zone and live with the difference.
The operational lesson arrives too late for most accounts. The zone is chosen during setup, usually by whoever clicked fastest, often an agency setting up a client account from a different country. Whatever gets picked then is permanent for the life of the account. Getting it right at creation costs thirty seconds and cannot be fixed later without losing history.
The one hour gap, worked through as arithmetic
Take a property set to Central European Time and an Ads account created in London. For most of the year the London clock runs an hour behind, so the final hour of the London day lands after midnight in the property's zone. A conversion at 23:30 London time is 00:30 the next morning in Warsaw, so it is booked to 14 November in Google Ads and to 15 November in GA4.
If roughly a twentieth of the site's conversions happen in that final hour, then the two systems will disagree by up to five percent on any single day, even though both are counting the same events. Weekly and monthly totals agree almost exactly, because the same events are counted on both sides with a one day offset inside the period.
That gives a clean diagnostic, and it is worth remembering: a daily discrepancy that disappears when you widen the date range is a boundary problem. A discrepancy that persists at every range length is a tracking problem. Run the weekly number before you open Tag Manager.
The gap is not always the friendly one hour. A London property compared against a New York account sits five hours apart for most of the year, and four hours for the few weeks when the two countries change their clocks on different dates. Anyone comparing those two numbers in the changeover window is comparing days with a different offset than last month, which is why spring and autumn produce a flurry of false alarms.
Search Console and BigQuery are separate clocks again
Search Console does not follow the property zone. Google's own page on discrepancies between Search Console and other tools lists time zones as a cause and names Pacific Time as the clock its daily figures use, with the flat warning that if other systems use different zones, daily views will not match exactly. For a UK site that offset is eight hours, and for a Polish property it is nine, baked into every daily comparison of organic clicks against organic sessions.
BigQuery carries both clocks in the same row, which is where the expensive mistakes happen. In the export schema, event_date is the date the event was logged, in the registered time zone of the property. event_timestamp is the time in microseconds, in UTC, when the event was received by Analytics. Two date fields, two different calendars, in one table.
The daily tables are named events_YYYYMMDD, and a row sits in the table named for its event_date. A query that groups by DATE(TIMESTAMP_MICROS(event_timestamp)) is therefore grouping on UTC days, and evening traffic from a European property lands in the following bucket. The totals still look plausible, which is what makes it dangerous. Group on event_date and the numbers line up with the interface. Where hour level detail is needed, convert the timestamp into the property's zone by name rather than relying on the server default.
The export also carries event_original_occurrence_timestamp, which Google documents as the original time on the device, populated for events ingested after the standard 72 hour processing window. It is the field that lets a late arrival be placed where it actually happened rather than where it was received.
Why yesterday is still changing this morning
Day boundaries explain the difference between systems. They do not explain a number that moves on its own. That second effect is latency, and the export documentation states it without hedging: devices do not always send data on the day the event was triggered, so Analytics updates the daily tables with events for those dates for up to three days afterwards. Events keep their correct timestamp when they arrive late. Events arriving after that window are not recorded at all.
Three days is the number to remember. A report on yesterday is a draft. A report on a day older than the late arrival window has stopped moving, which is why anything contractual, anything sent to a client as a monthly figure, and anything compared against an invoice should be pulled from closed days.
A reporting convention that survives contact with a client
None of this is hard once it is written down, and almost all of it is invisible until the first argument. The convention that holds up looks like this.
- Name the clock. Decide that the property time zone is the commercial day, write it in the measurement plan, and say it in the first reporting call. Every daily figure in every report then has one definition.
- Set the ad account zone at creation. Before an account exists, agree its time zone with the person who will read the reports. After creation the decision is permanent.
- Align new properties to the ad account where ads reporting is the main use, or accept a documented offset and stop treating daily parity as a goal.
- Compare closed days. Use a three day lag for anything that has to be exact, and let yesterday be approximate.
- Group on
event_datein BigQuery. Reserve UTC timestamps for hour level work, and convert with the property zone named explicitly. - Annotate zone changes and DST. A note on the date saves the next analyst an afternoon.
- Check the boundary before the tags. Widen the date range first. If the gap disappears, nothing in the container needs fixing.
The pattern behind all seven is the same. Two systems agreeing on a number requires them to agree on three things: what counts as an event, how long they wait for it, and where the day starts. Most reconciliation work focuses on the first and assumes the other two. The third is a dropdown in the property settings, a dropdown that cannot be changed in a live ad account, and a column name in BigQuery that looks almost identical to the one next to it.
Reports that disagree every single morning? Book a free tracking audit and we will check the property zone, the ad account zone and the BigQuery queries side by side, then tell you which differences are real.