query-fan-out-search-console
19 Sep 2026
10 min read
Error: Invalid Frontmatter
Path: D:/inetpub/webs/castaldosolutionsit/articles/user/pages/01.blog/79.query-fan-out-search-console/item.en.md
Failed to read D:/inetpub/webs/castaldosolutionsit/articles/user/pages/01.blog/79.query-fan-out-search-console/item.en.md: Unexpected characters near "s behalf.'" at line 7 (near "excerpt: 'From March to August this blog tripled its impressions and its clicks stayed flat. Looking closely, the explanation is not a ranking drop: a growing share of whoever finds us is not a person, it is an assistant searching on somebody else's behalf.'").
---
title: "One in three searches that find you is not a person"
seo_title: 'Query fan-out: the traffic that never clicks'
date: '2026-09-19 10:00'
author: Gaetano Castaldo
metadata:
description: 'Blog impressions go up, clicks do not. The reason has a name, query fan-out: on this site it went from 7% to 30% in six months. Here are the numbers.'
excerpt: 'From March to August this blog tripled its impressions and its clicks stayed flat. Looking closely, the explanation is not a ranking drop: a growing share of whoever finds us is not a person, it is an assistant searching on somebody else's behalf.'
taxonomy:
category:
- marketing
tag:
- geo
- ai-seo
- content-marketing
- business
image: query-fan-out.webp
image_alt: "A magnifying glass that lights up many pages in parallel instead of a single one"
---
When an AI assistant prepares an answer it does not run one search: it runs many at once, across different subtopics. Google calls this **query fan-out**, and those searches land in your Search Console impressions mixed in with the ones made by people. On this blog, from March to August 2026, they went from 7% to 30%.
This is not a forecast and not an article about what will happen. It is what the numbers of a real company blog show after a year of publishing.
## What happened to this blog's impressions in six months?
I split the searches by length. Short queries, one to three words, are what a person types: "claude skills", "crm for smes". Queries of six words or more have a different shape, and I will come back to that.
Total impressions more than tripled, from 4,691 to 16,255. But the part growing fastest is the blue one: from 327 to 4,924 in six months, that is from one impression in fourteen to almost one in three. In September, across the first sixteen days, the share holds at 29.7%. The growth was not steady: after the 22.7% peak in May the share fell back to 20.4% and stayed there for two months before climbing again.
Where you cut changes the result, and it is worth saying so instead of picking the most convenient number. If you count every query of four words or more as long, they are **48.2% of impressions**: almost half. From six words up they are 26%. From eight, 16.8%.
## Why do those impressions produce no clicks?
Because on the other side, almost always, there is nobody who can click.
Short queries return a **2.58% CTR**. Queries of six words or more return **0.11%**: 11,669 impressions and thirteen clicks in ninety days. It is not a position problem, because on many of those queries we sit between fourth and ninth place.
| Minimum query length | CTR | Share of impressions |
|---|---|---|
| 1 to 3 words | 2.58% | |
| 4 words or more | 0.62% | 48.2% |
| 5 words or more | 0.19% | 34.4% |
| 6 words or more | 0.11% | 26.0% |
| 7 words or more | 0.05% | 18.9% |
| 8 words or more | 0.05% | 16.8% |
The proof that these are not people is geographical. The single biggest query on the whole site, 2,889 impressions in ninety days at position 4.1 and one single click, comes from Pakistan, Germany, the United Kingdom and Hong Kong, at twenty or thirty impressions per country. Nobody in Karachi is searching for automation for Italian SMEs. It is the same question opened from several places at the same moment, which is exactly what an engine does when it is gathering material.
And they are shaped like a prompt, not like a search: `roi levers and quantification methods for ai in pharmacovigilance payback period case studies`. Nobody types that.
The mechanism is documented by Google, which also gives it its name: [*query fan-out*, that is several related searches issued together across subtopics and data sources](https://developers.google.com/search/docs/appearance/ai-features). The same page states that sites appearing in AI Overviews and AI Mode are included in the overall search traffic in Search Console.
Since June 2026 there is also a dedicated report inside Search Console showing impressions coming from AI features separately. It has two limits that matter for this question: it lives only in the interface, and it reports impressions but not clicks, so it does not give you the CTR. Through the API there is nothing: I queried the search appearance dimension on the domain and the only value that exists is the one for translated results.
So to answer this article's question, what a click on those searches is worth, an estimate is still needed. Query length is the approximation I chose, and like every approximation it has to be declared.
## What is actually changing with query fan-out?
The tool is absorbing the behaviour people used to have. There is no longer someone sitting in front of a search engine: there is someone sitting next to an assistant that searches on their behalf.
That shifts the weight onto something many underestimate: **sources**. An assistant that answers without saying where it got the answer is an assistant you cannot trust, and indeed the tools that cite sources by construction, Perplexity among the first, are the ones that made the idea of agentic search credible at all. That is the direction the others will take, and it means being a citable source matters more than most companies believe. On how you make yourself citable I wrote a [dedicated technical guide](/blog/comparire-chatgpt-seo-geo), and on how a model reads a page there is [a piece of its own](/blog/ai-legge-il-tuo-sito-geo).
## What did I stop writing for answer engines?
This is the part I care about most, and it is not good news.
Writing that delivers value gradually, the kind that builds an argument page after page and hands you the conclusion at the end, no longer works. Bots summarise extremely well: they open the article, pull out the best passage and drop the rest. They do not read the path, so whoever receives the answer will not read it either.
The consequence is that the summary goes on top. That is what I do, and it is what any answer-engine optimisation manual asks for: put the conclusion in the first lines, then explain.
It works. But it is worth saying out loud what it costs.
**Whoever reads an answer does not walk the path the writer had in mind.** They jump from zero to a hundred, get the result and close. And very often it is precisely the path that holds a person's added value: how they ruled out the alternatives, the mistakes they made before getting there, the reason an obvious solution did not work. Knowing how you get there is a fundamental part of growing in a craft, not an ornament.
If the collective habit becomes receiving conclusions without crossing any reasoning, the effect does not stop at a blog's traffic. It touches the way people learn. I write the summary on top anyway, because it is the medium we have. But it is not a neutral improvement, and it is honest to say so in an article that uses that very mechanism.
## What does query fan-out mean for a small business?
That informational articles are no longer enough on their own, and not because they are useless: because the top of the funnel is exactly the part an assistant absorbs and hands back without sending you anybody.
What you need alongside is a real, built funnel, of a kind you very rarely see out there. Information at the top, then actual tools in the middle, something a prospect really uses and walks away from with a result in hand, and at the bottom an offer that puts the value in front rather than behind, with the courage to lose money on it. That money is the marketing investment: it is not a discount, it is how you prove what you can do before asking for anything.
The most common mistake among the entrepreneurs I meet is thinking a paid campaign is enough. Underneath sits a deeper and more mistaken belief: that marketing works on feelings, likeability and word of mouth. Word of mouth works, but it is not a strategy, it is an effect. Marketing is an applied science, with hypotheses, experiments and measurements, which is the reason this article is full of numbers instead of adjectives.
If the click at the top of the funnel is disappearing, the answer is not to write less. It is to stop measuring the blog with the wrong yardstick and to build the two layers that come after it.
## Frequently asked questions
### What is query fan-out?
It is the technique by which an answer engine, instead of running the search you typed, launches several related sub-searches in parallel and then composes its answer from what it finds. Google uses this term to describe how AI Overviews and AI Mode behave. For a site owner it means appearing for broader and more specific questions than before, with more impressions and a lower CTR.
### Do AI impressions show up in Search Console?
Yes, and since June 2026 separately too. Google states that sites appearing in AI Overviews and AI Mode are included in overall search traffic, and a dedicated report shows impressions coming from AI features on their own. Two limits remain: that report lives only in the interface and reports impressions without clicks, so you cannot derive the CTR. Through the API no value exists yet that lets you isolate them.
### Why does CTR collapse if rankings do not get worse?
Because who sits on the other side has changed. On a search made by a person, appearing on the first page produces clicks: on this site one to three word queries return 2.58%. On long queries, largely generated by an engine gathering material, CTR drops to 0.11% while still sitting between fourth and ninth position. There is nobody clicking, there is a model reading and citing.
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "What is query fan-out?",
"acceptedAnswer": {
"@type": "Answer",
"text": "It is the technique by which an answer engine, instead of running the search you typed, launches several related sub-searches in parallel and then composes its answer from what it finds. Google uses this term to describe how AI Overviews and AI Mode behave. For a site owner it means appearing for broader and more specific questions than before, with more impressions and a lower CTR."
}
},
{
"@type": "Question",
"name": "Do AI impressions show up in Search Console?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Yes, and since June 2026 separately too. Google states that sites appearing in AI Overviews and AI Mode are included in overall search traffic, and a dedicated report shows impressions coming from AI features on their own. Two limits remain: that report lives only in the interface and reports impressions without clicks, so you cannot derive the CTR. Through the API no value exists yet that lets you isolate them."
}
},
{
"@type": "Question",
"name": "Why does CTR collapse if rankings do not get worse?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Because who sits on the other side has changed. On a search made by a person, appearing on the first page produces clicks: on this site one to three word queries return 2.58%. On long queries, largely generated by an engine gathering material, CTR drops to 0.11% while still sitting between fourth and ninth position. There is nobody clicking, there is a model reading and citing."
}
}
]
}
</script>