Discussion
  • Search Console
  • SEO Analytics

Anonymised queries hide a large share of clicks in Search Console. How do you account for them in keyword-level reporting?

Published 1 reply

Discussion starter: an SEO measurement question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • admin
    Admin
    Original poster

    Do not push anonymised clicks down to individual keywords. Keep keyword-level reporting to the queries Google shows, report the hidden clicks as their own line, and allocate them at the page level, where they can be measured, before rolling pages up into topics. In the bulk data export to BigQuery, anonymised rows are present with a null query and a flag, so you can count them per URL. With the API or the interface, you estimate them per page by subtracting a page’s visible query clicks from its total. Either way, every report reconciles to the property total, and a shift in the hidden share cannot pass for a change in demand.

    The complication is that “missing” query clicks come from three different gaps, and only one of them is anonymisation. If you book all of it as anonymised traffic, you will overstate the privacy gap on large sites and hide a problem in your own export method.

    Separate the three gaps before accounting for any of them

    Google’s help documentation defines the first gap plainly: “Some queries are omitted from the report to protect user privacy. These are called anonymized queries.” It adds: “They’re included in chart totals, unless a query filter is applied (for example, ‘queries containing’ or ‘queries not containing’ a given string)”. The same page says “Due to internal limitations, Search Console stores and shows only the most important data rows”, and notes that not every query beyond the anonymised ones appears in the table.

    Gap What causes it How it shows up How to handle it
    Anonymised queries Rare queries withheld for privacy Query-level sums fall short of chart totals; any query filter removes them from totals Measure per page and report as its own line
    Row limits The interface table stops at 1,000 rows; the API returns up to 50,000 rows per day per search type, sorted by clicks Visible but low-volume queries missing from your pull Pull one day at a time and paginate, or use the bulk export
    Aggregation Figures grouped by page are calculated differently from figures grouped by property Page-level sums that do not match property totals, especially impressions, where one search can count once for the property but once for each of your URLs shown Reconcile within one aggregation type

    The API details come from Google’s developer documentation, which gives the 50,000-row daily limit per search type, recommends “running a query each day for one day’s worth of data”, and states: “Impressions, clicks, position, and click-through-rate are calculated differently when grouping results by page rather than by property”. On a small site the row-limit gap may be close to zero, and the whole residual is anonymisation. On a site whose daily query rows exceed the cap, every pull loses its lowest-click rows, and labelling that loss “anonymised” hides a fixable problem.

    Use pages as the reconciliation layer

    Pages are the one dimension where anonymised clicks stay attached to something you can report on. A page-only view has no query dimension or query filter, so its totals include anonymised queries. A page-and-query view does not.

    With the bulk export

    The export’s table reference describes the flag: “Rare queries (called anonymized queries) are marked with this bool. The query field will be null when it’s true to protect the privacy of users making the query”. That turns the hidden share per URL into a count rather than an inference:

    SELECT
     url,
     SUM(clicks) AS clicks,
     SUM(IF(is_anonymized_query, clicks, 0)) AS anonymised_clicks
    FROM `your_project.searchconsole.searchdata_url_impression`
    WHERE data_date BETWEEN DATE '2026-08-01' AND DATE '2026-08-31'
     AND search_type = 'WEB'
    GROUP BY url
    ORDER BY clicks DESC

    Aggregate with SUM rather than assuming one row per key: the same reference says “Performance data is accumulated by Search Console incrementally, resulting in table rows with repeated keys.” Use the site-level table (searchdata_site_impression) for property totals and the URL-level table for page totals, and do not mix the two in one sum.

    With the API, Looker Studio or the interface

    1. Pull page-level totals (page dimension only) for the period.
    2. Pull page-and-query rows for the same period, one day at a time, paginating until the API returns no more rows.
    3. For each page, subtract the sum of its visible query clicks from its total. Call the result “unattributed” rather than “anonymised”, because it still contains any rows lost to limits.

    This is also why dashboards disagree with themselves. A chart that includes the query dimension drops anonymised rows, while a scorecard without it keeps them, so one Looker Studio report can show two click totals for the same dates. Neither is wrong; they answer different questions.

    Rolling pages up into topics

    Map each page to one primary topic or keyword group. Topic clicks are then the page totals summed, and the visible and hidden split becomes an attribute of each topic rather than a hole in the report. This page-first view will disagree with a query-first view, where you classify query text with regex. Both are legitimate. Query-first grouping is precise about what people typed but blind to anonymised clicks, and it splits pages that rank for mixed intents. Page-first grouping includes everything but assumes each page serves one topic. I would report totals and trends page-first, and use query-first analysis to explain what moved within the visible share.

    Allocating hidden clicks without inventing keyword data

    The tempting shortcut is to spread each page’s hidden clicks across its visible keywords in proportion to their clicks. That biases the report towards head terms. Anonymised queries are rare by definition, so they resemble the long tail of a page far more than the handful of terms that dominate its visible clicks. Pro-rata allocation hands most of the hidden volume to those head terms and inflates the keywords stakeholders already watch.

    Two better options:

    • A long-tail line per topic. Report “long tail (unattributed)” under each topic, sized from the page data, with no keyword breakdown. It is honest, and it shows how much of a topic’s traffic sits outside anything a rank tracker can follow.
    • A ratio from the visible long tail, when a split is required. If someone needs, for example, a branded versus non-branded estimate for hidden clicks, derive the ratio from the page’s low-volume visible queries (below a click threshold you set, such as five clicks in the period) rather than from all visible queries. Label the result as a modelled estimate.

    Hypothetical example:

    A software company’s site gets 120,000 Web clicks in a month, and the query rows in its API pull sum to 66,000. The page-level pull shows the documentation directory with 70,000 clicks, of which visible queries account for 31,000. The pricing and feature pages have 25,000 clicks, 21,000 of them visible, mostly branded and head terms. A keyword report built on visible rows says documentation drives 47% of attributable clicks (31,000 of 66,000). The page-level view says it drives 58% of all clicks (70,000 of 120,000). Both numbers are accurate, but only the second reconciles to the total, and the first would quietly shift budget towards the commercial pages.

    Filters and brand splits: where totals quietly change

    Any query filter, including a regex you run over exported rows, removes the anonymised share from the totals. A “branded plus non-branded” split built from your own regex will therefore fall short of the property total, and the gap is the anonymised and unattributed traffic. Report that gap as a third line rather than forcing the two halves to add up.

    Search Console’s built-in branded queries filter works differently, because Google classifies the queries rather than matching your text. Whether its totals include anonymised queries is worth testing rather than assuming, and the help page notes that “This filter isn’t available for sites with a low number of impressions”. Test it on your own property: if branded plus non-branded clicks equal the unfiltered total for the same dates and search type, the filter covers the anonymised share. If they fall short, treat it like any other query filter.

    A reporting layout that reconciles

    Report line Source Includes anonymised clicks? Use it for
    Property total (Web) Performance report with no query dimension or filter; site-level export table Yes Headline trend
    Visible query clicks Query rows; export rows where is_anonymized_query is false No Keyword and intent analysis
    Anonymised or unattributed clicks Export rows where the flag is true, or the page residual This line is the hidden share Coverage, and context for every keyword trend
    Topic clicks Page totals mapped to topics Yes Topic performance and investment decisions
    Tracked keyword positions Rank tracker or Search Console position Not applicable Visibility for chosen terms, not traffic

    The order matters. Readers who see the total and the hidden share first read the keyword table as a partial view, which it is. Readers who see the keyword table first assume it is the whole picture.

    Track the hidden share as a metric in its own right

    Put the anonymised (or unattributed) share of clicks on the dashboard by week, and segment it by template or directory, country and device. Because anonymisation removes rare queries, the share should be higher where demand is spread across many unusual queries: large catalogues, documentation and support content, and pages targeting many local or product variations. It moves when your query mix moves, and that movement distorts keyword-level trends.

    Diagnostic rule:

    If visible keyword clicks fall while the property total holds, check the hidden share before reporting a decline. Demand moving from visible to anonymised queries looks like lost keywords in a query-level report and like nothing at all in the total.

    One hypothesis worth monitoring rather than asserting: AI Mode activity is counted inside the Web totals, and conversational questions can be longer and more varied than classic keyword queries. If that pushes demand towards rarer phrasings in your market, the anonymised share would rise with no change in real demand. Search Console gives you no way to confirm that directly, which is one more reason to judge performance at the page and topic level.

    The decision rule

    Reconcile first, then analyse. Every keyword report should open with the property total, the visible share and the hidden share, then drill into keywords with the caveat that they describe only the visible part. Allocate hidden clicks to pages and topics, not to named keywords. Use the bulk export if you have it, because the anonymisation flag replaces an estimate with a count. And treat any keyword trend that the page-level trend does not confirm as a possible reporting effect until the hidden share has been checked.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.