The export and the reports come from the same events, yet their numbers rarely match exactly. The usual reasons:
- Estimates: reports count users and sessions with an estimating algorithm (HyperLogLog++);
COUNT(DISTINCT ...) in SQL is exact. Expect small differences.
- Late events: GA4 can update a day for up to 72 hours as late events arrive, in reports and in the daily table.
- Consent: in advanced consent mode, events from visitors who declined arrive in the export without a
user_pseudo_id, with privacy_info.analytics_storage set to No.
- Data filters: an Active filter removes events from both; a filter in Testing only marks them.
- Retention limits explorations, never the export, which keeps everything for as long as your project does.
Know these before anyone asks "which number is right?"