Search Console is the most trustworthy dataset in SEO because it comes from the source. It is also incomplete in specific, documented ways, and teams that do not account for them draw conclusions the data cannot support.
Anonymized queries
Google filters out queries it considers rare enough to be personally identifying. Those impressions and clicks still appear in your site-level totals, but the queries themselves are withheld from the query dimension.
The consequence is arithmetic: the sum of your query rows will be less than your site total, sometimes far less. Sites with long-tail-heavy profiles routinely find that 30–50% of their clicks sit in the anonymized bucket. Local businesses and niche B2B sites are affected most, because their query distribution is naturally sparse.
Two implications follow. First, never report "clicks by query" as if it accounts for all traffic — it does not, and the gap varies by site. Second, the anonymized bucket is not random noise; it disproportionately contains the most specific queries, which are often the highest-intent ones. Your visible query data is systematically biased toward the head.
Row limits
The Search Console UI caps exports at 1,000 rows. The API goes to 25,000 rows per request, with pagination beyond that. If you are exporting from the interface and your site has more than a thousand queries, you are analyzing an arbitrary top slice.
This matters most for the analysis people most often want to run — finding underperforming queries, striking-distance opportunities, content gaps — because all of those live in the tail that the UI truncates.
The fix is the API. It is genuinely accessible: a service account, the searchanalytics.query endpoint, and a loop over date ranges. A weekend of setup replaces a permanent ceiling on your analysis.
Data freshness and the final-day dip
Search Console data lags by roughly two days, and the most recent day in any report is usually incomplete. Reports that include "today" or "yesterday" will show an apparent decline that is purely an artifact of incomplete collection.
Every recurring report should end its window at least three days before the run date. A surprising number of "traffic is dropping" escalations resolve to this.
Property scope
Data is scoped to the property as verified. A URL-prefix property for https://example.com/ does not include http://, does not include www. if that is a separate hostname, and does not include subdomains.
Domain properties solve this by covering all protocols, subdomains, and hostnames — and they are the right default for almost everyone. Sites migrating from URL-prefix properties often discover the "traffic drop" they were investigating was a hostname sitting outside the property boundary the entire time.
The 16-month window
Search Console retains 16 months. That is enough for one year-over-year comparison and not enough for two.
If you care about multi-year trends — and for seasonal businesses you should — you need to export and store the data yourself before it ages out. The BigQuery bulk data export, configurable directly from Search Console, is the cleanest way to do this: it writes daily partitions to your own warehouse and keeps them as long as you do.
Anyone running SEO for a business with a multi-year horizon should turn this on now rather than discovering the retention limit at the moment they need history.
Discover and News are separate
Discover and Google News performance are reported in their own sections and are not included in the web Search performance figures. For publishers, Discover can be a substantial share of total Google traffic, and analyzing only the Search tab can understate Google's contribution significantly.
What this adds up to
Search Console is accurate about what it reports. It reports a filtered subset, with a known lag, within a defined property boundary, for a limited window.
Practical rules that follow: use domain properties, use the API rather than UI exports for anything analytical, end reporting windows at least three days back, enable the BigQuery export before you need the history, and state site totals and query-level totals as separate numbers rather than implying the latter explains the former.
None of this makes the data less valuable. It makes the conclusions you draw from it survive scrutiny, which is a different and more important property.