Quick answer: GA4 applies data thresholding to protect user identity when a report combines low-volume segments with demographic/interest data or User-ID. The threshold isn't a fixed number — it's a complex anonymization rule. The most common fixes are: switch reporting identity to "Device-based," exclude demographic dimensions, expand the date range, or use BigQuery export for unfiltered data.
What is data thresholding
Data thresholding is a privacy feature where GA4 withholds data from a report if showing it could let you identify individual users. When this happens, you'll see:
- A small orange triangle warning icon at the top of the report
- Hovering over it shows: "Data was withheld to comply with Google's data thresholds policy"
- Sometimes specific rows disappear; sometimes entire metrics return zero
When thresholding kicks in
Thresholding applies when ALL of the following are true:
- Reporting identity is set to "Blended" or "Observed" (uses signals like Google signals)
- The report includes demographic data, interest data, or User-ID dimensions
- One or more segments in the report has fewer users than the (undisclosed) minimum
It does NOT apply when:
- Reporting identity is set to "Device-based"
- The report uses only event-level data (no demographics)
- The user counts are above the threshold (~50+ users, though Google doesn't publish the exact number)
4 ways to work around thresholding
Method 1: Switch reporting identity to "Device-based"
The fastest fix. Go to Admin → Data display → Reporting Identity → switch from "Blended" to "Device-based." This tells GA4 to use only the device (cookie) for user identification, which doesn't require thresholding.
Trade-off: You lose cross-device user stitching. Users on phone + desktop count as 2 users instead of 1. For most reporting use cases this is fine; for cross-device journey analysis it's a real loss.
Method 2: Remove demographic dimensions from the report
If your custom report includes Age, Gender, Interest Categories, or User-ID as dimensions, those are what trigger thresholding. Removing them from the report's dimension list often unblocks the data.
Method 3: Expand the date range
Thresholding is triggered by low absolute user counts. Expanding from "Last 7 days" to "Last 30 days" or "Last 90 days" often pushes you above the (undisclosed) threshold.
Method 4: BigQuery export
The export from GA4 to BigQuery is not subject to data thresholding. Every row is there, regardless of how small the segment is. This is the only way to get truly complete data for sensitive low-volume analyses.
Setup: Admin → BigQuery Links → Link → choose a free-tier-eligible project → enable daily export.
How to know if thresholding is hurting you
Open any report and look at the top right corner. If you see a small triangle (sometimes orange, sometimes gray), data has been withheld. Click it for details.
Common reports affected:
- Demographic details (Age, Gender) — almost always thresholded for small segments
- Conversion paths with rare touchpoints (e.g., "podcast")
- User explorer showing fewer rows than expected
- Custom audiences with strict criteria
The big picture: thresholding is GA4 being conservative
Universal Analytics didn't have aggressive thresholding, so GA4 feels stricter by comparison. The reason is privacy regulation (GDPR, CCPA) and Google's own reputational risk after years of regulator scrutiny. The right framing isn't "GA4 is broken" — it's "GA4's defaults err on the side of privacy, and you can configure your way out of most of the restriction."
The one trade-off you can't escape: if you want truly cross-device unified user reporting AND zero thresholding AND demographic data, you'll need BigQuery export. There's no UI-only solution.
Audit your reporting identity and data quality automatically
Thresholding is one of dozens of subtle GA4 issues that quietly distort your data. Run a free Snifflytics audit to see whether your reporting identity, BigQuery export, and data retention settings are configured to minimize data loss — and exactly which reports are being affected.