How thresholding works
GA4 will not show you a row when the number of users behind it is small enough that the row could point at a particular person. Instead of printing the figure, it withholds it and marks the report with a notice saying data has been removed. Nothing has been deleted and nothing is broken; the platform is refusing to answer a question that is too specific.
What sets it off is a combination of a small audience and a detailed question. Reports that include demographic or interest dimensions — age, gender, interests — are the usual trigger, because those come from signed-in Google users rather than from your own site. Long date ranges make it less likely, because more users accumulate; narrow filters, a single city, a single landing page or a short window make it more likely. Switching the reporting identity away from the blended, signals-based setting often removes the notice, because the report no longer depends on that identity data.
Why thresholding matters
It quietly changes totals. A report with withheld rows adds up to less than the same report without those dimensions, so two views of the same period disagree and the difference is easy to blame on a tracking fault that does not exist. Once you know the notice is there, the disagreement stops being a mystery.
It also hits smaller advertisers hardest. A site in Nepal serving one city, or a niche B2B service anywhere, will trip thresholding constantly simply because its audience is small — the very accounts that most want to know who their visitors are get the least detail about them.
Common mistakes with thresholding
The first is reading a gap as a zero. A withheld row is not evidence that a city, an age group or a channel sent nobody; it is evidence that too few users sit behind it to report safely. Concluding that a segment does not convert, and cutting spend on that basis, is a real and avoidable mistake.
The second is mistaking it for sampling. Sampling estimates from a slice of the data; thresholding withholds complete rows on privacy grounds. A third is trying to engineer around it by slicing even more finely, which makes every row smaller and triggers it more often.
How to act on it
Check the report notice first, then widen the question. A longer date range, fewer filters and coarser dimensions will often bring the rows back on their own. If the report includes demographics or interests and you only wanted traffic and conversion figures, remove them — that alone resolves many cases.
Where a full, row-level answer genuinely matters, take it from a source that does not apply the rule: your own site’s records, your CRM, or a raw export queried elsewhere. And when you build the measurement plan during a GA4 setup, favour dimensions you collect yourself over inferred ones, because your own event parameters are not subject to identity thresholds.