TRACKING SPECIALIST — BARIS ASA · LONDON
Synctag
Start the check → TRACKING SPECIALIST — BARIS ASA · LONDON

READING ROOM › SYMPTOM FILES › SYM-056

SYMPTOM FILE SYM-056 FILED 10 NOV 2026 · REVISED 10 NOV 2026

GA4 BigQuery export doesn't match GA4 reports: what is expected and what is a fault.

ALSO FILED AS
  • “ga4 bigquery export doesn't match ga4 reports”
  • “bigquery users different from ga4”
  • “bigquery sessions don't match ga4”
  • “ga4 bigquery vs ui numbers”
  • “events_intraday vs events_ table different”
  • “why is bigquery data different from google analytics”
  • “ga4 bigquery export missing data”

Baris Asa, tracking consultant, London · Last checked: 10 November 2026

BigQuery counts the events GA4 collected. GA4 reports estimate, can merge one person’s devices, and include modelled data, so the two will never match exactly. Before you call a gap a fault, compare like with like: active users (people who engaged, not everyone who triggered an event), sessions counted as distinct user plus session ID, days cut in the property’s time zone, and data older than 72 hours. In my experience the first of those often does most of the work.

What is left after the four is the gap worth explaining. If it is still above roughly ten per cent, that is my judgement, not Google’s, and it is worth chasing: send me the date range and the two numbers, with the SQL if you have it, and I’ll tell you which of the causes below it looks like. If you have no analyst, ask your agency one question: “Are these active users, and is the date range closed?”

Google’s own article says the interface and the export aren’t expected to be fully reconcilable. It gives plus or minus 1.63 per cent on sessions, at 95 per cent confidence, as an example of the report’s precision, and no figure for a normal gap. Mine is an observation only: after the four fixes, a few per cent is usual.

The four comparison errors

QUESTIONWHAT TO MATCH
UsersActive users, not all users
SessionsUser ID plus session ID, whole range
Day boundaryReporting time zone, not UTC
FreshnessDates older than 72 hours
SAME NAME, DIFFERENT COUNT · what to match before you compare · from Google's BigQuery export and UI-versus-export articles

Users. The report counts active users. A distinct count of user_pseudo_id over every event is nearer total users, which includes people who never engaged. Google’s article limits its example query to is_active_user.

SELECT COUNT(DISTINCT user_pseudo_id) AS active_users
FROM `project.dataset.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20261001' AND '20261028'
  AND is_active_user

Sessions. Count unique combinations of user_pseudo_id and ga_session_id across the whole range. Don’t count per day and add the days up: a session that crosses midnight is then counted twice.

SELECT COUNT(DISTINCT CONCAT(
  user_pseudo_id, '.',
  CAST((SELECT value.int_value FROM UNNEST(event_params)
        WHERE key = 'ga_session_id') AS STRING))) AS sessions
FROM `project.dataset.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20261001' AND '20261028'

If you send user_id, use COALESCE(user_id, user_pseudo_id) in place of user_pseudo_id. Events with no ga_session_id drop out of this count.

Time zone. Google says event_date is in the reporting time zone and event_timestamp is UTC in microseconds. If you group by the timestamp, convert it first, for example DATE(TIMESTAMP_MICROS(event_timestamp), 'Europe/London'). Google’s export help adds that changing the property’s time zone will impact the BigQuery export.

Freshness. Google updates the daily tables for up to three days after the event date and doesn’t record later arrivals. Its article says to compare data older than 72 hours.

Here is an illustrative example, not client data. In it, the first correction closes most of the gap. How much each step moves the total depends on the property, and these figures are not a benchmark.

STEPBIGQUERYGA4 REPORTGAP
As found: all users63,40051,900+22%
Active users only56,10051,900+8%
Closed range, time zone matched55,00051,900+6%
ILLUSTRATIVE · one gap, two corrections · users, 28 closed days · illustrative worked example, not client data

Part of the six per cent that is left is what the next section covers. Whatever is still unexplained after that is for the export checks further down.

What the reports have that the export doesn’t

This part is by design. You can’t fix it, only label it.

  • Modelled data. Google’s comparison lists key event modelling and behavioural modelling as in reports and not in BigQuery. With consent mode on, the export holds the cookieless pings, and Google’s article says the pings can carry a different user_pseudo_id for each session. Because of behavioural modelling, it says, reports might show fewer active users than the export. So where consent denial is high, expect BigQuery to run higher and don’t chase that part of the gap. Google says user_id is exported whatever the consent status, so sending it narrows the gap. For what consent mode recovers, see Consent Mode v2 and your conversion numbers.
  • Google signals. This is Google stitching one person’s phone and laptop together in reports. Google says BigQuery still shows three user_pseudo_id values for one person on three devices, and that reports with signals data will most likely show fewer users.
  • Attribution. Google’s article says Analytics applies its own attribution model at session level. Data-driven attribution is marked as not in BigQuery. A source and medium table built from the export with your own last-click rule won’t match Traffic acquisition. That isn’t a tracking fault. It’s two models.
  • Approximate counting. Active users and sessions in reports are estimates. A COUNT(DISTINCT) in BigQuery is exact, so a small gap either way is expected.

Reports can also hide more than the export: Google’s article says most information that can be thresholded isn’t in the export, and BigQuery keeps the rows GA4 folds into (other) in reports. If your gap is a missing row rather than a different total, start with GA4 data thresholding and R-502 .

Which table are you reading?

Streaming export writes today’s events to events_intraday_YYYYMMDD. Google’s help says it is deleted once events_YYYYMMDD is complete, that streaming is best effort and may contain data gaps, and that it excludes traffic source data for new users and new sessions. A query on today’s intraday table will undercount sources and may undercount events. Compare against events_YYYYMMDD for closed days. Use the intraday table only to see that data is arriving.

The wildcard in the queries above, events_* with a numeric _TABLE_SUFFIX range, leaves the intraday tables out.

Is the export itself missing data?

Only after the four fixes. These are the causes Google documents, and one query to find the day a gap starts.

SELECT event_date, COUNT(*) AS events
FROM `project.dataset.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20261001' AND '20261028'
GROUP BY event_date
ORDER BY event_date

A day with no row, or a count far below its neighbours, is where to look. A day near 1,000,000 on a standard property points at the first cause.

  • The daily limit. Google’s export help says standard properties have a daily export limit of 1 million events. Google’s article warns that when export volume exceeds the limit, the export won’t match the standard reports.
  • Data filtering on the export. The same article says the export won’t match the reports if you’ve set up data filtering for your BigQuery export. Check what was changed and when. A filter added after the report was built explains a gap that begins on one date.
  • A late daily export. Google’s help says the daily export typically runs mid-afternoon in the property’s time zone, but can be delayed until later in the day or the next day. Query yesterday’s table the next morning, not at lunchtime.
  • Collection that changed. If the gap began the day a tag, consent banner or server container changed, the export and the reports may both be right about different things. The same logic on the ad side is in GA4 and Google Ads numbers don’t match.

The check, in order

Give this list to whoever runs your BigQuery. Each step says which symptom it answers.

  1. Use a closed date range ending at least three days ago. (Numbers that keep moving.)
  2. Count active users: is_active_user, then distinct user_pseudo_id. (BigQuery users far above GA4.)
  3. Count sessions as distinct user plus session ID over the whole range. (Sessions off by a few per cent or more.)
  4. Group by event_date, or convert event_timestamp into the reporting time zone. (Daily totals off, weekly totals fine.)
  5. Query events_YYYYMMDD only, not the intraday table. (Today’s sources missing.)
  6. Note Google signals, consent denial and user_id. (A user gap you can’t remove.)
  7. Run the events-per-day query and check the 1 million limit, data filtering and the export’s timing. (A gap that begins on one date.)

The fix that stops it coming back is writing down which source each number comes from. That’s the register in reporting people actually trust, and it is the R-501 fix. It matters most where a dashboard puts BigQuery tiles beside GA4 connector tiles under the same label. In Looker Studio reporting work, the register names the source and exact field behind every tile, so nobody has to guess which of the two they are reading.

Send me the two numbers and the date range, and I’ll say whether the gap is expected or a fault. The SQL helps, but you don’t need it.

Common questions

How big a gap between BigQuery and GA4 is normal?

Google gives plus or minus 1.63 per cent on sessions, at 95 per cent confidence, as an example and no normal gap. In my experience, after the comparison errors are fixed, a few per cent is usual. That’s my observation, not Google’s figure.

Why do my BigQuery numbers change a few days later?

Google updates daily tables for up to three days after the date. Compare data older than 72 hours.

Does the BigQuery export have thresholding or sampling?

Google describes the daily export as raw, unsampled event data, and says most thresholded information isn’t in it. Standard properties have a daily limit of 1 million events.

What is the difference between events_intraday and events_ tables?

The intraday table holds today’s events and is deleted when the daily table is complete. Streaming is best effort and may have gaps. Compare against the daily table.

Sources

Read on 10 November 2026:

  • Google Analytics Help, Set up BigQuery Export (daily export of raw, unsampled events; 1 million events per day for standard properties; typically mid-afternoon in the property’s time zone, can be delayed to the next day; streaming best effort with possible gaps; intraday table deleted when the daily table is complete; new user and session traffic source excluded from streaming; a time zone change will impact the export; consent mode cookieless pings appear in the export)
  • Google Analytics Help, BigQuery Export schema (daily tables updated up to three days after the event date; event_date in the registered time zone; event_timestamp in UTC microseconds; user_pseudo_id; is_active_user)
  • Google Analytics Help, Reporting surfaces comparison (modelling and data-driven attribution not in BigQuery; no sampling for BigQuery)
  • Google for Developers, Bridge the gap between the Google Analytics UI and BigQuery export (active versus total users; plus or minus 1.63 per cent session precision at 95 per cent confidence; Google signals and separate user_pseudo_id values; cookieless pings with a different user_pseudo_id per session and fewer active users in reports because of behavioural modelling; user_id exported regardless of consent; time zone; session-level attribution; thresholded information mostly absent from the export; (other); 72-hour rule; data filtering or export limit exceeded means no match; session counting by user and session ID across the range)

← back to the reading room