66% of our Google impressions came from searches no one ever typed

90 days of Search Console data for geoa.app: 16,784 impressions, 7 clicks — and two-thirds of the impressions came from 19 search terms that no human has ever typed. Here’s how to distinguish bot traffic from customers in your own console.

We analysed 90 days’ worth of Google Search Console data for geoa.app: 16,784 impressions across 179 keywords — and 7 clicks. When we broke down the keywords, 66.5% of all impressions came from 19 keywords that no human had ever typed. They ranked on average at position 4.2 on page 1 and generated exactly zero clicks. Your own console probably mixes these two types of traffic together — here’s how to separate them.

It’s one of the most common panic emails we receive: “My impressions are rising, but my click-through rate has plummeted. What have I done wrong?” Often, the answer is: nothing. It’s just that something new has crept into the figures that isn’t actually customers.

What do the figures show?

We used our own dataset, as it’s the only one we’re allowed to share in full detail. The period is 1 May to 29 July 2026, the source is Google Search Console for geoa.app, and the breakdown is at keyword level. An impression is counted every time your link appeared in a search result — Google itself describes the definition here — so it doesn’t take a click or a human to make the figure go up.

We sorted the keywords into two groups according to a single, easy-to-follow rule: does the search contain our own domain name within a longer phrase, or does it use search operators? If so, it’s machine-generated. Everything else is ‘the rest’.

Group	Search terms	Impressions	Share	Average position	Clicks
Machine-shaped searches	19	11,155	66.5%	4.2	0
The rest	160	5,629	33.5%	15.1	7
Total	179	16,784	100%	10.9	7

Note the ranking. The machine-generated searches are ranked 4.2 — at the top of page 1 — whilst the genuine searches are ranked 15.1, i.e. on page 2. If you only look at the overall average position of 10.9, it looks as though the site is performing well. It’s just doing so for the wrong searches.

This isn’t a fault in Search Console. Google’s average position is defined as an average across all impressions, so a cluster of 11,155 impressions around position 4 inevitably drags the whole average down with it. The tool is doing exactly what it’s supposed to — it just doesn’t realise that the two clusters are vastly different things.

What does a machine-generated search look like?

Here are the top five from our own console, quoted exactly as Search Console recorded them. They are Danish-language queries and are left untranslated, because the wording is the evidence:

Search	Impressions	Position	Clicks
billigste geoa.app danske virksomheder	4,309	4.2	0
bedste danske virksomheder til geoa.app	2,767	3.8	0
bedste geoa.app danske virksomheder	2,685	4.0	0
geoa.app anmeldelser danske brugere	898	5.3	0
danske virksomheder anbefalet til geoa.app	358	3.9	0

Read them aloud. “billigste geoa.app danske virksomheder” isn’t Danish. It’s a template where a domain name has been inserted into a gap: cheapest [X] Danish companies. The same search was displayed 4,309 times over 90 days — around 48 times a day, every day, for three months, without anyone ever clicking on it.

The second type is even more obvious. Among the 19 were also searches such as:

• "brand24" -site:reddit.com -site:twitter.com -site:x.com
• "brand24" -"24-7" -"24/365" -"24/7" -"brand 24 apparel"
• ("kunstig intelligens" or "ai" or "maskinlæring") and ("kosm…

No customer writes Boolean logic with exclusion operators in Google. These are tools that do the work. However, the twelve operator searches combined accounted for just 91 impressions — the bulk of the volume lay in the seven template searches, which together accounted for 11,064 impressions and zero clicks.

Who is behind these searches?

The honest answer: we cannot determine that. Search Console tells us which search query triggered an impression — not who or what sent it. We can see the signature, not the sender.

There are at least three plausible explanations, and they are not mutually exclusive:

Explanation	What speaks in favour	What speaks against it
AI engines’ own lookups (fan-out behind an AI response)	The phrasing resembles the sub-questions an AI asks to formulate a response	We have nothing linking the search to a specific AI response
Monitoring and ranking tools	The operator searches are textbook examples of media monitoring	Fail to adequately explain the 4,309 impressions on a single template search
Competitor analysis	Two of the search terms contain competitors’ brand names	Very low volume — 91 impressions in total

We could have written “Google is now sending AI searches into your Search Console” and come up with a more exciting headline. But we can’t show that, and we’ve seen how badly things go when someone makes a guess and presents it as fact. Nor is the sender the interesting part. What is interesting is that the traffic is there, that it accounts for two-thirds of the total, and that it looks like genuine customers in any report you pull.

Why does this ruin your reporting?

Because the three figures most people look at all become incorrect at the same time — and they point in different directions:

• Impressions are rising. It looks like growth. It isn’t.
• The click-through rate is falling. In our figures, the overall click-through rate is 0.042 per cent. For ‘other’ traffic alone, it’s 0.124 per cent — three times better. The bot traffic is dragging the average down without having done anything wrong.
• The average position is improving. The automated searches rank at position 4.2 because they’re ultra-specific and face almost no competition. This pulls the average up from 15.1 to 10.9.

Note how toxic this combination is: your dashboard simultaneously shows more impressions, a better ranking and a lower click-through rate. This is precisely the pattern people tend to interpret as ‘Google is showing me more, but my content is missing the mark’. That conclusion leads people to start rewriting titles and descriptions that weren’t actually at fault.

How to clean up your own figures in ten minutes

You don’t need a tool for this. The procedure is the same as the one we used:

1. Open Search Console → Results → Search Queries and set the time range to the last 3 months. Sort by impressions.
2. Look for your own domain name within a sentence. Searches such as “best [yourdomain.dk] Danish companies” are templates, not natural language. No customer would type your domain name in the middle of a sentence.
3. Look for operators: quotation marks, a minus sign before words, site:, -site:, or and and lower-case letters. Anything of that sort is a tool.
4. Look for zero clicks at high rankings. A keyword in positions 1–5 with hundreds of impressions and zero clicks over three months isn’t a bad title. It isn’t a human.
5. Add up the keywords you’ve found and subtract them from your total. Recalculate your click-through rate on the remainder. That’s the figure you should report on and base your decisions on.

Make this a regular quarterly exercise rather than a one-off clean-up. The pile is growing, and it grows fastest on small sites, where a few hundred bot impressions can account for the majority of all impressions.

Does that mean impressions are now useless?

No — but they should be interpreted as two figures rather than one. Impressions from humans still tell you how people are finding you. Impressions from machines tell you something else, and something useful too: that your site has become a reference point for a system that’s researching the market. We’re listed under ‘bedste danske virksomheder til geoa.app’ because something out there is trying to work out who we are and who’s similar to us.

It’s not turnover. But it’s not just noise you need to filter out either — it’s a different metric that simply shouldn’t be added to the first one. The same distinction applies in our previous analysis of what determines whether AI mentions your company: being used as a source and being recommended as a company are two different outcomes, and they must be measured separately.

What this doesn’t show

This is data from a single website — our own — over 90 days. It is an observation, not a market study, and we cannot say whether 66.5 per cent is high or low for a typical small or medium-sized business.

Specifically, we do not know: who is generating the searches, whether the proportion is just as high on larger sites (it is probably lower, because genuine traffic accounts for a larger share), or whether the pattern remains stable over a period longer than a quarter. Our own click volume is low enough that the click-through rates in this article should be interpreted as orders of magnitude, not precise figures — 7 clicks is too small a sample for statistical analysis.

We’ll repeat the measurement next quarter and publish the result, whatever it shows. Should you run the exercise on your own site, we’d love to hear what you find — that’s how we’ll get a proper sample rather than just one.

Frequently Asked Questions

Why are my impressions increasing whilst my click-through rate is falling?
Often, this is because some of the impressions aren’t coming from people at all. In our own figures, 66.5% of all impressions came from 19 machine-generated search terms with zero clicks. They drive the click-through rate down and the average position up at the same time, even though there’s nothing wrong with your titles or content.

What does a machine-generated search look like in Search Console?
There are two recurring patterns. One is template phrases with your own domain name inserted in the middle of the sentence. The other is search operators: quotation marks, a minus sign before a word, site: and -site:, as well as or and and in lower case.

Who is behind these machine-generated searches?
We cannot determine this from Search Console, which only shows the search string and not the sender. There are at least three plausible explanations, which are not mutually exclusive: AI engines’ own lookups to generate an answer, monitoring and ranking tools, and competitor analysis.

How do I clean up my own figures?
In the Search Queries report, find the search terms that contain your domain within a phrase, use operators, or appear in positions 1–5 with hundreds of impressions and zero clicks. Add them together, subtract them from your total, and calculate the click-through rate for the remainder. That’s the figure you should report.

So should I ignore impressions altogether?
No — but they should be treated as two figures rather than one. Impressions from humans tell you what people are finding you for. Impressions from machines tell you that your site has become a reference source for a system that’s researching the market. It’s not revenue, but it’s not noise either — it just mustn’t be added to the first figure.

May I quote the figures?
Yes. Please cite “Geoa, July 2026, geoa.app, n=179 search terms over 90 days” and feel free to link to this page. We note the date of each update, so older citations can be traced back to the version from which they were taken.