6.9 Million Impressions, One Click: What Was GSC Actually Counting?

As of 31 October 2026 · Google Search Console data from 23 sites

One German-language query. Between January and 20 September 2026 it brought my client’s site 6,935,544 impressions and one click. Just the one. From 21 September, impressions for that query fell from around 200,000 a week to 14,000, and a week later to 116. Over three weeks, the whole site lost 42% of its impressions and 8% of its clicks.

My first reaction was: what broke? But that wasn’t the question to ask. The right ones were: what exactly had we been measuring for those nine months? And is the drop in impressions after September something to worry about?

Clue one: 225 “countries” and a bot that switches devices

I started with countries. From February to September, the query had impressions in 190⁠–⁠225 countries and territories every week. There are fewer countries than that, but Search Console counts places like Greenland, Puerto Rico and Hong Kong separately. Most impressions came from Brazil, Saudi Arabia, Morocco and South Africa. The result usually sat on page one. Clicks: none. Interesting.

Then devices. This is where it got funny:

From desktop to mobile and tablets in two weeks, almost nothing a week later. One German-language query on a client’s site. Source: Google Search Console.

Entire countries don’t switch from desktops to tablets in two weeks. A bot can – you just change its settings when fetching results on one device becomes a problem. From the week of 28 September, 100⁠–⁠130 impressions a week remained, almost all on desktop, with an average position of 2⁠–⁠3. Maybe that’s what genuine interest in this query looks like?

Clue two: the same template on 17 sites

I found the second clue in queries that look human, only too long. They follow this template:

"phrase" -site:reddit.com -site:twitter.com -site:x.com -site:wykop.pl -site:tripadvisor.com -site:youtube.com -site:yelp.com -site:booking.com -site:facebook.com -site:instagram.com -site:tiktok.com

The same template shows up on 17 of the 23 sites I can see in Search Console: online shops, blogs, local services and my travel guides. In total: 594 queries, 5,303 impressions and zero clicks. Of those 5,303 impressions, 5,301 were on desktop. On 13 sites the template first appeared in March or April 2026. (Wykop, on the exclusion list, is a Polish Reddit-style site.)

The operator alone proves nothing. People type -site: too. Maybe not everyone, but SEOs certainly do – I have a few searches like that bookmarked myself. But someone who, across hundreds of phrases and 17 sites, manually excludes the same eleven websites and never clicks anything? More likely a tool checking what Google shows outside Reddit, YouTube and Booking.

The best one turned up on my Bornholm travel guide: "ewelina podrez" -site:reddit.com … Someone is monitoring my name, and the guide pops up in their results. I was about to write “warm regards, whoever you are”, but it turned out I was sending my regards to myself – it’s my own Brand24 project. The tool that tells me where people write about me was, as a side effect, adding impressions to my sites in Search Console. Good thing I figured that out before publishing this post.

Traces of the Brand24 queries stop around 10 September, a few days before the blocking reported by other tools. Whether Google cut off this route or Brand24 changed it itself, I don’t know. I still see mentions from websites, news and blogs.

Clue three: what comparing 19 sites showed

A single query on one client’s site is an anecdote. So I checked every site with enough data to measure: 19 of 23, because four have fewer than 300 desktop impressions a week. This isn’t a representative sample of the market, just the shops, blogs, local services, B2B companies, travel guides and the agency I have access to.

Desktop: in the week of 21 September the median drops from 100 to 64 and stays around 60 until the week of 19 October. Mobile: no step change. Each grey line is one site, the red one is the median of 19 sites. 100 = average of the three weeks before 21 September.

The median change in clicks (three weeks before 21 September vs three weeks after) is −3.6%. 12 of the 19 sites show the full pattern: fewer desktop impressions, a “better” average position and a smaller desktop share of all impressions.

It’s most visible at Fox Strategy, the agency I founded. In the week of 21 September it lost 46% of its desktop impressions. Average position “improved” from 18.1 to 15.1, and there were more desktop clicks than the week before: 35 instead of 26. Looking only at impressions, I’d have a crisis. Looking at clicks, I’d had a pretty good week. That said, Fox Strategy also migrated to a new website in early September, so this case alone would be a weak basis for conclusions. That’s why I show it alongside the other 18.

What was happening on the other side of the results

Humans aren’t the only ones “looking” at Google results. Rank trackers, scrapers and data vendors for AI companies query the search engine non-stop, and Google has been fighting this for a long time. In January 2025 it started requiring JavaScript, citing protection against bots (TechCrunch). In December 2025 it sued SerpApi, a company that sells Google results collected by bots. According to the lawsuit, this meant hundreds of millions of fake queries a day (Google blog).

That some of this traffic ends up in Search Console has been known for a while. In 2021, John Mueller said the report doesn’t filter out all bot traffic (SER). The strongest clue came in September 2025. Around 10⁠–⁠12 September, the &num=100 parameter, which trackers used to fetch 100 results at once, stopped working. In Tyler Gargula’s analysis, 87.7% of 319 sites lost impressions, and the average position of many of them suddenly “improved” (SER). On my sites, 11 of 18 showed the same pattern back then.

And September 2026? From around 13 September, Google has been blocking scrapers much more effectively. Nozzle lost about 80% of its data and Sistrix reported limited data collection (SER). Google also started feeding some tools substituted results. Nozzle checked 56 million results pages from 53 countries: from 13 September, an average of 8.7% of first pages were poisoned, compared with 0.17% before (Nozzle).

In Poland, this was described in detail by Damian Sałkowski of Senuto, a Polish SEO platform. From 19 September the tool was getting pages completely unrelated to the query, and from 21 September Google introduced new safeguards that made fetching results much harder. For the query “plastic fruit crates 20 kg”, Senuto got, among other things, tram no. 20 in Wrocław, a contactor and a glue with “20” in its name. The company paused publishing data, introduced a quality gate on 10 October and has been collecting data normally again since 13⁠–⁠14 October (Senuto blog, in Polish).

21 September. The same day Senuto starts seeing Google’s new safeguards is the first day of the week in which desktop impressions disappear on my sites and the German query drops from 200,000 to 14,000.

What fits and what I don’t know

Quite a lot fits, though not all on the same day. The desktop drop on my sites falls in the same week Senuto starts seeing Google’s new safeguards. The Brand24 traces stop earlier, around 10 September, so that thread shows where such impressions may come from rather than explaining the date of the drop. The device fits too: it’s mostly desktop impressions that disappear, and queries with the Brand24 template are 99.96% desktop. Clicks hold steady. Average position “improves”, just like after num=100, which would be consistent with impressions from distant positions disappearing.

The question I’m left with

How many of our SEO success stories were told to us by the tools that measure SEO?

A similar question came up after num=100. Brodie Clark wondered then whether part of the “great decoupling” of impressions and clicks, which the industry attributed to AI Overviews, was simply down to scraping (brodieclark.com). I think he’s not far from the truth.

For me, that’s the most interesting consequence of the whole story. A tool built to measure visibility may itself have been adding impressions to the report we later used to judge that visibility. More queries could mean a bigger chart. Fewer successful queries – a drop.

This doesn’t mean AI Overviews don’t take clicks, or that every big change in Search Console can be explained by bots. It does mean that a gap between impressions and clicks alone isn’t enough to assign a cause. First, you need to know what changed in each of those numbers.

With 6.9 million impressions and a single click, asking “why did we lose visibility?” seems premature to me. First, I’d like to find out whose attention we had in the first place.

If you’re looking at a similar chart right now

1. Split the data by date and device. If it’s mostly desktop impressions that fall while clicks and mobile hold steady, check whether it’s a reporting change before you treat it as lost traffic.
2. Check countries and top queries. One query from dozens of countries with no clicks is probably not your customers.
3. Type -site: into the query filter and look for a repeating template.
4. If clicks fall on desktop and mobile at the same time, that’s a different story. Then look for causes in rankings, demand and specific pages.

If you rely on data from SEO tools, also read Nozzle’s research on poisoned Google SERPs.

And if you see a similar trace on your own sites – almost click-less queries from all over the world, the same string of exclusions or a sudden desktop drop in the week of 21 September – let me know. I’m curious what this looks like beyond my 23 sites.