Sampling is analysing a subset of data and scaling the result up to estimate the whole, instead of processing every record. In Google Analytics 4 it happens when an exploration asks for more data than Google will process in one go, so the figures you see are estimates rather than exact counts.
How sampling works
GA4 has two kinds of reporting. The standard reports in the Reports section are built from pre-processed tables and are not sampled. Explorations, the custom analysis area, query the raw event data, and when a query is too large GA4 reads only a portion of it. At the time of writing (October 2026), a standard (free) property samples an exploration once the query covers more than 10 million events; GA4 360 has much higher limits.
The data quality icon at the top right of an exploration tells you whether the result is based on all the data or on a percentage of it. Hover over it and GA4 shows how much was used.
Sampling error is the gap between the estimate and the true figure. It is small for large totals and grows as you drill into small groups. A sampled exploration may estimate total sessions within a fraction of a per cent, yet be well out on the conversions from one campaign in one city, because only a handful of real records sit behind that row. The same principle applies outside analytics: a survey of 40 customers carries far more sampling error than a survey of 400.
Sampling is often confused with thresholding, where GA4 hides rows with very few users to stop individuals being identified. Sampling estimates; thresholding withholds.
Why it matters
Most small UK business sites will rarely see it. If a typical visit generates ten to twenty events, 10 million events means hundreds of thousands of sessions in a single date range. A busy online shop running a year-on-year exploration, or a publisher with heavy traffic, can easily cross the line.
The risk is decisions built on estimates of small numbers. If you cut a campaign because a sampled exploration showed it produced three sales, you may have cut it on a figure that was really six, or one. Sampled numbers also will not match the standard reports exactly, which leads to wasted hours reconciling two “correct” figures.
Common mistakes
- Ignoring the data quality icon and treating every exploration figure as exact.
- Drilling into tiny segments in a sampled exploration and acting on the result.
- Expecting explorations and standard reports to match to the unit, then assuming tracking is broken when they do not.
- Confusing sampling with thresholding, then changing settings that make no difference.
- Pulling large date ranges through a dashboard connector without checking whether the source query was sampled or hit quota limits.
How to act on it
When you see a sampled exploration, first try shortening the date range and running it in pieces, for example month by month, then combining the results. Use standard reports for headline totals wherever they answer the question, and keep explorations for analysis they cannot do.
For regular analysis of large volumes, link GA4 to Google’s data warehouse through the BigQuery export, which gives you the raw, unsampled events to query directly. Standard properties have a daily export limit, so check it against your event volume before relying on it.
When a decision rests on a small group, check the underlying sample size as well as the sampling: an unsampled figure built on twelve conversions is still a weak basis for moving budget. Setting up reporting that answers business questions reliably is part of the digital marketing strategy and consulting work I do.