How we read the findings: Infrastructure, Information, Influence
At VML Finland we sort AI visibility work into three parts. They are a useful way to read this report back, because the findings don't spread evenly across them.
Infrastructure: can AI read you. Whether the engines' bots get in, whether the page text reaches them, and whether the code on the site tells them which company this is. In this report: a third of the companies have no schema markup, six turn away at least one engine's bot, and the companies with schema were no more likely to have their website used (Finding 3).
Information: does your content give AI something to quote. Pages an engine can lift a clear, sourced answer from, with a named author and claims it can check. In this report: when the engines did read a company's site, about half the time it was the About page or the homepage, and a product page one time in six at most (Finding 1). Asked as a buyer, they went deeper and often landed in an investor PDF (Finding 5). Companies with a named author on their pages had their own site cited more often on two of the three dates, a lead we'll keep following.
Influence: what the rest of the web says about you. Wikipedia, directories, industry press, ranking lists, data vendors, everything the engines treat as a source that isn't yours. In this report this is where most of the evidence sits:
- one source in fifteen being the company's own site, and a slightly smaller share when asked in Finnish (Finding 1)
- a German directory cited for more of the companies than their own websites in English, and Finnish third parties taking its place in Finnish (Finding 1)
- an English Wikipedia article going with being named first, among the 50 largest on every date (Finding 3)
- old corporate structures still described as current, one of them more than two years after the change (Finding 2)
- two new sources entering the pool within days of being published and cited for more companies on each date since (Finding 4)
- unlisted companies named as often as listed ones but almost never read from their own site (Method)
Most of what this report found sits in the third one.
What good looks like
When a buyer asks AI who the leading suppliers in their field are, or asks Google's AI Mode to keep track of a category for them, the answer is put together before anyone from your company is in the room. What the engine reads about you is mostly not yours. That's the consequence in one line, and every finding in this report is a version of it.
In August, in The Invisible Leader, I put Infrastructure first. It was the foundation: if AI can't get into your site, everything else is moot. That is still true. But once the door is open, the outcome in this data went with what others write about a company. So the order flips.
Influence first. A company that knows which sources describe it, and has decided which of them to correct, which to outweigh and which to leave alone. Today, for most companies, a directory, a data vendor's list and an outdated Wikipedia article tell the story in English, a different set of outsiders tells it in Finnish, and nobody inside has looked at either. Done well, someone owns the list, the reference sources say what the company is today, and the next time a sales software startup publishes a ranking of your industry, somebody notices within the week.
Then Information. Before: the engine reads the About page because there is nothing else on the site that answers the buyer's question. After: the questions buyers actually ask have a page each, with a name on it, a date and claims the engine can check, and the references and certifications a tender asks for are somewhere an engine can find them, not only in a PDF.
Then Infrastructure. Still needed. An engine that can't read a site can't use it, and a third of these companies have no markup at all. When an engine looks a company up, it reads the words on the page, not the code behind them (see "What the markup says, and who reads it"), so the page has to say plainly what the company is and does, and the markup has to say the same. In this data it is where the work starts rather than what decides the answer, and in most companies it is the part already in someone's backlog.
Three things to check next week
None of these needs a tool or a developer. Each one ends in a decision for whoever owns the brand story.
1. Find out who is describing you today. Take the question your buyer would ask ("Who are the leading [your category] in Finland?"), in English and in Finnish. Put it to at least three engines, three times each, and write down every source under the answers. Count how many of them are yours. Across this study it was about 6%, and for 24 of the 100 companies it was none, on any engine, on any of three dates. Then decide which of those sources to correct, outweigh or ignore.
2. Check whether AI still knows the old you. If you have merged, demerged, renamed, sold a division or changed owners in the last five years, ask each engine what your company is and who owns it. In this study 8 of 20 such companies were still described the old way in September, and one engine called a company a subsidiary of its former owner more than two years after the demerger.
3. Ask the buyer's question and read why you made the list. Ask for a supplier shortlist in your category, the way a procurement manager would, and read the reason given next to your name, not just where you are on the list. Then put it next to what your sales team leads with. In this study the engines gave size as a reason in 72% of answers and references in 38%, certifications in 8%. If your strongest proof is references or certifications, the research your buyer's AI does may not find it. Whether that is a content question, a PR question or a sales question is the decision this check hands you.
Your own row
A single answer about one company wouldn't tell you much anyway: only 9 of the 100 got the same result on all three dates.
The free AI Visibility Check does that part. We ask your category's question across the engines several times, and show you who is named, who is first, and which sources the answers come from, including how many are yours.
The report shows your sector. The Check shows your row.
[Contact line: juhani.vaataja@vmlfinland.com, the study page URL]
About the author
Juhani Väätäjä is Tech Lead at VML Finland. In August 2026 he gave The Invisible Leader, a talk on why the leading company in a category is often not the one AI describes best, and this study is the follow-up he promised that room. He designed the study, ran it and wrote this report.
To cite: Väätäjä, J. (2026). State of AI Visibility in Finnish B2B. VML Finland. https://doi.org/10.5281/zenodo.23493119
[email] · [LinkedIn]
Published by VML Finland, Helsinki.