Two users, one session: counting GA4 sessions in BigQuery with ga_session_id alone

The GA4 export has no sessions table and no session_id column. What each event carries is a parameter, ga_session_id, inside event_params. The obvious way to count sessions is to count its distinct values. The problem is that this ID doesn’t mean what its name suggests.

What an undercounted session looks like

A query with COUNT(DISTINCT ga_session_id) for a day or a month:

• It gives fewer sessions than it should. Two events from different users with the same ga_session_id count as a single session.
• The gap grows with traffic: the more users, the more chances that two of them share a value.
• Metrics built on that count (conversion rate per session, pages per session, revenue per session) are off too, in the other direction.

Why ga_session_id isn’t enough

ga_session_id identifies a session within a user, not across the whole property. Two different users can carry the same value. Google’s own method, explained on the Google Analytics developers blog, is to count the unique combinations of user_pseudo_id (or user_id) and ga_session_id.

Two related mistakes are worth knowing:

• Adding up sessions per day. A session that crosses midnight has events on two days. If you count sessions day by day and then add the days, it counts twice. Count the combinations over the whole range in one query.
• Expecting the exact GA4 figure. The GA4 interface estimates sessions with the HyperLogLog++ algorithm (Google gives a precision of about ±1.63% at 95% confidence), and its reports can include modeled data and thresholds that the export doesn’t have. BigQuery gives the exact count of what was exported, so a small gap is normal even with the right query.

How to find a wrong session count

1. Search your queries, views and dashboards for COUNT(DISTINCT ga_session_id) or for a ga_session_id used as a session key on its own, also in joins.
2. In the same query, compare the count by ga_session_id alone with the count by the user_pseudo_id + ga_session_id pair. If the second one is higher, the first one is merging sessions.
3. Look for monthly totals built by adding daily session counts.

How to count sessions

Count the unique pair of user and session ID, over the whole range at once:

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 `analytics_123.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20261001' AND '20261028'

• CONCAT needs text, so the session ID (an integer in int_value) is converted with CAST. The separator avoids two different pairs producing the same string.
• If you use user_id for signed-in users, use the same rule: user and session together.
• For a daily breakdown, count per day knowing that sessions crossing midnight appear on both days; for the total of the period, don’t add the days.

Documentation: Bridge the gap between the GA UI and BigQuery export and BigQuery Export schema.

How to check sessions are counted right

• The count by user_pseudo_id + ga_session_id is equal to or higher than the count by ga_session_id alone, and it’s the one your reports use.
• For a closed period, the result is close to the GA4 sessions in the interface, within a small percentage.
• The total for the period comes from a single count, not from adding days.

The user and session IDs in the example are illustrative.