Search Console's four numbers are honest, but each one counts something other than what owners assume

One clinic's export, and what it could not carry

On 23 August 2026 I downloaded the Search Console performance export for my own clinic's website, covering 1 March to 21 August, web search only, and published it as a case study with its method, limitations and file fingerprints ^[20].

The headline numbers read like an advertisement. Clicks per day rose from 2.6 in the first 28 days of structured work, which began on 24 April, to 13.9 in the latest 28 days; impressions per day rose from 79 to 387 over the same windows; and the weekly average position moved from 19 to 9 between late April and August ^[20].

The same page lists what those numbers cannot carry. The website was new, and a young site maturing in Google's index would have produced some improvement regardless ^[20]. The clinic ran offline and social marketing in the same period, which drives brand searches ^[20]. Search Console counts Google web search only ^[20]. And correlation, the page says, is still not causation ^[20].

I am a prescribing pharmacist, not a marketer. I built the clinic's website, Google Business Profile, analytics and tracking myself during 2026, on a small budget and without an agency. The reason Search Console deserves an owner's attention is that it comes from Google and cannot be edited by whoever presents it. That is worth something only if you know what each column counts.

What the four numbers actually count

Google's definitions are short. 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 "the click count divided by the impression count"; and average position is "the average position of the topmost result from your site" ^[1].

Each hides a rule. An impression means "that a user has seen (or potentially seen) a link to your site" ^[2]. The words in brackets matter: a result near the bottom of the page counts whether or not anyone scrolled to it ^[2]. Impressions record Google choosing to show you, not anyone noticing. Two results from the same site on one page count as a single impression in the chart ^[1].

Position is the least intuitive. Google defines it as "the topmost position occupied by a link to your property or page in search results, averaged across all queries in which your property appeared", with an example: positions 2, 4 and 6 for one query count as 2, positions 3, 5 and 9 for another count as 3, and the average is 2.5 ^[2]. An AI Overview occupies a single position and every link inside it is given that position; AI Mode follows the same method ^[2].

The position value is a complex metric that can be misleading if you don't understand the subtleties.

Why average position is the most misread figure

The phrase that matters is "averaged across all queries in which your property appeared". If a result gets no impression, because it sits on page three and the searcher only looked at page one, its position is not recorded for that query ^[2]. The average describes only the occasions on which Google showed the site at all.

A site that begins to appear at position 40 for hundreds of new searches will see its average get worse while nothing has deteriorated; a site that stops appearing for its weakest searches will see its average improve while its visibility shrinks. This is why the case study's weekly average moving from 19 to 9 ^[20] has to be read beside its impressions line, which rose 4.9 times over the same windows ^[20]. My reading, an inference rather than a measurement, is that the average improved while the site was shown far more often, the pattern one would expect if rankings improved rather than weak queries falling away. The export cannot prove that; it shows two lines moving together.

The average is also weighted by volume. Google's bulk export documentation gives the calculation: the sum of the topmost positions divided by the sum of impressions, plus one ^[9]. One query with thousands of impressions dominates, and for a local business whose most frequent search is its own name, the headline average largely describes how it ranks for that name. It also changes meaning between chart and table: the chart covers the whole property and each table row covers one URL or grouping ^[1].

And it blends devices, countries and people. The report can be split by device and by country ^[3], but the headline mixes them all. Google says the figure is the average for all searches and "for your specific search your position might be different" ^[2]; its guidance on why results differ lists time, location, language, device, personalisation and recent searches ^[15]. For a clinic in Bradford, the searcher two streets away and the searcher across the county see different pages; the average is a blend of both.

Impressions without clicks are neither failure nor success

Because an impression only requires that a link could have been seen ^[2], a growing impressions line often arrives weeks before a growing clicks line: Google is testing a site against more searches, often at positions where few people click. Owners read the gap as failure or as success; neither reading is safe.

The useful reading is in the table. Google's documentation suggests using the Query dimension to find high-impression, low-CTR rows ^[3]. Those are searches where the business is considered but not chosen, the most specific thing the report can say about where improvement is possible. Whether a better page changes the outcome is a hypothesis to test, not a promise the report can make. Some low-CTR rows are not problems at all: several links to the same site inside a knowledge panel count as one impression when grouped by property ^[2], and a searcher who wanted the phone number got it without clicking.

Brand and non-brand tell different stories

Most local businesses receive a large share of clicks from people searching their own name, and those people were going to find the business anyway. The case study's rule for separating them is stated exactly: non-brand queries are the exported queries whose text does not contain the brand terms 'pulse', 'clarity', 'p&c' or 'pandc', case-insensitive, and 946 of the 1,000 exported queries qualify ^[20]. Its summary: brand searches show marketing is working; non-brand searches show visibility is working ^[20].

The two lines respond to different causes. The case study notes that the clinic's offline and social marketing drives brand searches ^[20]; a leaflet or a recommendation raises the brand line without any change to the website. Mixing the two in one total and attributing it to search work is the most common way a flattering report is assembled, and it needs no false number. The query filter accepts strings and regular expressions ^[4], but filtering has a cost, covered next.

The queries you are not allowed to see

Search Console omits some queries to protect user privacy and calls them anonymised queries; they are included in chart totals unless a query filter is applied, and they never appear in the table ^[3]. Google says plainly that filtering by query or URL can affect the totals because of data truncation and the omission of anonymised queries ^[4].

How much is hidden has been measured independently. Ahrefs, which sells SEO software and has an interest in the answer, examined 22 billion clicks across 887,534 properties and found that in April 2025 anonymised queries accounted for 46.77 per cent of clicks, slightly higher than its 2022 study ^[16]. Sites with lower and with higher traffic appeared to be missing more than sites in the middle ^[16]. Browser Media, a UK agency, found 46.21 per cent across ten client sites in 2022 ^[17]. Google has not published the threshold; when the documentation changed in 2022, Danny Sullivan of Google said it was meant to make clearer that very rare long-tail terms can add up, and that privacy concerns lie behind the omission ^[18].

For a small business this cuts both ways. Roughly half the clicks may belong to searches the owner will never see, and those are likely to be the specific, local, multi-word searches that matter most to a clinic or a trade ^[16]. But because anonymised queries are in the unfiltered chart totals ^[3], the headline clicks figure is complete; it is the query table, and any total built by filtering it, that is partial. A brand versus non-brand split will add up to less than the chart, and the difference is not an error.

Rows, limits and the difference between the interface and the API

Google states that "due to internal limitations, Search Console stores and shows only the most important data rows" and directs anyone who needs more to the bulk export ^[3]. The case study's query export is 1,000 rows ^[20], which is what the report's download produces. For a larger site it is a selection chosen by Google, sorted by clicks, and the rows that fall off the bottom are the rare ones.

The API goes further, not all the way. Its default is 1,000 rows and its maximum 25,000 per request, and it "does not guarantee to return all data rows but rather top ones" ^[7]. Google's guide to retrieving all of a site's traffic says the method exposes a maximum of 50,000 rows per day per search type, and that grouping by page or query may drop some data to keep calculation time reasonable ^[8]. Google does not use the word sampling; what it describes is truncation, keeping the top rows. Only the bulk export escapes the cap: it sends daily tables to BigQuery in the owner's Google Cloud project, starts up to 48 hours after configuration with no history, and incurs Google Cloud storage and query costs above a free usage level ^[10].

Sixteen months, three days and Pacific Time

Search Console keeps data for the last 16 months, and Analytics reports built on it inherit the limit ^[12]. The only way to keep a longer record is to export on a schedule and store the files, as the case study does ^[20].

The data is not live. Google says collected data should normally be available in two to three days, and a newly added site can take up to a week to generate data at all ^[5]. Fresher figures exist under a separate state in the API, with a warning that they are not yet final and may change ^[19]. Nor is the day a UK day: a time zone can be chosen in Analytics but not in Search Console, where the default is Pacific Time ^[11].

Google keeps a public log of events that distort the data; entries in 2026 include a logging error, resolved on 3 April, that had affected impression reporting from May 2025 to April 2026; the disappearance of FAQ rich results from 7 May, with a corresponding drop in impressions; and two days missing from bulk exports for some properties at the end of February ^[6]. Anyone comparing periods that straddle those dates is comparing two measurement conditions. Google's advice when comparing date ranges is to use weekly or monthly granularity ^[4]; the case study's 28-day windows, with weeks starting on Monday ^[20], follow the same logic.

Why Search Console and GA4 never agree

Google's explanation for the two organic numbers never matching is that they measure different events: a click in Search Console, a session in Analytics, and "clicks and sessions are calculated differently, which means that when you compare the data, you'll likely see different numbers" ^[11].

Google lists the reasons. Analytics depends on how its tag is implemented; if a site asks users to accept tracking and they opt out, the Analytics data is skewed; Search Console reports only on the Google Search canonical URL whereas Analytics reports on any URL carrying the tag; Analytics excludes known bots while Search Console does not necessarily filter them out; and the time zones differ ^[11]. The channel Analytics calls Organic Search is defined by rule, as traffic whose source matches a list of search sites or whose medium exactly matches organic ^[13]. Linking the accounts does not reconcile them ^[12].

There is a wrinkle in Google's own pages. Its general page on Search Console data says the tool does additional processing, for example to eliminate duplicates and visits from robots ^[5], while its page comparing the tools says Search Console does not necessarily filter bots out ^[11]. What matters is that the two tools should move in the same direction and should not be expected to match.

A ten-minute monthly routine

The routine below is the one I use on the clinic's own property. It uses nothing outside the standard report, and each step is a comparison rather than a single number.

1. Compare the last complete month with the previous one in the Performance report for web search, using the monthly granularity Google recommends ^[4]. Write down total clicks and impressions for both; these unfiltered totals include anonymised queries and are the only complete figures on the page ^[3].
2. Apply a saved query filter excluding your business name and its misspellings, and write down non-brand clicks and impressions for both months. These exclude anonymised queries and will not add up to the chart ^[4]; the direction of change is what you are recording.
3. Remove the filter, switch the table to Queries and sort by impressions. In the top twenty rows, mark the ones that are not your name and sit between roughly position 8 and 20. Those are searches where you are shown but rarely chosen ^[3], and they are the month's shortlist.
4. Switch to Pages, compare the two months and sort by the Difference column ^[4]. A page that lost a large share of its clicks is a technical problem, a changed page, a competitor's improvement or a measurement event; check the anomalies log first ^[6].
5. Open the Business Profile performance report separately and note calls, direction requests and website clicks for the same month ^[14]. Search Console does not count Maps ^[20].
6. File the numbers in a spreadsheet with the date. After 16 months you hold data Google will no longer give you ^[12].

The routine leaves average position out. As a monthly figure it is dominated by the brand query, blended across every device and location, and moved by changes in which queries the site appears for at all ^{[2] [9]}.

What Search Console cannot tell you

The report has a boundary, and most disputes about agency reporting live there. Search Console describes appearances in Google Search, Discover and News ^[2]. It does not describe Maps or the Business Profile, which have their own report counting searches, views, calls, direction requests and website clicks, with views counted once per person per day ^[14]. It does not describe other search engines or AI assistants ^[20]. And it stops at the click: whether the visitor called or booked belongs to the analytics tag, which depends on consent and implementation ^[11].

This boundary is the test for any report from any provider, including HighRegard. A report that quotes clicks without saying whether they are brand or non-brand is quoting the marketing budget as well as the search work. A report that headlines average position is headlining a figure Google itself calls potentially misleading ^[2]. A report claiming every query is either using the bulk export or overstating ^[3]. And a report that does not come with access to the underlying property is asking to be taken on trust.

The safeguard is ownership. An owner who holds the property, exports monthly and keeps the files can check any report against the source, the standard the case study sets for itself by publishing fingerprints ^[20].

What this means for a local business

A local business is affected by these rules more than a national site is. Its brand query is a large share of its total, so the blended average position and the unfiltered clicks line say more about reputation and offline marketing than about search visibility. Its valuable searches are specific and local, which places many of them below the threshold at which Google anonymises the query ^[3] and inside the roughly half of clicks that independent measurement found hidden ^{[16] [17]}. Its customers act on Maps and the Business Profile, which Search Console does not count ^[20].

The consequences are modest and mostly free. Hold the property in the business's own Google account. Compare whole months or 28-day windows, never days. Split brand from non-brand before judging a trend, and accept that the split will not add up to the total ^[4]. Read impressions as Google's willingness to show the business. Keep the Business Profile report beside Search Console ^[14], and export monthly, because the window closes at 16 months ^[12]. None of this needs software or a consultant.

What the evidence does not settle

Several things here rest on inference. Google has not published the threshold at which a query becomes anonymised, nor the ranking that decides which rows survive its "most important data rows" limit ^[3]. The independent estimates of the hidden share come from companies with a commercial interest in the answer, and the largest measured April 2025 and expected the share to rise as searches lengthen ^[16]; whether it has since risen is not, as far as published measurement goes, known. Google's own pages differ on whether Search Console removes bot traffic ^{[5] [11]}, and its documentation on how AI Overviews and AI Mode are positioned is one sentence each ^[2]. The anomalies log shows impression reporting was affected by a logging error for nearly a year before its correction in April 2026 ^[6], so some comparisons made in that period set a faulty count against a corrected one.

The case study I published is one business in one market, and the same work elsewhere will produce different numbers ^[20]. It shows clicks, impressions and position moving together after the work began; it cannot separate the work from a new site maturing in the index or from the other marketing that ran alongside ^[20]. What Search Console can do, read carefully, is put an honest boundary around what is known. Where that boundary sits after the next change to Google's results page is something the report will only show in retrospect, and two to three days late ^[5].