Underneath those steady totals, only 21 of the 100 companies got exactly the same result on the two September dates, on every engine. From September to October it was 24. Across all three dates, 9.
What this means in practice
Who this lands on:
- The marketing director who was sent a screenshot on Monday saying "we're first on ChatGPT". It was true when it was taken. It says very little about what the next person who asks will see, and it proves nothing about the other engines.
- The buyer's team. Two people on the same buying team ask the same question on different engines, or on the same engine a week apart, and come back with different shortlists. Neither of them knows the other's list was different.
- Investor relations and communications after a merger, demerger or rename. The engines keep describing the old company for a long time. Investors, recruits and buyers hear about a structure that no longer exists, and they hear it with full confidence.
And a single check on a single engine tells you almost nothing about where you stand.
Same question, different answers
The four engines regularly disagree on the identical question.
In software and IT services on 4 September, one listed consultancy was the first name on two engines and not mentioned at all on the other two. Another was first on one engine and missing on two. A third was named by two engines and first on none. On 12 October the same three looked different again: the first was named by three engines and first on none, the second was named by all four and first on one, and the third was named by two.
Across the whole study, the company named first in a category was the same from one date to the next in about two answers out of three: 66% between the September dates, 62% from September to October. That varied by engine, and no engine kept the same first name more than about three times in four:
| Engine | 4 to 22 Sept | 22 Sept to 12 Oct |
|---|---|---|
| Claude | 73% | 55% |
| Gemini | 73% | 68% |
| Google AI Mode | 58% | 59% |
| ChatGPT | 57% | 65% |
8 companies were mentioned somewhere on 4 September but never among the first three names on any engine. On 22 September it was 9, and on 12 October 14. Only three companies were in that group on all three dates.
The repeat
We counted a change whenever a company was named on one date and not the next, had its own website cited on one date and not the next, or was the first name on one date and not the next, engine by engine. Each company could change in up to 12 ways between two dates: three measures on four engines.
| Changes | 4 to 22 Sept | 22 Sept to 12 Oct |
|---|---|---|
| None | 21 | 24 |
| 1 | 29 | 29 |
| 2 | 22 | 21 |
| 3 | 15 | 16 |
| 4 or more | 13 | 10 |
Where a company was named on both of two dates, its position in the answer was exactly the same 71% of the time in September and 73% into October. About half the companies gained or lost an own-website citation on at least one engine at each step (55, then 48).
Own website cited on at least one of the three dates: 76 companies. On all three: 51. On none: 24. So 25 companies were read from their own site on some dates and not on others. The 24 are the clearer group: for them, AI didn't use their website on any engine on any date.
Four companies weren't named by any engine on 12 October, seven on 22 September and four on 4 September. Only one was missing all three times.
This is why a single answer tells you so little about one company, while the totals across 100 are reliable. A company's position is a draw. To know where you stand you need to ask more than once, on more than one engine.
Who takes your place
In 54 to 56% of answers, depending on the date, the first name was the company we asked about. In the rest:
- two thirds of the time it was another Finnish company
- one third of the time it was a company with its headquarters or parent outside Finland
That split is very uneven between sectors. In industrial machinery the company itself was first in 83% of answers on both September dates and 81% in October, so Finnish companies hold their own categories almost completely. We asked about "leading companies in Finland", and that wording helps Finnish companies.
In professional and business services, a foreign or foreign-owned firm was first in 21% of all answers on 12 October (31% on 22 September, 21% on 4 September), with Swedish, Danish and German owners among them. In construction it was the largest share of all on every date, but the sector has seven companies, so read it as a direction.
Software is the other pattern. There the first name is usually another Finnish company, in about half of all answers. The consultancies that take those slots are mostly ones whose reputations were built while they were private, and two of the most frequent are still unlisted. They aren't in the study, and they are still the answer to the question about the companies that are.
The largest listed company isn't the default answer
Nine of our questions are shared by two or more of the 100 companies, because they compete in the same market: forest industry, construction, infrastructure building, software consulting, engineering consulting and accounting among them. That allows a direct question. When AI names the leading companies in a category, does it start with the largest listed one?
Mostly it doesn't. Across the 96 answers to those nine questions on 12 October, the largest listed company in the category was named first in 39% (35% and 34% on the September dates). In about half, 48% (52% and 56% before), the first name was a company outside the 100: a privately owned company, a foreign group or its Finnish unit. In four of the nine questions, the largest listed company was first in fewer than half of the answers.
Some of those outside companies are large in their own right, so this doesn't mean the leader is overlooked. It means that being the biggest listed company in your category doesn't make you AI's default answer. We added this comparison after the third date, so read it as a pattern on nine questions.
The old company
For 20 companies in the study we know that the name or the corporate structure changed in the last few years: a demerger, a rename, a business sold or spun off. On 4 September, 8 of the 20 were still described in the old form by at least one engine. On 12 October it was 5.
The sharpest case is a large industrial group that was described as part of its former parent by three of the four engines, more than two years after the demerger. Three weeks later, two of those engines had caught up and now called it "historically part of" the old parent. The third still described it as a fully owned subsidiary, and on 12 October it said so again, in the same words.
The engines aren't uniformly out of date. On the same question, one engine will describe a recent demerger correctly and another will list the old company as if nothing happened. So you can't check one engine and assume the rest agree.
The old structure also appeared when we asked as a buyer, on six companies. A procurement team building a shortlist gets the old company too.
For a listed company this is not only a marketing question. The people asking AI what the company is and who owns it include investors.