On 6 September an American B2B data vendor refreshed its "Top 25 companies based in Finland" pages, one per category, some of them down to city level.
| 4 Sept | 22 Sept | 12 Oct | |
|---|---|---|---|
| The startup's ranking pages, companies cited for | 7 | 12 | 22 |
| The data vendor's pages, companies cited for | 2 | 18 | 41 |
On 12 October the startup's pages were cited by all five engines we asked. The data vendor's pages were cited by Claude for 28 companies, after none in September, and 78 different pages of the vendor's site turned up among the sources. With Perplexity included, the two were cited for 23 and 51 of the 100.
Asked the same questions in Finnish on 12 and 15 October, the startup's pages were cited for just as many companies as in English, 19 and 20 on the four engines that answered in both languages (19 in English). They come in Finnish too. The American vendor's English-only pages fell from 44 to 12 and 23.
Almost nobody reads these pages. By Semrush's October estimate, the startup's site gets about 70 visits a month from Finnish Google and the vendor's fewer than 200. AI reads them anyway.
Both companies sell sales prospecting tools. The companies described on their pages are, in a sense, the product.
What this means in practice
The German directory in Finding 1 could look like a one-off, an old site that happened to be well placed when AI search arrived. These two show that it isn't. The same thing happened in front of us, on a clock of days.
None of this is a fault in the engines. Search-enabled AI is built to find relevant new pages and use them, and it did exactly that. The point is who is publishing them. What it took was a page built for a machine to read. A clear title that matches the question a buyer asks, a list, a date, a short note on method. Nothing on these pages needed to be read by a person to work.
This works both ways. Anyone with a reason to publish a ranking of your industry can become one of the sources AI uses to describe your company within a week, whether you know about it or not. And a company that publishes clear, dated, well-structured material about its own field can become a source the same way. In this study, almost none of them did.
This doesn't show in search rankings, website analytics or media monitoring. It only shows if someone reads the sources under the answers.
One company, three dates
A Finnish company in the top half of the 100 by revenue. Its website has schema markup. AI knows it well: most of the engines named it on every date, and in October two of the five put it first in its category.
On 12 October Claude named it first and described its main business and a recent investment. Its sources for that answer were three pages of the German directory, three trade news sites in its field, a 2022 article on a Finnish business news site, an industry cluster's member list, and the English Wikipedia article about one of its competitors. None was the company's own website.
Across the three dates, the five engines cited 40 different sources in answers about this company, 112 citations in all. Not one of them was the company's own site, or the site of any of its brands. The sources were the German directory, international trade press, Finland's official country image site, a market research company, a national research institute, a Finnish sourcing company, and in September, a travel blog.
The company is described well, and it is first in two of the five October answers. It just isn't the one doing the describing.