Google Search Console: the measurement surface that is not a guess

Search Console is Google's own reporting on properties you have proved you own. Almost every other number in search marketing is modelled: estimated from browsing panels, extrapolated from samples, or inferred by scraping results. Search Console reports what Google recorded, the only non-modelled search data you have. The distinction is provenance, not precision. You see only verified properties, so a competitor's Search Console data does not exist.

Property types and verification

Google's property types documentation defines two. A domain property covers a domain including all subdomains and multiple protocols, so example.com covers http and https, www and m. subdomains, and every path beneath. It is verified by DNS record only. A URL-prefix property covers only URLs beginning with the prefix you enter, protocol included, so https://example.com/pets/ excludes both the http version and the m. subdomain. Verification methods include an HTML file, an HTML tag, Google Analytics, Google Tag Manager and your domain provider.

The domain property is usually right, because it cannot quietly omit traffic. A site serving both www and non-www, or still holding legacy http URLs, under-reports in a URL-prefix property and the shortfall is invisible. Add URL-prefix properties only to report a section separately.

The Performance report

The four metrics

Google's Performance report documentation defines them. Clicks are the number of times a user clicked your site from Google Search results. Impressions are how many times your site appeared in Search results. CTR is clicks divided by impressions. Average position is the average position of the topmost result from your site.

Average position, which is widely misread

Google's definition of the metrics gives a worked example. A site at positions 2, 4 and 6 for one query counts as position 2; at 3, 5 and 9 for another it counts as 3; the average is 2.5. Three consequences follow.

It is an average across queries, not a rank. Nobody saw you at 2.5, and it moves when the query mix changes without any result changing.
Aggregation changes it. In Google's example, three URLs from one site shown and clicked for one query give a property-aggregated CTR of 100 percent at position 1, and a page-aggregated CTR of 33 percent per URL at position 2. The chart aggregates by property, the table by page.
Only queries with an impression count. Newly appearing at position 40 for a batch of queries lowers the average although nothing existing got worse.

Google's own guidance is to watch trends in impressions and clicks rather than position alone.

Dimensions and filters

Six dimensions group the table: queries, pages, countries, devices, search appearance and dates. The search type filter selects web, image, video or news; the date filter sets the range and offers a comparison against a previous period. Filters combine, and query and page filters accept a substring or a regular expression.

How AI features are reported

Per Google's documentation on AI features, sites appearing in AI features such as AI Overviews and AI Mode are included in the overall search traffic in Search Console, reported in the Performance report within the Web search type. No filter isolates them, so a page's numbers blend classic results and AI features.

Search Console also carries a separate generative AI performance report. It covers AI Overviews and AI Mode, and reports impressions only, defined as how many times links to your site were shown in a generative AI feature. Clicks, CTR and average position are not included. Its dimensions are pages, countries, dates and devices, with no query dimension. It draws on the Web search type data, excludes Search Labs experiments, and is still rolling out to some owners. What generated answers do to clicks is covered in AI Overviews and AI Mode.

Four analyses worth running

Queries where you rank 11 to 25. Open the queries tab, set the last three months, enable average position, and export. There is no numeric position filter, so filter the export to rows between 11 and 25. Those are queries with demand where you sit on page two. See keyword research.
Pages whose impressions hold while clicks fall. Open the pages tab, compare against the previous period, sort by the difference column for clicks, then read the impression difference on the same rows. Google's traffic drop guidance states that if impressions remain the same but clicks drop, you might not be generating the best title and snippet.
Period over period, to detect decay. Compare the last three months against the previous period, then repeat year over year. Google recommends viewing the last 16 months so a fall can be checked against seasonality first. A page shedding impressions across consecutive comparisons needs updating; see freshness.
Query export as prompt research input. Export the queries table, or the bulk data export for the fullest list. It is recorded demand rather than invented phrasing, and the honest starting point for prompt research.

The other reports worth knowing

URL Inspection reports Google's indexed version of one URL: whether it is on Google, the last crawl date, whether robots.txt or noindex blocks it, and the Google-selected canonical. A live test fetches the page in real time, and requesting indexing carries a daily quota and no guarantee of inclusion.
Page indexing shows how many URLs Google indexed, and why the rest were not. "Crawled, currently not indexed" means Google fetched the page and chose not to index it; "Discovered, currently not indexed" means it has not been crawled yet.
Sitemaps shows each sitemap you submitted, whether Google could fetch and parse it, the last read date and the URL count. Sitemaps found through robots.txt are absent.
Manual actions reports whether a human reviewer judged pages non-compliant with the spam policies, and which patterns are affected.
Core Web Vitals reports Chrome User Experience Report field data for Largest Contentful Paint, Interaction to Next Paint and Cumulative Layout Shift, grouped by sets of similar URLs.

The documented limits

Row limits and truncation. The table displays a maximum of 1,000 rows, so rare and long-tail rows are omitted from it while remaining in the chart total. Google states that Search Console stores only the most important rows, and that the fullest query list comes from the bulk data export. The API accepts up to 25,000 rows per request.
Anonymized queries. Some queries are omitted from the query table to protect user privacy, though Google includes them in chart totals unless a query filter is applied. That is one reason a filtered "match" and "does not match" pair does not sum to the total.
Delay. Collected data is usually available in two to three days, the newest can be preliminary, and days are labeled in Pacific Time.

The totals are authoritative and the table is incomplete at the tail, which is worth knowing before building a report on a row count.

It reports Google only

Search Console covers Google Search. It says nothing about other search engines, and nothing about how any answer engine fetched your pages. For that the first-party evidence is your own server logs, which record every request including those from AI crawlers; see server logs.