Data Sampling
Data sampling is the practice of analyzing a subset of a dataset and scaling the results to estimate the whole. Analytics platforms use it to return answers quickly when a query covers more data than they will process in full. The numbers you see are estimates: close enough at the top level, and increasingly unreliable the smaller the segment you look at.
In Google Analytics 4, sampling is triggered by query size, not by property size. Standard reports are never sampled. Explorations are, once a query crosses roughly 10 million events on a standard property.
Why Data Sampling Matters
Sampling changes what a number means. An unsampled report counts events. A sampled report counts a fraction and multiplies up, so every figure carries a margin of error GA4 does not display.
The error is not spread evenly. Totals hold up well because they average across the whole sample. Small segments do not. A conversion path with 40 real completions might be built from three sampled events and reported as 60, or from one and reported as 20. The rarer the behavior, the wider the swing.
It matters most on decisions made from a segment worth a few percent of traffic. Two sampled explorations of the same date range can return different numbers, which is usually the first sign a marketer notices.
Types of Data Sampling
Sampling methods fall into two groups.
Probability sampling gives every record a known chance of selection:
- Simple random sampling: every record has an equal chance. Analytics platforms use this most.
- Stratified sampling: the dataset is split into groups (by channel, country, device) and each is sampled separately, which protects small segments.
Non-probability sampling does not. Convenience sampling takes whatever data is easiest to reach and quota sampling stops at a fixed count per group, so neither supports a reliable estimate of the whole.
GA4 applies random sampling at the event level within the queried date range. It does not stratify, which is why low-volume segments degrade first.
The GA4 Sampling Threshold
GA4 samples when a single query would need to read more events than its quota allows.
| Surface | Standard property | Analytics 360 |
|---|---|---|
| Standard reports | Never sampled | Never sampled |
| Explorations | Sampled above 10 million events | Sampled above 1 billion events |
| Data API | Sampled above 10 million events | Higher quota |
Standard reports escape sampling because they are served from pre-aggregated tables built during processing, not queried live. That is also why they allow fewer dimension combinations.
The 10 million count is the number of events the query must scan, not the number of rows returned. A 90-day exploration on a property recording 150,000 events a day scans 13.5 million events and gets sampled. The same exploration over 60 days scans 9 million and does not. High-cardinality dimensions and multiple segments push the scan cost up further.
Analytics 360 properties can request unsampled results when a query exceeds 1 billion events. Requests draw on 20,000 tokens per property per day, capped at 5,000 per query, and the resulting tabs expire after 50 days.
How to Check If Data Is Sampled
Look at the data quality icon in the top right of any exploration. A green check means the report used everything: hovering it shows “This report is based on 100.0% of available data.” A yellow icon means sampling was applied, and hovering it shows the percentage used, such as “This report is based on 43.2% of available data.”
Treat anything below 100% as an estimate. Below roughly 30%, segment-level numbers are directional at best.
The icon is the only warning GA4 gives. Data pulled through the Data API into a dashboard or a scheduled export can be sampled with no indicator at all, so connected reports need the same date-range discipline.
How to Avoid Sampling in GA4
Sampling is a function of the query, so the fix is always to make the query cheaper.
- Shorten the date range. The most effective change. Run three 30-day explorations rather than one 90-day exploration and add the results.
- Use standard reports where they answer the question. Traffic acquisition, landing pages, and events reports are unsampled at any volume.
- Remove segments and comparisons. Each one multiplies the events the query must evaluate.
- Cut dimensions. Drop high-cardinality fields such as page path or transaction ID unless the question needs them.
- Filter before you query. A filter set on the exploration itself narrows the scan; a filter applied to the output does not.
- Export to BigQuery. The GA4 BigQuery export is event-level and never sampled at any volume. It is the only complete answer for large properties, though it needs SQL.
Data Sampling vs Data Thresholding
Sampling and thresholding both make GA4 numbers incomplete, for opposite reasons.
Sampling is a processing limit. All the data qualifies, GA4 reads part of it, and scales the answer up. Widening the query makes it worse.
Data thresholding is a privacy control. The data is read in full, then rows with too few users behind them are withheld from the view. Widening the query usually makes it better.
The diagnostic is simple: if shortening the date range fixes the problem, it was sampling. If lengthening it fixes the problem, it was thresholding.
Frequently Asked Questions
What is data sampling in analytics?
Data sampling in analytics is when a platform analyzes a portion of your data and scales the result to estimate the full figure. It exists to keep queries fast on large datasets. The numbers are estimates rather than exact counts, so totals stay roughly accurate while small segments can be substantially wrong.
Is GA4 data sampled?
Sometimes. GA4 standard reports are never sampled, whatever the traffic volume, because they are served from pre-aggregated tables. Explorations and the Data API are sampled once a query exceeds 10 million events on a standard property, or 1 billion on Analytics 360. Most small and mid-size properties never hit either limit.
What does sampled data mean?
Sampled data means the figures you are reading were calculated from a subset and multiplied up to represent the whole. A report based on 25% of available data examined one event in four and scaled the results by roughly four. The direction of a trend is usually reliable; the exact value is not.
What is the GA4 sampling threshold?
The threshold is 10 million events per query for standard properties and 1 billion events per query for Analytics 360. It counts events scanned by the query, not rows returned, so date range, segments, and dimension cardinality all affect whether a report crosses it.
How do you avoid sampling in GA4?
Shorten the date range first, then remove segments and high-cardinality dimensions. Use standard reports instead of explorations where they answer the question, since standard reports are never sampled. For properties that exceed the threshold on any useful query, the BigQuery export is the only way to get complete event-level data.
To keep a clean, unsampled record of campaign clicks alongside your GA4 reports, track tagged links with linkutm.