GA4 data sampling: when it happens, how to spot it, how to avoid it

GA4 data sampling happens when a report, exploration or API request needs more events than your property’s quota, so Google Analytics calculates the answer from a subset of events and scales it up. On a standard property the quota is 10 million events per event-level query; on Analytics 360 it goes up to 1 billion. The data quality icon tells you when a result is sampled and what share of the data was used. Small sites rarely hit the limit. Busy stores that build explorations over months of data hit it all the time.

Data sampling is the practice of computing a report from a subset of the underlying data and scaling the result to estimate the total. It makes large queries fast, at the cost of precision, and the loss of precision is largest for small segments.

When does GA4 sample data?

GA4 samples when the number of events needed to answer a question exceeds the quota for your property. Google’s help page puts it plainly: sampling “may occur when the number of events used to create a report, exploration, or request exceeds the quota limit for your property.” What counts is the events inside the date range and filters you chose, not your monthly total.

Explorations are where most teams run into sampling, because they let you pick any date range, segment and breakdown. The trigger is the query, so the same exploration can be unsampled for last week and sampled for last quarter. Google also notes that filtering large datasets by country “may result in data sampling in standard Google Analytics reports and Explore,” so standard reports are not completely immune either.

How can you tell if a GA4 report is sampled?

Check the data quality icon at the top of the report or exploration. “When your results use sampling, it is indicated in the data quality icon with the percentage of data used to create the results,” per Google. On Analytics 360, the icon is red or yellow while results are sampled and green when they are not.

Make it a habit before you screenshot or export anything:

1. Open the data quality icon and read the percentage of data used.
2. If it is below 100%, note it next to the number you share, or change the query.
3. Re-check after every change to the date range, segment or breakdown, because any of them can push the query over the quota.

A sampled exploration often looks perfectly normal. Totals are scaled back up, so nothing on the chart warns you. The icon is the only signal.

How much can sampling distort your numbers?

Totals are usually close, while small segments can be well off. The error of an estimate depends on how many events the sample actually contains for that segment, not on how big the sample is overall.

A worked example with hypothetical numbers. An online store’s GA4 property records about 400,000 events a day. A marketer builds an exploration of the last 90 days:

• Events in the query: 400,000 × 90 = 36 million.
• Share GA4 can use on a standard property: 10 million ÷ 36 million ≈ 28%.
• A small referral channel drove 200 purchases in that period. In a 28% sample, GA4 estimates that figure from about 200 × 0.28 = 56 purchase events, then scales it up.

Using the usual approximation that a count estimated from n sampled events has a relative error of about 1 ÷ √n:

The store-wide total is fine. The small channel’s number now moves by around 13% either way, which is enough to make a decent campaign look like a failure, or the other way round. This is why sampling hurts most exactly where marketers look hardest: new campaigns, small cities, one landing page, a single UTM content variant.

What is the difference between sampling, thresholds and the (other) row?

They are three separate mechanisms, and fixing one does not fix the others. Sampling estimates numbers from part of the data. Thresholds hide rows to protect user privacy. The (other) row groups the least common values when a table has too many rows.

Thresholds are applied, in Google’s words, “to prevent anyone viewing a report or exploration from inferring the identity or sensitive information of individual users.” They apply when a report includes demographic data or audiences built on it, and to search query information when there are not enough users. Google’s advice is to adjust the date range, because low user or event counts in a short range are what trigger them.

The (other) row appears “when the number of rows in a table exceeds the table’s row limit.” Google treats any dimension with more than 500 values as high cardinality. Full page URLs with query strings, free-text custom dimensions and per-user IDs stored as custom dimensions are the usual causes. Reports with secondary dimensions, filters or comparisons are more likely to show it.

How do you avoid sampling in GA4?

On a standard property you cannot raise the quota, so you work under it. Five methods, from quickest to most thorough:

1. Shorten the date range. In the example above, 10 million ÷ 400,000 events a day = 25 days. Run three 25-day explorations instead of one 90-day one. Event and purchase counts can be added across periods; user counts cannot, because the same person can appear in more than one.
2. Use standard reports for routine numbers. Google recommends them because they have aggregate tables, and they are less likely to show (other) rows than an exploration with several breakdowns.
3. Cut cardinality. Strip query strings from page paths before they reach GA4, keep UTM values to a fixed list (the UTM parameters guide has a naming convention), and use the User-ID feature rather than a custom dimension for per-user identifiers.
4. Export raw events to BigQuery. The daily export “exports all the raw, unsampled event data once per day from the previous day.” Standard properties are capped at 1 million events a day in that export; above it, use filters to stay under the limit or the streaming export, which Google describes as best effort with possible gaps. The guide to exporting GA4 data covers the setup and cost arithmetic.
5. Move to Analytics 360 or a tool that does not sample. On 360, unsampled explorations are available only when a query would return more than 1 billion events, and they draw on 20,000 tokens per property per day, with at most 5,000 per query. For most teams that is a contract-sized answer to a smaller problem.

Does every analytics tool sample data?

No. Sampling is a design choice to keep big queries fast and cheap, not a law of analytics. Matomo lists “no data sampling” among its On-Premise features, as covered in the Matomo vs Google Analytics comparison, and AdPix reports are not sampled either.

AdPix is a web analytics platform that replaces Google Analytics 4: realtime, acquisition, engagement, conversion and e-commerce reports, plus attribution, lift testing and marketing mix modelling, with no data sampling and a free plan. For the problem in this article, that means a 90-day view of a small referral channel is computed from all of its events, so the 56-versus-200 gap in the example does not arise.

What changes by plan, per adpix.io:

• Free plan: acquisition reports by channel, source / medium and UTM campaign, realtime data and conversion funnels, with 100,000 events a month and 30 days of raw data.
• Starter plan and up: segments, audiences and custom reports, the features you would use to look at small slices of traffic, plus online sales and revenue reports.

AdPix has limits of its own. Each plan has a monthly event cap and a fixed retention period, so long histories need an export routine or a bigger plan. AdPix is also hosted inside Iran, which is the reason to choose it for an Iranian site and something readers elsewhere should check against their own data-location rules.

What should you do this week?

Open the three explorations your team shares most often and read the data quality icon on each. If any uses less than 100% of the data, rebuild it on a date range under your quota, or move it to a standard report. If the segments you care about are small and the warnings keep coming back, run a tool that does not sample next to GA4 for a month and compare the numbers for those segments. The Google Analytics alternative guide explains how to run that side-by-side test without losing your GA4 setup.

Measure your site with AdPix today

The free plan covers 100,000 events a month, no credit card. Setup takes under five minutes.

Start freeSee pricing