Position 1, Not a Single Click: How to Spot Machines Searching for Your Website

The Google Search Console for this website contains a search query that looks like this:

"chatbot" -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

Eleven platforms excluded, the search term in quotation marks. The website ranked at position 1 for it. Nobody clicked.

The Bing data for the same website holds a second query:

anthropic/claude-opus-4-8#N##N#heute um 15:04#N##N#klar, hier ein paar alternative überschriften mit anderem thematischen fokus:#N##N##N##N#klassiker mit netz

The #N# is an encoding for a line break. A search box does not accept line breaks; it submits. This query did not come from a browser. It came from a program passing a piece of text along.

In brief: A growing share of the search queries a website is served for no longer comes from people but from AI systems researching on a person's behalf. They are recognisable by operator syntax, by characters a search box will not accept, and by conversational fragments with no topic. In this website's data, that class grew from 2.95 to 23.1 impressions per day between 2 August and 7 September 2026, a factor of 7.8, while total volume grew only by a factor of 3.5. It sits at an average position of 6.58 instead of 10.03, yet produced exactly one click across 218 impressions. The sharpest comparison comes from the same period on the same page: the human typo faible 5.1 returns a click-through rate of 17.6 percent at position 3.02, while two queries carrying a full model identifier in quotation marks return zero clicks across 54 impressions at position 1.98. Click-through rate therefore no longer measures whether a page convinces. It measures what share of demand is human.

Which queries could no person have typed?

The question sounds like hair-splitting, but it underpins everything that follows. Anyone who cannot separate the human part of their visibility from the machine part is reading their Search Console incorrectly.

Three features rule out a human origin in practice.

First, operator syntax on a scale nobody types by hand. The query quoted above contains eleven consecutive -site: exclusions. That is roughly 200 characters of pure filter logic for a single search term. A person wanting to keep forums out of their results types two or three exclusions and then gives up.

Second, characters a search box will not pass through. The encoded line break in the Bing example is the clearest piece of evidence in the entire dataset. A query containing line breaks does not arise from typing, because the return key triggers the search. It arises when a program hands over a multi-line text as a search term. The embedded timestamp "heute um 15:04" and the half-finished sentence reveal the origin: a chat transcript in progress.

Third, conversational leftovers with no topic of their own. The past two weeks of data include also bis wann?, ja beides, wie nutzt man es?, when did it come out?, er det lovligt, daj link do artykulu and что ты путаешь. Translated: "so by when?", "yes both", "how do you use it?", "is it legal", "give me the link to the article", "what are you confusing". The presence of Danish, Polish and Russian has a simple explanation: only 42.5 percent of this website's impressions go to Germany, the rest spread across India, Korea, Spain, Greece, Norway and a dozen other countries. Its articles on new AI models are served worldwide, and someone asking their assistant about a model in Danish produces a Danish row in a German Search Console.

Not one of these is a search query. Each is an answer or a follow-up inside a dialogue. They appear in Search Console because an assistant passed the conversational turn verbatim to a search engine.

Taken individually, each feature can be explained away. Some people do know their operators, and nonsense input exists. Taken together they form a pattern that can be measured cleanly.

How large is the share, and how fast is it growing?

For this analysis I compared two non-overlapping windows and converted both to impressions per day, because they differ in length. A row counts as machine-shaped if it begins with a plus or percent sign, contains quotation marks, or carries a -site: exclusion. That is a deliberately narrow definition: conversational fragments and fully worded briefs fall outside it, because they cannot be captured reliably by automatic means. The true figure is therefore higher than the one reported here.

The factor of 3.5 in total volume has an identifiable cause: a new AI model was released on 1 September, and this website reported on it early. Part of the increase is pure news traffic and says nothing about a structural shift.

That is precisely why the comparison matters. The machine-shaped class grows at 7.8x, more than twice as fast as total volume. It is therefore not merely riding the news surge; it is gaining share on top of it. From 0.61 to 1.36 percent within five weeks.

One percent sounds like nothing. At that rate of increase, the class reaches double digits within six months.

Why do machines rank better and still not click?

The finding looks contradictory at first: the queries that bring no visitors are the ones that rank best.

The better position has a straightforward explanation. A search system can serve a precise query precisely. Putting "chatbot" in quotation marks and excluding eleven platforms narrows the result set so far that specialist pages rise to the top. The same holds for a query such as "us.anthropic.claude-fable-5-1": for a complete model identifier in quotation marks there may be a dozen matching pages worldwide. Run one of them and you sit at position 2.

The missing click has an equally simple explanation: there is nobody at the other end to click. The system reads the result list, fetches the page directly where needed, and composes an answer inside the chat. The retrieval happens. It simply does not run through a counted click but through a separate fetch that Search Console does not record as one.

This shows most sharply in a paired comparison from the same period on the same domain.

Four queries, all near the top, all pointing at the same page. The upper two are recognisably human, one of them a typo. Together they return 12 clicks across 57 impressions. The lower two are complete technical identifiers in quotation marks. Together they return zero clicks across 54 impressions.

Near-identical impression counts, near-identical positions, the same destination page. The only difference is who was asking.

What Google has shown since 31 August, and what it has not

Google has responded to this development. On 3 June 2026 the company announced dedicated performance reports for generative AI features on the Search Central blog, and the rollout to every property was completed on 31 August 2026.

What the report shows: impressions inside AI Overviews and AI Mode as well as generative features in Discover, broken down by page, country, device and date.

What it does not show: clicks. Click-through rate. Position. And above all, the queries. It is also unavailable through the Search Console API and can only be exported by hand.

For a managing director wanting to know whether the effort pays off, that is an unsatisfying answer. The report confirms that your page appears in AI answers. It withholds which question prompted it, with what result, and whether a conversation ever came of it.

What is notable is what Google concedes implicitly with that selection. A report that surfaces impressions and omits clicks is saying: impressions exist, clicks in meaningful numbers do not. Otherwise they would be in there.

The analysis presented here therefore takes a different route. It looks for the machines not in the AI report but in the ordinary query report, where they remain identifiable by the shape of their queries. There the queries appear verbatim, and there position and click-through rate are available.

Why AI systems exclude Reddit of all places

The eleven exclusions in the opening query are no accident. Excluded are Reddit, Twitter, X, Wykop, Tripadvisor, YouTube, Yelp, Booking, Facebook, Instagram and TikTok. Forums, social networks, review and booking portals.

That inverts what many visibility guides have said over the past two years. The argument there was that AI systems cite Reddit disproportionately often, so that is where you should be. For fan-out queries, with which an assistant verifies facts, the opposite now applies: for a factual question these platforms deliver a great deal of text with little verifiable content, and the systems actively filter them out.

Published analyses of ChatGPT fan-out queries show the same movement from the other side. The site: operator, which restricts a search to exactly one domain, rose there from around 0.3 percent to about 23 percent of all fan-out queries after ChatGPT 5.6 became the default model; a second analysis reached 64 percent in its own dataset. Fan-out queries per user prompt rose from 2.17 to 7.61, and the longest observed chain went from 4 to 29 individual searches. Among the most frequent added terms are "official" and "gov".

Taken together this gives a clear picture of how these systems behave: they exclude platforms with a low density of evidence, they target domains they credit with authority, and they ask not one question but seven or more.

What this means for your own domain

The most practically important point sits in the site: operator. When an assistant is asked to answer a question about your company, it increasingly searches not the open web for you but your domain specifically.

That shift has an uncomfortable consequence. In an open search, a system can fill gaps from third-party sources: a trade directory, a press article, an old profile. In a search restricted to your domain, that fallback does not exist. What is not there does not exist for the purposes of that check.

In concrete terms: a system asked to verify whether you are the right provider for a particular need will search your domain for scope of service, area covered, target audience, named contact and explicit limits. If it finds three of those five, it will either leave you out of the answer or name you with a caveat.

The gaps that turn up most often are strikingly mundane:

• Who the service is for. Not "companies of every size", but the actual range you are willing to be measured against.
• Where it is delivered. On site, remotely, within what radius, and what that means in practice for working across longer distances.
• What triggers it. Which problem inside a company leads someone to look for this service in the first place.
• What is not included. An honest limit is worth more to a verifying system than another list of strengths, because it answers a question conclusively.

None of this is a search-engine measure in the classical sense. These are the details an attentive prospect looks for anyway. The difference is that a person fills gaps from context or asks a follow-up question. A system does neither. It moves on.

How do you find these rows in your own data?

The test takes a few minutes and can be reproduced on any property.

Open the performance report in Search Console, add a new filter on "Query" and choose "Custom (regex)". Then enter each of the following in turn:

Note impressions, clicks and average position for each group. What matters is not the individual figure but the comparison between the three groups and the movement across two time windows.

Two notes on the analysis. Convert your windows to impressions per day, or you will be comparing periods of different length. And read the rows themselves, not just the totals. The insight sits in the wording: it shows which question somebody asked before an assistant turned it into a search.

For background: the observation that fully worded briefs appear in Search Console at all was set out in detail in this article on brief-style prompts at the end of August. Back then that group averaged position 59.4. In the current window it sits at 18.6. The impressions have stayed the same, the position has improved by 40 places, and still nobody has clicked. That too is a finding: a position can be improved; a machine sender cannot be made to click.

What metric replaces click-through rate?

If a growing share of demand structurally does not click, then click-through rate increasingly measures one thing only: the human share of senders. When it falls while impressions and position rise, that is not a quality problem with the page.

That is exactly the case in this website's data. Between the two windows, volume rose by a factor of 3.5, average position improved from 10.31 to 9.72, and click-through rate fell from 2.08 to 0.50 percent. Read the last figure alone and it looks like a collapse.

Three metrics still carry meaning.

The number of distinct queries you are served for. It shows how many different situations your pages are considered a possible fit for. This number rises before positions rise, and it rises long before clicks arrive.

Position split by class. The overall position blends two groups with entirely different behaviour and is therefore barely interpretable. Split apart, you can see whether the work on your content is landing where decisions are actually made.

The number of first contacts who name an AI assistant as their source. This number appears in no tool, because there is no referrer for it. It exists only if somebody asks, in the first conversation, how the contact came about. For enquiries with no traceable origin in the statistics, that question is the only reliable source.

The third is the most inconvenient and the most important. It is the only direct evidence that machine visibility turns into business.

What to do now

Four steps, ordered by effort.

Measure once how large your machine-shaped class is. The three regex filters from the previous section, two time windows, half an hour of work. Without that figure, every further measure is guesswork.

Introduce the origin question into your first conversation. One line in the call guide. Without it the entire AI channel stays unprovable, however good the impression figures look.

Check your most important pages against the five details. Scope, area, audience, contact, limits. Not spread across several pages but complete on each one, because a search restricted to your domain examines individual pages, not your website as a whole.

Stop reporting overall click-through rate as a measure of success. As long as the machine-shaped class grows, that number falls even when everything is going right. Keep it in the monthly report and you will regularly have to present good work as a setback.

Where the line runs between technically possible and commercially sensible depends on the individual case. If you would like to know what your own figures look like and what follows from them, we can go through it together in a free initial consultation.

Frequently asked questions

How can I tell that a search query came from a machine? By three features that almost never appear in a search box. First, operator syntax: quotation marks around whole phrases, a leading plus sign, several minus-and-site exclusions in a row. Second, characters a search box will not even accept, such as encoded line breaks or a timestamp in the middle of the query. Third, conversational fragments with no topic of their own, such as "yes both" or "so by when?", which only make sense inside a dialogue. Any one feature is weak on its own; together they are conclusive.

Why do machine-shaped queries rank better yet never click? They rank better because a search system can serve a precisely worded query with quotation marks and exclusions more accurately than a vague keyword. In our data this class sits at position 6.58 against 10.03 across the whole property. The click never comes because there is nobody at the other end to click. The system reads the result list, fetches the page directly if it needs to, and delivers the answer inside the chat. The visit happens; it simply is not counted as a click.

Does Search Console not show these AI impressions by now? Partly. Google announced dedicated performance reports for generative AI features on 3 June 2026 and completed the rollout to every property on 31 August 2026. Those reports show impressions from AI Overviews and AI Mode, broken down by page, country, device and date. They show no clicks, no click-through rate, no position and no queries, and they are not available through the API. The report confirms that something happened, not what for and with what result.

Why do AI systems exclude Reddit and YouTube of all places? Because forum and video platforms return a great deal of text with little verifiable substance for a factual question. Our data contains one query that excludes eleven platforms at once, among them Reddit, X, YouTube, Facebook, Instagram and TikTok. Published analyses of ChatGPT fan-out queries show the same movement from the opposite direction: the site operator, which restricts a search to a single trusted domain, rose there from roughly 0.3 to about 23 percent of all fan-out queries.

What does this mean for my own website? When an AI system is asked to answer a question about your company, it increasingly checks your own domain specifically. What is not there, it cannot confirm, and what it cannot confirm it either omits from the answer or replaces with an older figure from a third-party source. Scope of service, area covered, target audience, named contact and explicit limits therefore belong in plain text on your own pages, not only in a contact form or a brochure.

What metric replaces click-through rate? No single one, but three together still carry meaning. First, the number of distinct queries you are served for at all: it shows how many situations you are considered a possible fit for. Second, position within the machine-shaped class, measured separately from the rest. Third, the number of first contacts who say they found you through an AI assistant. That third number appears in no tool; it has to be asked for in the first conversation.

Do I now have to write for machines instead of people? No, and attempting it would backfire. What machine-shaped queries reward is exactly what rewards a reader in a hurry: a clear statement up front, verifiable figures with a date and a source, clean headings, and no claim without evidence. The difference lies not in style but in completeness. A person fills gaps from context; a system does not.

Is this worth the effort at barely one percent of impressions? The share is small today; the direction of growth is not in doubt. In our data this class grew almost eightfold within five weeks, while total volume grew three and a half times. What matters more than the share, though, is the intent behind it: someone asking an assistant for a provider is closer to a decision than someone looking up a technical term.

Sources and status

All figures relating to this website come from Google Search Console for the property arkeontech.de, retrieved on 7 September 2026. The time windows are stated in each case. Classification used the regular expressions given in the section on detection, which makes it reproducible on any other property.

The queries quoted from Bing data come from Bing Webmaster Tools for the same property, likewise retrieved on 7 September 2026. They are reproduced verbatim, including the lower-case spelling and the #N# encoding for line breaks.

The details of the performance reports for generative AI features come from the announcement on the Google Search Central blog dated 3 June 2026 and from reporting on the completion of the worldwide rollout on 31 August 2026.

The figures on ChatGPT fan-out queries, that is the rise of the site operator from around 0.3 to about 23 percent, the diverging figure of 64 percent from a second dataset, and the increase in fan-out queries per user prompt from 2.17 to 7.61, come from a published analysis by Lily Ray, retrieved on 7 September 2026.

How fully worded briefs to AI assistants appear in Search Console, and what distinguishes them from keyword queries, is covered in the article on brief-style prompts. Why a company can go unmentioned in AI answers despite good positions is examined in the article on the mention threshold. Which sources AI systems actually assemble their answers from is shown in the article on the sources behind AI answers.