What Are Anonymous Queries in Google Search Console?

What Google hides, and why

Google's documentation calls them anonymized queries: searches made by very few people over the date range. The query text is withheld because a rare, specific search can identify the person who typed it. Think names, street addresses, medical situations.

The key detail: the clicks and impressions from those searches are real and still counted. Your page ranked, someone clicked, the totals chart recorded it. Only the words are missing, and only from the row-level report.

Not the same thing as "(not provided)"

If you have been doing SEO long enough, this problem has a familiar shape. It is not the same one:

• "(not provided)" was analytics-side. From 2011, Google stopped passing the search keyword to analytics tools, and that one label eventually swallowed nearly every organic keyword. Search Console became the replacement window into query data.
• Anonymized queries are Search Console-side. The replacement window itself holds back its rarest rows, for privacy rather than policy.
• The scale is different. "(not provided)" hid almost everything. Anonymization hides a minority share that varies by site, and unlike its predecessor, you can measure yours exactly.

The practical difference matters: "(not provided)" ended keyword-level analytics. Anonymized queries only trim it, and the trim is knowable.

Why your query rows never add up to the totals

This is the symptom that brings most people to this question. Export your queries, sum the clicks column, and the sum lands short of the totals line at the top of the report. Every time, on every property.

In the example above, the totals line says 10,000 clicks, the visible query rows add up to 8,700, and the missing 1,300 clicks belong to anonymized queries. The numbers are made up; the arithmetic is exactly what your own export will show.

How to measure your own hidden share

No published number tells you how big your share is, because it depends on your query mix. Sites with lots of rare, specific, long-tail searches hide more; head-term sites hide less. Measuring yours takes three steps:

1. Pick a date range and note the unfiltered total clicks from the Performance report.
2. Export the queries for the same range and sum the clicks column.
3. Subtract, then divide by the total. That percentage is your anonymized share.

Worth writing down once a quarter. A stable share is background noise; a share that suddenly grows says your traffic is shifting toward rarer searches, which changes how much you should trust query-level analysis.

One precision note: run the subtraction on clicks and on impressions separately, and label which one you are quoting. The two shares usually differ, because hidden queries earn impressions more easily than clicks, and a report that mixes the two definitions will contradict itself a quarter later.

The filter trap

The behaviour that actually distorts analysis is not the missing rows. It is what filters do. The moment you apply any query filter, anonymized queries are excluded from the totals, not just the row list.

• Filtered views use a smaller denominator. A "queries containing X" view and a "queries not containing X" view will never sum to the unfiltered total.
• The branded split shrinks both halves. Your branded versus non-branded views each lose the anonymized clicks entirely. Compare each half against itself over time, never against the property total.
• CTR comparisons stay honest only like-for-like. Filtered CTR against unfiltered CTR is two different populations wearing the same label.

Where you can and cannot see them

The one place with any visibility is the BigQuery bulk export, which writes a single combined row per day for anonymized queries: their clicks and impressions, never their words. If you already run the export for keeping your data past 16 months, you get the hidden share charted for free. Nothing else, interface, API, or export file, recovers them.

A common false hope is the API. It does return far more rows than the interface's 1,000-row table, so it feels like it sees deeper, but it applies exactly the same anonymization: no row for a hidden query, and the same exclusion from any filtered total. More of the visible long tail, none of the invisible one.

What to actually do about them

Mostly: stop worrying that something is broken, and adjust how you compare numbers. Three habits cover it:

• Compare like with like. Rows to rows, totals to totals, each filtered view against its own history. The gap only bites when you mix denominators.
• Note your share quarterly with the three-step measurement above, so a shift in it reads as a signal instead of a mystery.
• Earn more visible queries. You cannot reveal hidden ones, but as pages gain impressions on more common searches, more of your traffic lands in rows you can actually analyse. That is ordinary SEO work: the five monthly Search Console jobs are where it happens.

Check your share with one prompt

With CrawlRaven's MCP server connected, the measurement is one question instead of an export and a spreadsheet:

For [your site] over the last 3 months: compare total clicks with the sum of clicks across all visible queries, and tell me what share of my clicks comes from queries I cannot see. Has that share changed versus the previous 3 months?

When you analyse my queries, always tell me whether a comparison mixes filtered and unfiltered views, since anonymized queries drop out of filtered totals.

The Search Console MCP guide covers what a connected assistant can and cannot read; anonymized queries stay hidden from it too, which is the point of them.