Ranking in the top five and still not cited: one Dutch site, Google's own numbers

The data

We combined three sources for one Dutch home-security company: Google's standard Search Console report, its generative AI report, and five weeks of AI answers to 100 fixed buying questions. The first gives each page's average position, the second how often Google showed a link to it in AI answers, the third which pages were cited.

Three sources, three periods

Source	What it counts	Period	Size
Search Console, performance report	Clicks, impressions and average position per page	3 June – 2 September 2026	51 pages
Search Console, generative AI report	Impressions of links in AI Overviews and AI Mode, per page	30 August – 5 September 2026	37 pages; 984 impressions for the site (1,033 summed per page)
Our measurement, 100 buying questions	Pages cited as a source in the answer	10 August – 14 September 2026	1,616 Google searches, 1,599 Perplexity answers

The periods differ because Google's AI report was exported for one week, while average position needs a longer window to be stable. That is a limitation, and we return to it below.

Why this site

The company sells home-security systems in the Netherlands and has 60 addresses in its sitemap, 21 of them blog posts. It ranks reasonably in Google, gets roughly a thousand AI impressions a week, and is cited in AI answers often enough to count, which makes it a useful case. It is one site, and the pattern on another site may differ.

Position does not predict AI impressions

Among the 39 pages with at least 100 search impressions, the Spearman correlation between a page's average position and its AI impressions was 0.06. The number of ordinary search impressions did track AI impressions, with a correlation of 0.62. How often a page appeared in Google tracked its AI impressions; where it ranked on average did not. With 39 pages the uncertainty is wide: the 95% interval runs from about −0.26 to 0.37 for the first correlation and from about 0.37 to 0.78 for the second.

Top-ten pages with next to nothing

Of those 39 pages, twelve average a position in the top ten. Six of them had four or fewer AI impressions in the week, including a product page for an accessory at 8.4 with none and a sizing guide at 6.1 with three.

Page	Average position	Search impressions (3 months)	AI impressions (1 week)
Blog: a crime-statistics question (informational)	4.3	7,942	217
Blog: a household-situation question	5.1	1,608	18
Blog: explainer of a service component	5.3	7,271	217
Blog: a sizing question	6.1	253	3
Blog: a feature question	6.2	239	49
Product page: accessory	8.4	103	0
Blog: comparison of the best options for home use	15.0	19,909	162
Main service page	16.4	8,331	49
Price page of the main service	23.9	8,981	56
Service page: business customers	42.6	11,146	32

Low-ranking pages that still show up

The reverse also holds. The service page for business customers averages position 42.6 and still collected 32 AI impressions in the week. The price page averages 23.9 and had 56. Google's documentation offers a possible explanation: AI Overviews and AI Mode may run several related searches on subtopics and show a wider set of links than a classic search.

Shown is not the same as cited on a buying question

The best-ranking article, on an informational crime-statistics question, averages position 4.3 and was shown 217 times in a week in Google's AI features. In five weeks of buying questions it was cited zero times, by Google AI Overviews, Perplexity, ChatGPT and Claude alike. It answers a question people ask, just not a question people ask right before they buy.

The page that did the work

The page cited most often on buying questions is a comparison of the best options for home use. It averages position 15.0 in Google, well off page one. In our measurement it accounted for 168 of the 406 citations of this site in Google AI Overviews, 41%, and 353 of the 1,329 in Perplexity. It is also, after the homepage, the page with the most search impressions on the site, which fits the 0.62 correlation above: it is found often, just not at the top.

What the question decides

The two sources measure different questions. Google's AI report covers every search that led to an AI answer with a link to the site, presumably including informational searches; the report does not show which. Our measurement covers 100 questions a buyer asks: which system, which provider, what it costs. A page can win the first and be absent from the second. For a business, the second is closer to the moment of purchase.

What larger studies found

Large studies disagree on how often AI answers draw from Google's top ten, and the answer moved sharply within a year. They measure per query, not per page, so they answer a slightly different question than our page averages. Read side by side, they show that a top-ten ranking is not required; our own data adds that it is not sufficient either.

Ahrefs on AI Overviews: 76%, then 38%

In July 2025 Ahrefs reported that 76.1% of pages cited in AI Overviews ranked in Google's top ten for the same query, based on 1.9 million citations from 1 million AI Overviews, counting the three most visible citations. In March 2026 it published an update on 4 million citations across 863,000 results pages: 38% of cited pages appeared in the first ten results blocks; counting organic links only, 37.1% ranked in the top ten, 26.2% between 11 and 100 and 36.7% below that. Ahrefs has since improved its parsing, and the first study counted only the three most visible citations, so the two figures are not a clean trend; Ahrefs suggests, with a caveat, that the move to Gemini 3 and heavier query fan-out may explain the drop.

Ahrefs on ChatGPT and other assistants: 12%

For assistants outside Google the overlap is smaller. In an August 2025 study of 15,000 long-tail queries, 12% of URLs cited by AI assistants ranked in Google's top ten for the original prompt. For ChatGPT's in-text citations it was 8.0%, for Perplexity 28.6%. Ahrefs collected the data in early July 2025; we have not measured this overlap for Dutch queries.

What this means for your own site

Do not read a good average position as evidence that AI systems use a page, and do not read a poor one as evidence that they ignore it. Check it directly: in Search Console's generative AI report for Google, and by asking your own buying questions for the other systems.

A three-step check

1. Export both reports. In Search Console, export the performance report by page and the generative AI report by page, for overlapping periods if you can.
2. Put them side by side. Sort by average position and look at the AI impressions column. Pages with a strong position and no AI impressions are the interesting ones.
3. Ask the buying questions. Put the ten questions a customer asks just before choosing to ChatGPT, Perplexity and Google, on three different days, and note which of your pages appear as a source.

What the Dutch context adds

In the Netherlands, ChatGPT had 81.7% of the AI chatbot market in August 2026, according to Statcounter. None of that is visible in Search Console. On this site, ChatGPT cited its pages 25 times in five weeks of the same buying questions, against 1,329 times for Perplexity. A Google ranking tells you nothing about that gap either.

Limitations

This is one site, three sources and three periods, and a correlation over 39 pages. It shows that position and AI visibility came apart here. It does not show why, and it does not show that they come apart everywhere.

• Average position is an average. It mixes every query a page ranked for, including brand and long-tail searches, so a page can average 15 while ranking first for a few valuable queries.
• The periods differ. Position covers three months, AI impressions one week, citations five weeks.
• The two counts differ. An AI impression counts each time a link to the page was on a results page with an AI Overview or AI Mode answer; our citation count is how often the page appeared in the sources for our 100 fixed questions.
• Gemini is not in the per-page counts. It returns sources as redirect links, so the page behind a citation is not visible in the address.

Frequently asked questions

Does a high Google ranking help at all?

Probably, but not in the way the ranking suggests. On this site, how often a page appeared in Google search correlated with AI impressions (0.62); its average position did not (0.06). Being found for many searches seems to matter more than ranking high on average.

Why was the best-ranking page never cited on buying questions?

Because none of those questions was about crime statistics. The page answers an informational question, and our measurement asks buying questions. It was shown in Google's AI features 217 times in one week, presumably on other searches; the report does not show which.

Can I see this in Search Console?

For Google, yes. Since 31 August 2026 every site with enough AI impressions has a generative AI report under Performance, with impressions per page, country, device and date. It shows no queries and no clicks, so you see which pages are used but not on which searches.

Is this the same in ChatGPT?

We cannot tell from Search Console. On this site, ChatGPT cited the company's own pages rarely: 25 times across roughly 1,600 answers to the same buying questions.

Where we stand

We measure which pages AI systems cite, per question and per system, rather than inferring it from rankings. AIGeschikt measures 100 fixed buying questions per client, three times a week (Claude three times a month), and records the full source list of every answer, so each question shows which page of which company is cited. The Dutch version of this analysis, which counts which of the same site's pages were cited, is Twee van de 60 pagina's leverden bijna de helft van alle AI-citaties. Our other English pieces are Is GEO just SEO? and Should you move budget from SEO to GEO?.