Open your analytics dashboard, filter by a date range longer than 90 days, add a secondary dimension, and check the top-right corner. If you see a green or yellow shield icon in Google Analytics 4, or a percentage beneath your report in other platforms, you’re looking at sampled data—not your full dataset.
Analytics sampling means the platform processes a subset of your traffic data and extrapolates the rest. It speeds up queries on large datasets, but it also introduces margin-of-error ranges that most operators never notice until a metric suddenly doesn’t match their records.
When platforms trigger sampling
Google Analytics 4 samples when your property exceeds 10 million events in the selected date range and you apply filters, secondary dimensions, or custom explorations. Standard reports stay unsampled; explorations and custom breakdowns do not.
Adobe Analytics samples at the report-suite level when queries exceed processing thresholds. Matomo samples only if you enable the setting manually. Plausible and Fathom don’t sample at all—they process every event—but they also cap the complexity of queries you can run.
The trigger isn’t daily traffic—it’s cumulative events in the query window. A site logging 120,000 events per day hits the GA4 sampling threshold after 84 days. Add a UTM breakdown or device-category filter, and the platform samples to return results faster than five seconds.
What sampling skews most
Aggregate metrics like total pageviews or sessions stay fairly accurate under sampling. Google claims 95% confidence intervals, which means your reported total might land within ±5% of the true figure.
Segmented comparisons break down faster. If you’re comparing conversion rates between traffic sources, and one source represents 2% of your total traffic, sampling can miscount that slice by 20% or more. Small segments, long date ranges, and multiple filters compound the error.
Funnel reports suffer worst. If your checkout funnel has five steps and you’re filtering by device type across six months, you’re stacking sampled data at every stage. A 3% error at step one becomes 8% by step four, and your abandonment-rate diagnosis points you in the wrong direction.
How to spot sampling before it distorts decisions
GA4 displays a green shield icon when sampling is under 10%, yellow when it’s higher. Click the icon to see the exact sample size and confidence interval. If your report says “based on 8.4% of sessions,” your margin of error is roughly ±3.4% for aggregate numbers and wider for segments.
Check sampling before you:
- Compare two time periods to measure campaign impact
- Segment by UTM source, device, or geographic region
- Build a funnel or path exploration longer than three steps
- Export data to a spreadsheet for further analysis
If you see yellow or red sampling warnings, narrow your date range, remove secondary dimensions, or export raw data to BigQuery and query it directly. GA4’s BigQuery export is unsampled and free for properties under 1 million events per day.
Alternatives when you need unsampled precision
Switch to standard reports instead of explorations when possible—those stay unsampled in GA4. For custom breakdowns, export to BigQuery and use SQL. The learning curve is steeper, but you get exact counts.
Plausible and Fathom don’t sample, but they limit filtering depth. You can’t build multi-step funnels or apply three-layer segments, which keeps queries fast without statistical shortcuts.
Matomo’s on-premise version lets you disable sampling entirely if your server can handle the query load. Expect 8–15 second page loads for complex reports on datasets over 50 million events.
For conversion tracking and funnel analysis, many operators now run a dedicated event pipeline to a data warehouse—Snowplow, Segment, or RudderStack feeding into Redshift or ClickHouse—and only use GA4 for directional traffic trends. It’s overkill for a solo newsletter, but it’s standard past 500,000 monthly sessions.
If you’re making budget or product decisions from a report with a yellow sampling badge, re-run the query with a tighter date range or fewer dimensions. A 5% error on total traffic is noise. A 15% error on your highest-converting UTM source sends you optimising the wrong channel.
Want more breakdowns on the tools operators actually use? Subscribe to One Two Three Send—one article every morning, no sampling required.