The picture didn't change. 95 of the 100 were named by at least one engine (94 with the plain question that afternoon). 71 were the first name on at least one engine (72). The company's own website was among the sources for 69 (67). Asked in English, the German directory was still the most cited source, for 80 companies (82).
What changed was the reasoning. We read what the engines gave as the reason a company belongs on the list:
| Reason given | Share of answers |
|---|---|
| Size, revenue, market position | 72% |
| International reach | 68% |
| Local presence or Finnish ownership | 58% |
| Advice to check for yourself, ask for quotes, run a tender | 39% |
| References, track record, named customers | 38% |
| Sustainability | 31% |
| Price | 27% |
| Listed, financially stable | 22% |
| Innovation, R&D | 12% |
| Certifications, quality standards | 8% |
What this means in practice
Many B2B sales teams lead with references and certifications. They are what a tender asks for and what a buyer's procurement process scores. In these answers they are close to the bottom of the list. Size and international reach come first, and a company can't publish its way into either.
We can't tell from the answers whether the engines don't value references and certifications, or can't find them. The second is likely to be part of it. Reference cases and certificates are usually the content that sits in a PDF, behind a tab, or on a page nothing else links to. In the whole run, only one company's own certificate was cited from its own website, and it was a PDF.
The engines also differ on how much they trust their own list. Gemini and Perplexity told the buyer to check the shortlist themselves or ask for quotes in about three answers out of four. Claude almost never did. So the same buyer gets a shortlist with a warning on one engine and a shortlist presented as settled on another.
What the buyer question changes
The answers get longer. On every engine they were longer than the plain question's answers, from a third longer to two and a half times as long, while naming about the same number of companies or fewer. The extra length goes into explaining the choice.
The sources move towards reference material. English Wikipedia was cited for 73 companies instead of 61, LinkedIn company pages for 50 instead of 42, and financial data sites gained: one went from 6 companies to 13. The newest ranking pages lost ground: the American data vendor from Finding 4 went from 37 companies to 23. The German directory didn't move.
When the engines do read the company's own site, they go one level deeper. Product and solution pages were cited 36 times instead of 16, news 40 instead of 26, About pages 85 instead of 65. Own PDF documents were cited 45 times for 21 companies, against 26 times for 10. Most of those PDFs are investor material: annual reports, a bond prospectus, roadshow presentations, an offering circular from 2021. When a buyer asks why a company should be on the list, part of the answer comes from documents written for investors, some of them several years old.
Not every engine moves the same way. Claude and ChatGPT cited the companies' own sites for more companies under the buyer question (33 instead of 26, 39 instead of 34). Gemini cited them for half as many (17 instead of 35) and gave fewer sources overall.